STAMP project

MASTER THESIS TITLE: STAMP project MASTER DEGREE: Master's degree in Applications and Technologies for Unmanned Aircraft Systems AUTHOR: Isaac Hoyas Ester ADVISOR: Mario García Lozano DIRECTOR: Xavier Banqué Casanovas DATE: June 2018 Title: STAMP project Author: Isaac Hoyas Ester Advisor: Mario García Lozano Director: Xavier Banqué Casanovas Date: June 2018 Abstract During the historical evolution of aerial photogrammetry all the production processes involved in its workflow have, in one way or another, been improved or optimized by the contemporaneous technological breakthroughs. There is however one area that, until today, continues unaltered since its inception, the Ground Control Point survey process which is still manually performed by a human. The motivation of this master thesis is twofold, in one hand to study a method to automate the Ground Control Point collection process by using Remotely Piloted Air Systems as both, visual targets as well as Global Navigation Satellite System receivers and, on the other hand, study if the technological concept may become a viable business or not. TABLE OF CONTENTS INTRODUCTION ................................................................................................ 1 1. TECHNICAL ANALYSIS ........................................................................... 5 1.1. Drone technological state of the art ................................................................................ 5 1.2. Drone selection .................................................................................................................. 6 1.2.1 Price ........................................................................................................................ 7 1.2.2 Body Size ................................................................................................................ 8 1.2.3 Drone weight......................................................................................................... 11 1.2.4 Flight time ............................................................................................................. 11 1.2.5 RF Range ............................................................................................................. 12 1.2.6 Lens Field Of View (FOV) ..................................................................................... 12 1.2.7 Lens aperture........................................................................................................ 13 1.2.8 Drone selection conclusions ................................................................................. 14 1.3. Rokubun STAMP.............................................................................................................. 14 1.3.1 Drone enhancement ............................................................................................. 14 1.3.2 GNSS software enhancement .............................................................................. 15 1.3.3 Tablet software ..................................................................................................... 16 1.4. Legal framework implications ........................................................................................ 16 1.5. Technical analysis conclusions ..................................................................................... 19 2. FIELD TESTING ...................................................................................... 19 2.1 Master drone camera resolution study ......................................................................... 20 2.1.1 Problem presentation ........................................................................................... 20 2.1.2 Preliminary Field Tests ......................................................................................... 21 2.2 STAMP extensive field test ............................................................................................. 23 2.2.1 Location ................................................................................................................ 23 2.2.2 Status of CATUAV GCP network ......................................................................... 23 2.2.3 Survey of CATUAV old GCP network .................................................................. 25 2.2.4 Final survey area bounds ..................................................................................... 29 2.2.5 STAMP slave deployment rationale ..................................................................... 30 2.2.6 STAMP slave availability ...................................................................................... 30 2.2.7 STAMP field survey .............................................................................................. 31 2.2.8 Argonauts GNSS post processing ........................................................................ 33 2.2.9 Photogrammetric processing ................................................................................ 33 2.3 Field testing conclusions ............................................................................................... 38 3 COMMERCIAL STUDY ........................................................................... 40 3.1 Potential figures and trends ........................................................................................... 40 3.2 Analysis and segmentation ............................................................................................ 43 3.3 The Competition .............................................................................................................. 43 3.4 Comparative product analysis ....................................................................................... 45 3.5 Revenue Streams............................................................................................................. 46 3.6 Supply chain .................................................................................................................... 49 3.7 Key Partners ..................................................................................................................... 50 3.8 Risk Mitigation ................................................................................................................. 51 3.9 Cost Structure & Financial Projection ........................................................................... 52 3.10 Commercial Study conclusions ..................................................................................... 54 4 CONCLUSION ......................................................................................... 55 5 REFERENCES ......................................................................................... 56 5.1 Acronyms ......................................................................................................................... 56 5.2 Bibliography ..................................................................................................................... 58 Introduction 1 INTRODUCTION When a person or entity wants to create cartography of the surface of a given subject, like a portion of the planet Earth, there are two different approaches that can be followed. The first approach would be to manually survey significant points of the surface to be mapped. This allows to create a dataset in the field that would only contain the data required to build the cartography. This method is human resource intensive because it is very time consuming on the field but, on the other hand, the processing of the data is relatively simple and fast. The second approach would be to systematically scan the surface to be mapped by using Synthetic Aperture Radar (SAR), LIght Distance And Ranging (LIDAR) or orthorectified imagery also known as photogrammetry [1]. This second approach of collecting data is typically much quicker on the field than the first method but it also generates very large amounts of data that has to be processed in powerful workstations. Among all the methods previously mentioned the most cost efficient when mapping large terrain areas is photogrammetry. This method is widely used in the booming Unmanned Air Vehicle (UAV) sector because digital cameras are becoming cheaper, smaller and lighter with ever increasing pixel counts. In parallel to the cameras improvements there are also Global Navigation Satellite System (GNSS) receiver improvements. Those are becoming cheaper, smaller and lighter allowing for very low cost photogrammetric setups able to provide end results that rival in quality with what a few years back was only possible with a multimillion dollar aircraft. However, there are areas that still need improvement, the most important of which is how to accurately relate the UAV photogrammetric end result with an absolute coordinate system tied to the planet Earth, process also known as georeferencing [2]. Georeferencing in photogrammetry can be tackled in two different ways, one is by providing accurate coordinates to each UAV camera shutter release typically achieved using very expensive dual frequency GNSS receivers and Inertial Measurement Unit (IMU), also known as direct georeferencing. The second method is by providing coordinates to Ground Control Points (GCP) scattered throughout the aerial survey zone. This task is usually performed by sending a human surveyor equipped with either a total station ranging from 10 k€ to 30 k€ or using a high end dual frequency GNSS receiver (typically the preferred method) ranging from 10 k€ to 20 k€. This method is known as indirect georeferencing. 2 STAMP Project One example of human made GCP can be seen in Figure1 where a black and white high contrast pattern has been positioned against green grass providing good contrast and easy identification in the resulting aerial picture even when the UAV is flying at 120 metres Above Ground Level (AGL) and using wide lens (78.8° Field Of View). Figure 1 50x50 cm GCP as seen in an aerial picture taken at 120m AGL This GCP survey process, where a human operator has to visit each GCP location one by one, usually takes between half and a full working day. Although time may vary depending on area size, the number of GCPs, the desired accuracy of the survey and the orography of the terrain. In very rough terrain, it may be impossible to deploy GCPs or it may imply severe risks to the surveyor and to the expensive and fragile equipment. In cartographic flights intended to provide a final metric document usually both, direct and indirect georeferencing are used simultaneously providing the highest accuracy but also the most robust / reliable result. In addition to establish the relationship between the photogrammetric end product and a known coordinate system, the GCPs are also used to assess how closely the photogrammetric final model fits the real world which is crucial to certify that the end result meets certain quality criteria. In April 2016 PhD. Miquel Garcia Fernandez and MSc. Xavier Banqué Casanovas, the two founders of Rokubun, started to work in the basic outline of a system that could replace the human surveyors that collect the Ground Control Points (GCPs). That system was also meant to replace the man made targets by a set of drones. Xavier and Miquel made a general sketch of the idea and named the concept STAMP but not long after the project was set in standby. A schematic overview of the STAMP system operation can be seen in Figure 2 Introduction 3 Figure 2 STAMP overview. In January 2018 the project was recovered and reassigned to me with the errand to further develop the idea and study its technical and economic feasibility. The idea behind STAMP is that it should be possible to reduce the GCP field survey time to just a few minutes by substituting the human land surveyor by a set of six to eleven small drones all controlled by a single operator through an easy to use app running on a tablet. The STAMP drones tasked as GCPs would be equipped with Rokubun’s SelfPowered Argonaut (SPA) single-frequency affordable GNSS receiver. The drone role would be to land on the location designated by the operator and remain landed throughout the photogrammetric flight; these drones acting as GCPs are known as “slave” drones. The remaining drone, also known as master drone, would be tasked to perform the photogrammetric flight after all the slave drones have landed meaning that the master drone would capture every slave drone landed on the ground in the aerial pictures. At the end of the flight of the master drone the slave drones would sequentially take off and return to the coordinates designated as “home point” always under the supervision of the operator. After all the drones are back to the home coordinates the GNSS data collected by each drone would be downloaded to a PC (or tablet) and then be uploaded to Rokubun’s Positioning as a Service (PaaS) through Internet to obtain a few minutes later an accurate positioning solution for each STAMP slave drone. The post processing of the GNSS raw data collected on the field using the classic methods would likely take a few hours in the office however instead of using a multi-thousand euro package running locally on a computer the GNSS 4 STAMP Project post processing software as conceived by Rokubun would be a cloud based, easy to use, pay per use Positioning-as-a-Service (PaaS). In addition to the cost savings in workflow optimization, it is also expected that STAMP will reduce the costs in hardware and rationalize the expenses in software. Beyond enhancing today’s photogrammetric workflows, STAMP will also allow to map areas were, previously, it was not possible to access, like plains over cliffs or lands beyond a river without bridge to cross to the other side. The expectative is that STAMP, in addition to the increased GCP collection speed, should provide external accuracies (positioning in an absolute reference frame like ETRS89) orders of magnitude better than uncorrected GNSS Single Point Positioning. Depending on the accuracy obtained in the stamp extensive field test STAMP would be suitable for different applications. For instance, STAMP may be able to match GIS expected accuracies (better than 30 centimetres) but not land surveying accuracies (centimetre level). This document has been divided in three main chapters: chapter number one is a theoretical study of the technical feasibility of STAMP, chapter number two will present the results of the empirical testing’s and the third chapter presents a study of the economic feasibility of the STAMP project. . Technical analysis 11 Figure 4 DJI Mavic Pro visibility simulation at 90, 100, 120 and 150 meters AGL Those qualitative results presented in Figure 4 can be summarised in a quantitate table as follows: AGL height Image Vertical Scale GSD (mm/pixel) Image footprint width over a flat surface Image footprint height over a flat surface Slave STAMP UAVs body pixel size 90 meters 1/19000 29 118 m 88 m 7 x 3 100 metres 1/21150 33 131 m 98 m 6 x 2 120 metres 1/25350 39 157 m 118 m 5 x 2 150 metres 1/31700 49 196 m 147 metres 3 x 2 Table 3 DMP/DMPP ground footprint vs GSD performance. At a later stage the theoretical results being presented here will have to be reviewed with an empirical test to take in to account non-simulated factors like non optimal contrast, Complementary Metal–Oxide–Semiconductor (CMOS) Bayer pattern caused artefacts due to irregular quantic efficiency, optical induced deformations, etc. In any case the 150 meters simulation of the DMP/DMPP is at the limit of visual perception therefore the smaller DMA has been discarded as a STAMP slave. 1.2.3 Drone weight At 430 grams the DMA is, by far, the best contender, unfortunately, its small size and its battery life outweigh the benefits brought by the smaller weight. 1.2.4 Flight time When evaluating the flight time we have to separate the two different roles that play the STAMP Master and Slave drones. 12 STAMP Project For the STAMP Master it is mandatory to use the best combination of weight to flight longevity ratio as this factor limits the extension of the terrain that can be covered by the STAMP master which in turn limits the size of the terrain that can be surveyed with STAMP. A reduction in the number of transported extra batteries is important because each extra battery has a protective enclosure and electronics that add to the final weight of the system but don’t add anything to the final solution. So the drone selected for the STAMP master is the DMPP because every DMPP battery provides 3 additional minutes which is an improvement of 10% over the closest competitor, the DMP. From STAMP slave point of view a cheaper drone with a shorter flight time but equal size as the DMPP is acceptable It must be noted that some countries like Spain legally limit the maximum distance a drone can reach from its operator to 500 metres, this rule sets a limitation on how far the slave drones may have to fly back and forth. This means that a single battery per slave drone is enough in most flights 1.2.5 RF Range The radio efficiency summarised as the maximum free space distance at which the drone would be able to downlink vide imagery / telemetry and the radio controller would be able to upload control commands to the drone, is a relevant aspect of the STAMP operation. However not because the operator is ever going to reach such long ranges but rather because the higher the sensitivity of the radios, the better the antennas and the higher the robustness of the modulation the more reliable the radio link will be when the radio operation conditions are sub optimal. This is significant to STAMP, because STAMP slaves are expected to regularly operate near the ground where obstructions are common. Considering that the legal frame of most countries requires the pilot to be in control of the drone at all times, even when it is operating autonomously, a highly attenuated radio link may mean the inability to operate. Because of this DMP and DMPP have been selected. 1.2.6 Lens Field Of View (FOV) Lens FOV or lens focal length is an intricate parameter that cannot and should not be gauged based on it’s numerical magnitude. Technical analysis 13 Generally speaking photogrammetry is able to efficiently reconstruct a threedimensional object out of a pair of bi-dimensional images as long as there is sufficient overlap (common ground) between the two images. In short focal length optics (wide FOV) there are two main inconvenients: 1. The GSD of a camera equipped with a wide FOV lens is going to be worse (larger GSD) when compared to the same camera fitted with narrower FOV lens (tighter GSD), this means that that the final model will have less resolution when using wider lens all the other parameters of the flight being equal. 2. The outer parts of an image captured with a wide angle lens may be distorted in such a way that may not be recoverable using the standard Brown frame camera model meaning that usually are discarded. The DMP and the DMPP both use the same camera equipped with an optic focal length of 28 mm in a 35 mm equivalent format, the lens can be considered wide enough for aerial photogrammetry, there is no need to use the wider 24 mm of the DMA. 1.2.7 Lens aperture The “f number” parameter or aperture is a quantification how much light is able to pass a given optic in a pre-set time interval, the smaller the “f number” the more light that lens let go through. The lens aperture has a direct effect on how much time a camera will have to leave the shutter open to gather enough photons to excite each CMOS sensor photocell to complete the picture. For photogrammetric applications it is important to use lenses with a small “f number” because the camera is moving with the drone at a considerable forward speed and if the shutter is open for too long the pixels will suffer motion blur also known as ground smear. Good photogrammetric practices dictate that ground smear must be kept below the GSD so that it does not negatively affect the final quality and accuracy of the orthomosaic typically this is achieved by reducing the forward speed of the drone. This is especially important in rolling shutter cameras, like the ones in DJI drones, where each pixel line is sequentially scanned by the imaging sensor as opposed to higher end cameras using global shutter where all the pixels of the sensor are measured at once. The combination of sensor/drone forward movement and sequential scan of the pixels can severely affect the image integrity causing relevant geometric 14 STAMP Project deformations that must be dealt with in the processing software increasing the computational time. Because the battery life of a drone is typically a limiting factor of the terrain extension that can be covered by a drone, reducing the speed is not a desirable solution because it means that the same battery will cover less terrain so every effort should be made to use the largest lens aperture possible. To further complicate the problem in photogrammetry it may be desirable to fly in an overcast day to eliminate the hard shadows casted by the objects directly illuminated by the Sun because this may improve the quality of the model derived from the pictures as shadowy areas are typically noisier than well-lit areas of the image. 1.2.8 Drone selection conclusions The DMPP has been selected as STAMP master mainly because it provides the best flight times of the three candidates being the only downside the slightly higher cost. The DMP has been selected as STAMP slave mainly because of the lower cost compared to the DMPP additionally the shorter flight time should not hinder the ability of those drones to reach its designated positions within the photogrammetric survey area while still keeping a minimum size. 1.3. Rokubun STAMP Rokubun will have to modify different parts of the currently existing drone systems, those are. 1.3.1 Drone enhancement The DJI DMP and DMPP are equipped with a GNSS receiver able to track GPS and GLONASS constellations in single point positioning. This means that the GNSS receiver within the drone is able to track the satellites in view but there is no way to provide corrections to compensate for ionospheric delays, tropospheric wet and dry delays, inaccurate ephemeris or inaccurate satellite clocks. The position provided by the GNSS receiver is expected to be within a bidimensional circle with a radius of ±2.5 m 50% of the time assuming that the receiver is stationary 1 . This kind of accuracy is good enough for drone 1 Assumes no correction source at all, if Satellite Based Augmentation System is enabled, which currently is unclear, then that figure would decrease to ±2 metres 50% of the time assuming a stationary receiver but it would still be insuficient to meet most photogrammetric constraints. Technical analysis 15 navigation but it is not good enough to determine the position of the drone for photogrammetric uses where accuracies better than ±10 centimetres may be expected depending on the application. Because of this, Rokubun will equip each STAMP drone with a derivative of the compact and light Argonaut GNSS receiver based on the u-blox NEO-M8T GNSS chipset. The M8T chipset is actually very similar to the receiver on board of the DMP/DMPP drones but with a slight difference, instead of just provide the precomputed position based on the signals observed from the different satellites it also provides the raw data in which the position calculation is based on. The Argonaut was designed to store the GNSS raw data in a microSD card with the objective to transfer that raw data in to a post-processing software to further enhance the positional accuracy and eventually meet the photogrammetric stringent requirements assuming that the raw GNSS data was captured under benign conditions. The way that Rokubun proposes to post-process the raw GNSS observables is by using the web frontend “Positioning-as-a-Service” or PaaS for short. Behind the PaaS web frontend there is a server running a GNSS post-processing engine that tries to provide an as accurate position as possible with very little user intervention and returning results within minutes. 1.3.2 GNSS software enhancement In the past, it used to be necessary to acquire a post processing software that typically costed around 6000 €. The software required the end user to be trained to properly configure the parameters and then it forced the end user to manually select, find and download the raw data files from a GNSS reference station that had to exactly match (or exceed) the time range of the GNSS rover. Rokubun proposes a “pay per use” approach where the PaaS users would only pay for each post processing performed avoiding to pay large quantities upfront without knowing if there would be a return of investment that justifies the expense which is a critical consideration for small companies. The PaaS has been built to automate all the tedious work related to finding the GNSS base stations saving precious operator time and avoiding possible mistakes. The PaaS is able to perform its task automatically so there is no need to train the operator because all what is required is to upload the GNSS rover raw data files to Rokubun’s server. 16 STAMP Project 1.3.3 Tablet software One of the pillars of the STAMP system would be a tablet app based on the flexible DJI API that would: 1. Provide a point and click interface similar to Goggle Earth that would allow to plan the photogrammetric flight in an easy and intuitive way but providing enough technical details for advanced users and with the ability to cache maps to be used at a later moment in the field without internet. 2. Enable the user to select the lateral and the longitudinal overlap as well as the mission altitude so that the parallel flight lines typical of photogrammetric flights are automatically created. 3. Provide the possibility to use NASA’s Shuttle Radar Topography Mission (SRTM) Digital Surface Model (DSM) [3] to allow terrain flight following for the DMPP. 4. Display a step by step check list for every step involved in the deployment of STAMP. 5. Control the status of the drone battery health parameters to ensure that there is enough battery to fulfil the mission and return to the landing point. 6. Send each DMP slave towards its targeted objective in a sequential fashion (one after the other, never simultaneously) with a preprogramed return flight profile so that if the slaves lose connection they could automatically return to the launch point autonomously. 7. Upload and execute the photogrammetric flight to the DMPP master. 8. Continuously display telemetry information for the drone in the air. 9. Provide the ability to the operator to immediately stop the procedures at any point, freezing the drone in the air and allowing the operator to provide manual controller inputs. 10. Record every command, input, log or event generated by the drone for legal purposes as well as for further analysis. The design of this application has not been yet outlined and is outside the scope of this thesis. 1.4. Legal framework implications There are very few legal limitations regarding drone sales, therefore STAMP itself would not be legally bound by any other regulations other than the ones that regulate the sale of commercial products. The lack of laws affecting the sale of drones does not mean that there are no rules to fly them, quite the opposite. The regulation affecting STAMP operation, being a drone system exclusively targeted at the professional sector, is actually quite complex because there is no homogenized legal framework regarding drones, each country has its own variation. Technical analysis 17 Within the European countries the general trend is to conform each national regulation to the European Aviation Safety Agency (EASA) recommendations, but in the end every state member has its own law. In the Spanish case, the Spanish Aviation Safety Agency (AESA) is the state agency defining the drone regulation framework within the country. The latest regulation iteration is written in the Royal Decree of the 29th December 2017. The law requires each commercial pilot to: 1. Be over 18 years old 2. Have a pilot license (may be specific for RPAS or a Private Pilot License). 3. Have a medical certificate of suitability. 4. Be insured by a government authorized insurer. Each drone dedicated to commercial operations must have: 1. A license plate. 2. A certificate of characteristics. 3. A maintenance log. 4. An approved operation manual. Each commercial operation must have an individually approved security study. The basic legal limitation factors for drone operation are: 1. Day light operation only. 2. Maximum altitude 120 metres Above Ground Level (AGL). 3. Maximum distance from the pilot is 500 metres. 4. It is mandatory to operate within line of sight. 5. It is forbidden to fly over inhabited areas (villages, towns, cities…). 6. It is forbidden to fly over people with the exception of those related to the drone operation. 7. The UAV pilot must be in control of the drone at all times. 8. The drone must fly farther than 8 km from airports. 9. Sensitive facilities like military bases, nuclear power plants, jails, stadiums … should not be overflown Most of these limitations can be softened or completely lifted if specific permissions are obtained from the competent authorities. The legal procedure to obtain exemptions is slow and cumbersome that most companies do not even consider such possibility, this implies that the vast 18 STAMP Project majority of today’s commercial drone operations are executed under the previously mentioned legal constraints. The only legal limitation that affects STAMP from technical perspective is the height limitation. The practical effect of that limitation has already been discussed in the drone selection title. Most countries legally require the drone pilot to be in control of the UAV at all times. The objective of such rule is to ensure that an autonomous preprogramed flight using waypoints is not an excuse for the pilot to not to be able to immediately recover the drone control in case of an abnormal behaviour or if there is an air space intrusion by another aircraft. Because one pilot should only control a single drone at any given time this means that swarms of drones are illegal because there is no way for a single pilot to effectively control all the flying drones simultaneously. This implies that STAMP users will have to send slave drones one by one and wait until each drone lands before sending the next one so the master drone would only be able to take off after all slave drones have ben landed and vice versa. The “one drone in the air at any given time” is not really a setback for STAMP because the inaccuracies inherent to the GNSS systems when operating without corrections make impossible to launch drones simultaneously without separating them 12 metres or more. 𝑀8𝑇𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦=±2.5𝑚 𝐶𝐸𝑃(50%)→2𝐷𝑅𝑀𝑆(95%)=𝐶𝐸𝑃×2.4=±6𝑚 2 Equation 3 GNSS Single Point Positioning (uncorrected) expected accuracy figures. In addition to the separation between drones at the moment of take off the mobile application controlling the drone trajectories would have to intelligently adapt the path of each drone so that there are no crossings between slave drones routes. An alternative solution to avoid mid-air collisions between drones would be to force each slave to fly at a different flight level separated 10 meters from each other to avoid mid-air collisions. This last approach however is inconvenient because assuming that all the slave drones take off at the same moment the drone flying the highest would have to fly 120m AGL but the bottom drone would have to fly at 20 m AGL which is dangerously close to the trees or buildings. 2 CEP stands for Circular Error Probability. 2DRMS stands for Horizontal Root Mean Square 2σ Technical analysis 19 1.5. Technical analysis conclusions As previously discussed, we can conclude that, from drone technology standpoint the hardware available on the market today fulfils the STAMP needs, since DMP and DMPP are suitable candidates for STAMP from specification sheet point of view. However, in the process of reviewing the market hardware platform availability a few DMP / DMPP weaknesses were unveiled, most of them linked to the DJI API and DJI firmware flexibility. Those limiting factors are discussed in greater detail along the COMMERCIAL STUDY chapter. On the other hand, the modifications that Rokubun is expected to perform to the DMP / DMPP are limited to the addition of a COTS GNSS receiver manufactured by Rokubun, which is easy to implement. The PaaS is already functional, however, the engine will gradually evolve towards a more refined post processing engine able to combine data from a GNSS receiver and from an IMU. The tablet app project has not yet been started and its development targets listed in the present document are just a preliminary draft. The app concept will be further developed in a likely future document. The legal framework does impose limitations to the drone operators but those limits do not hinder the usability of STAMP drones except for operation in urban areas, as you know, where special permissions would be required to overfly inhabited buildings. 2. FIELD TESTING To ensure the practical viability of STAMP the following two chapters will verse about the preparation and end results of the two field tests that were carried out to determine if STAMP, as conceived in the technical analysis section, performs as expected:  Field testing 1 - Master drone camera resolution study  Field testing 2 - STAMP extensive field test This field tests were necessary to reduce as much as possible the risk that arises from the untested hypothesis presented in the technical analysis section. If the two field tests are successful it would mean that the DMP/DMPP camera is suitable for the project and also that the STAMP general concept can be deployed in the field with a reasonable success chance. 20 STAMP Project Even if the two tests presented in this section are successful, further analysis and field work would be required by Rokubun to ensure that the tablet software fulfils the customer expectations. The main reason is because the tablet app coding process has not yet been started, so obviously there is no way to perform testing on that part of the system. 2.1 Master drone camera resolution study To ensure that the STAMP field test (see chapter 2.2) is successful a preliminary test must be performed to ensure that at least the DMP slaves are visible in the master imagery. 2.1.1 Problem presentation The requirements to perform a STAMP field test are: 1. A test field far from buildings, urban areas, and inhabited areas in general that provides easy vehicle access and with no other drones flying around. 2. A network of highly visible and accurately measured Ground Control Points scattered through the survey zone to provide a trusted geometric network to compare the STAMP photogrammetric result against. 3. Ten DMP slave drones or if it is not possible to acquire such large amount of equipment ten silhouettes that simulate the presence of such drones on the ground. 4. Ten Rokubun SPA or Argonaut + battery GNSS receivers to accurately record the GNSS raw data as seen by each drone. 5. DMP or DMPP drone to perform the photogrammetric flight (unlike the slaves this drone must be real and it must also be of the category selected to obtain relevant results). 6. Flight planning app with photogrammetric planning capabilities on a tablet to control the flight of the DMP/DMPP as well as control the camera shutter releases and monitor telemetry. 7. A suitable GNSS fixed base station positioned over a point of well-known coordinates. It can either be provided by Rokubun or use one of the stations part of Institut Cartogràfic i Geològic de Catalunya (ICGC) CatNet network. 8. Software used by Rokubun to post process GNSS raw data. 9. Properly licensed photogrammetric software, preferably Agisoft PhotoScan. 10. Workstation running Windows to perform the GNSS calculations as well as the very CPU / GPU / memory intensive photogrammetric calculations. 11. A person who can help to spread GCPs and slave test subjects (assuming they are not real drones). 12. A full working day (around 8 hours) to perform all the field operations. 13. A full working day (about 8 hours) to perform the GNSS post-processing. Note that due to the special field conditions not currently contemplated in Technical analysis 27 Figure 8 ICGC datasheet for the closest geodesic mark to CATUAV facilities. To follow the IGCG example a 3D standard deviation 1σ of 5 cm was assumed for all the geodesic coordinates (latitude [λ], longitude [Φ] and height [h]) of the GCPs that were to be used in the STAMP trial. The conjunction of Setat’s GCPs with the old CATUAV GCP’s does create a well-balanced dense network of 33 GCPs more or less evenly distributed on the terrain that can be used to test the STAMP accuracy against. A summary of the coordinates of the centre of each GCP used for the STAMP test is shown in Table 4.  In grey colour the coordinates of SBAR ICGC base station and the forced centring pillar right over CATUAV underground building roof.  In blue colour the coordinates of the 11 GCPs surveyed with the Septentrio AsteRX receiver.  In green colour the coordinates of the 22 GCPs surveyed by Setat for CATUAV. Greyed fields correspond to information that is not typically used in photogrammetric work. 28 STAMP Project Table 4 summary of GCP coordinates. scale Geoid EGM08D595 X (m) Y (m) Z (m) Degrees Minutes Seconds Degrees Degrees Minutes Seconds Degrees X (m) Y (m) Degrees Minutes Seconds Degrees factor (k) Marker Undulation Marker SBAR 4745755.10 180181.14 4244588.95 41 58 48.3885 41.9801079 2 10 27.4551 2.1742931 431595.37 4647897.33 033 8.33 -0.55231 0.99965758 937.89 50.485 887.41 CATUAV 4758451.99 179719.46 4230446.95 41 48 33.0558 41.8091822 2 9 46.6057 2.1629460 430470.18 4628928.89 033 28.97 -0.55805 0.99965949 925.85 50.317 875.53 astrx_026 4758509.66 179630.48 4230350.85 41 48 29.5619 41.8082116 2 9 42.6599 2.1618500 430378.09 4628822.03 033 31.56 -0.55877 0.99965965 902.23 50.316 851.92 astrx_027 4758389.34 179653.59 4230483.90 41 48 35.3543 41.8098206 2 9 43.8567 2.1621824 430407.44 4629000.40 033 30.82 -0.55856 0.99965960 901.96 50.318 851.65 astrx_028 4758431.23 179529.01 4230403.26 41 48 32.6037 41.8090566 2 9 38.3955 2.1606654 430280.60 4628916.80 033 34.43 -0.55957 0.99965981 875.90 50.316 825.58 astrx_029 4758366.64 179531.23 4230461.93 41 48 35.4136 41.8098371 2 9 38.5972 2.1607214 430286.10 4629003.41 033 34.33 -0.55954 0.99965980 866.97 50.318 816.65 astrx_030 4758509.85 179684.03 4230370.89 41 48 29.9983 41.8083329 2 9 44.9775 2.1624938 430431.69 4628834.96 033 30.02 -0.55834 0.99965955 917.24 50.316 866.93 astrx_031 4758310.98 179752.40 4230568.84 41 48 39.0175 41.8108382 2 9 48.2621 2.1634061 430510.19 4629112.38 033 27.93 -0.55776 0.99965942 903.00 50.320 852.68 astrx_032 4758283.18 179868.99 4230624.09 41 48 40.8570 41.8113492 2 9 53.3542 2.1648206 430628.23 4629167.97 033 24.55 -0.55682 0.99965922 922.42 50.321 872.10 astrx_033 4758239.21 179851.09 4230648.83 41 48 42.4186 41.8117829 2 9 52.6513 2.1646254 430612.48 4629216.29 033 25.04 -0.55695 0.99965924 905.66 50.322 855.33 astrx_034 4758199.24 179749.74 4230656.22 41 48 43.5429 41.8120953 2 9 48.3296 2.1634249 430513.11 4629251.94 033 27.93 -0.55776 0.99965941 877.96 50.321 827.64 astrx_035 4758242.72 179692.01 4230603.33 41 48 41.3736 41.8114927 2 9 45.7595 2.1627110 430453.15 4629185.61 033 29.62 -0.55823 0.99965952 873.46 50.321 823.14 astrx_037 4758298.34 179663.04 4230539.94 41 48 38.6651 41.8107403 2 9 44.4144 2.1623373 430421.31 4629102.38 033 30.49 -0.55847 0.99965957 871.81 50.320 821.49 D01 4758417.54 179728.85 4230477.31 41 48 34.5250 41.8095903 2 9 47.0682 2.1630745 430481.29 4628974.10 033 28.67 -0.55796 0.99965947 920.70 50.318 870.38 D02 4758450.10 179714.47 4230447.62 41 48 33.1169 41.8091991 2 9 46.3927 2.1628869 430465.28 4628930.83 033 29.11 -0.55809 0.99965950 924.75 50.317 874.43 D03 4758462.47 179744.12 4230432.64 41 48 32.4635 41.8090176 2 9 47.6558 2.1632377 430494.23 4628910.39 033 28.26 -0.55785 0.99965945 924.81 50.317 874.49 D04 4758454.76 179730.51 4230442.82 41 48 32.8871 41.8091353 2 9 47.0793 2.1630776 430481.06 4628923.58 033 28.65 -0.55796 0.99965947 925.47 50.317 875.15 D05 4758459.04 179710.07 4230430.49 41 48 32.5136 41.8090316 2 9 46.1874 2.1628298 430460.37 4628912.27 033 29.24 -0.55812 0.99965951 919.87 50.317 869.55 D06 4758511.06 179689.23 4230370.91 41 48 29.9684 41.8083245 2 9 45.2005 2.1625557 430436.83 4628833.99 033 29.87 -0.55830 0.99965955 918.31 50.316 867.99 D07 4758518.42 179785.79 4230370.31 41 48 29.7162 41.8082545 2 9 49.3678 2.1637133 430532.91 4628825.28 033 27.09 -0.55752 0.99965938 926.10 50.316 875.79 D08 4758528.63 179828.99 4230357.60 41 48 29.1534 41.8080982 2 9 51.2213 2.1642281 430575.51 4628807.50 033 25.85 -0.55718 0.99965931 926.46 50.316 876.14 D09 4758507.37 179855.69 4230368.17 41 48 29.8458 41.8082905 2 9 52.4118 2.1645588 430603.19 4628828.59 033 25.06 -0.55696 0.99965926 918.42 50.316 868.10 D10 4758402.08 179869.70 4230498.52 41 48 35.2561 41.8097934 2 9 53.1903 2.1647751 430622.77 4628995.27 033 24.60 -0.55683 0.99965923 927.28 50.319 876.96 D11 4758318.65 179906.92 4230595.02 41 48 39.3581 41.8109328 2 9 54.9382 2.1652606 430664.33 4629121.39 033 23.48 -0.55652 0.99965916 930.53 50.320 880.21 D12 4758329.37 179919.62 4230579.15 41 48 38.7329 41.8107591 2 9 55.4702 2.1654084 430676.42 4629101.99 033 23.12 -0.55642 0.99965914 928.29 50.320 877.97 D13 4758294.48 180024.09 4230626.74 41 48 40.5504 41.8112640 2 10 0.0495 2.1666804 430782.62 4629157.02 033 20.08 -0.55558 0.99965895 936.97 50.321 886.65 D14 4758326.85 180016.74 4230583.88 41 48 38.8224 41.8107840 2 9 59.6780 2.1665772 430773.53 4629103.81 033 20.31 -0.55564 0.99965897 932.30 50.321 881.98 D15 4758267.10 180073.96 4230654.54 41 48 41.7722 41.8116034 2 10 2.2528 2.1672924 430833.82 4629194.21 033 18.63 -0.55517 0.99965887 936.51 50.322 886.19 D16 4758268.24 180087.91 4230648.57 41 48 41.5922 41.8115534 2 10 2.8551 2.1674598 430847.67 4629188.52 033 18.22 -0.55506 0.99965884 933.78 50.322 883.46 D17 4758140.07 180106.86 4230780.91 41 48 47.5406 41.8132057 2 10 3.8855 2.1677460 430873.22 4629371.74 033 17.60 -0.55489 0.99965880 927.08 50.324 876.76 D18 4758143.15 180157.24 4230782.33 41 48 47.4669 41.8131852 2 10 6.0608 2.1683502 430923.38 4629368.98 033 16.15 -0.55449 0.99965872 931.74 50.324 881.42 D19 4758144.60 180174.08 4230780.98 41 48 47.3893 41.8131637 2 10 6.7877 2.1685521 430940.13 4629366.43 033 15.66 -0.55435 0.99965869 932.40 50.324 882.07 D20 4758158.90 180200.57 4230760.62 41 48 46.5671 41.8129353 2 10 7.9110 2.1688642 430965.80 4629340.82 033 14.91 -0.55414 0.99965864 930.22 50.323 879.90 D43 4758433.57 179505.04 4230388.51 41 48 32.2164 41.8089490 2 9 37.3544 2.1603762 430256.47 4628905.09 033 35.12 -0.55976 0.99965985 867.15 50.316 816.83 D44 4758252.53 179692.08 4230591.99 41 48 40.8877 41.8113577 2 9 45.7465 2.1627074 430452.71 4629170.63 033 29.62 -0.55823 0.99965952 873.21 50.320 822.89 Elipsoidal Height Meridian Convergence Point ID UTM ECEF Marker Latitude Longitude Technical analysis 29 2.2.4 Final survey area bounds The area to be covered with STAMP was readjusted to fit the GCP network as depicted in Figure 9. Figure 9 GCPs and aerial survey area 30 STAMP Project The 500 metres radius is depicted in red colour, within the circle you can see the area to be covered by the aerial pictures which is bounded by the green line so finally the planed flight looked as shown in Figure 10. Figure 10 DMP flight path in white colour. Note that not all GCPs are visible in the image. 2.2.5 STAMP slave deployment rationale The very minimum of slave DMPs that can be used to georeference an orthophoto is 5, assuming a good / even distribution over the working area, and the maximum number is typically considered as 10. More than 10 GCPs is considered an overkill. To ensure that the metric results obtained with the DMPs were not affected by the errors inherent to the photogrammetric processes it was decided that all the DMPs must be positioned near existing GCPs. To avoid extrapolation errors the DMPs silhouettes located in the exterior ring of the GCP network were always positioned towards the internal side of the ring. 2.2.6 STAMP slave availability Due to budgetary constraints it was decided that it was not possible to purchase 10 DMP to operate as STAMP slaves so an alternative was to be found. Technical analysis 31 In the visibility tests paper silhouettes were used but there was always a person to watch the test subjects so wind or humidity was not a real problem but in the field test at CATUAV elements could have been less forgiving. The best solution was to draw the DMP silhouette in a 2D CAD (as seen in Figure 11) and then order to a third party company to cut a 2 mm thick stainless steel plate using laser to create the 10 DMP test subjects. Figure 11 2D CAD drawing of the DMP. The end result was 10 consistent and solid metallic plates that were not susceptible to wind, humidity or other external factors. 2.2.7 STAMP field survey After a failed attempt on May 10th 2018 due to rainy weather conditions a successful attempt was performed on May 17th 2018. The first step was to deploy the 33 pre-existing GCPs in addition to the 10 metallic silhouettes simulating the DMP accompanied by the 10 Rokubun Argonaut GNSS receivers plus the corresponding batteries. The DMPs were distributed as follows: 32 STAMP Project Table 5 GCP and Argonaut serial number, power on and power off UTC times. In the example picture below can be seen the relative size of a 50x50 cm GCP set on the ground and centred over the D15 surveyed point besides a DMP silhouette equipped with an Argonaut receiver on top of the metallic plate. The Mazda 2 car (small) may serve as a reference to gauge the proportions of the elements in the image. Figure 12 D15 GCP, DMP stainless steel silhouette and car. After all the GCPs and DMPs were deployed the DMP that was going to be used to perform the photogrammetric flight was prepared at CATUV facilities and was launched from the roof of the building. In 23 minutes and 26 seconds the drone captured 518 aerial pictures in Adobe digital negative format also known as DNG based on tiff version 6 open standard. On average the image size was 23.2 Megabytes for a total of 11.7 Gigabytes. GCP ID Serial Number Power on Power Off Astrx30 261017-1-0006 8:45:35 12:11:35 Astrx32 261017-1-0056 9:01:05 12:17:10 Astrx34 261017-1-0021 7:36:35 14:23:00 D01 261017-1-0208 7:55:40 23:59:59 D09 261017-1-0201 8:53:05 12:13:05 D14 261017-1-0121 9:13:10 17:13:00 D15 261017-1-0055 9:23:05 12:19:40 D17 261017-1-0051 9:48:30 13:44:55 D20 261017-1-0048 9:40:10 13:40:30 D43 261017-1-0013 7:21:05 14:31:25 Technical analysis 33 2.2.8 Argonauts GNSS post processing The settings used to post-process the 10 Argonauts over the DMP silhouettes were:  Constellations: GPS and GLONASS  Frequencies: L1  Epoch sampling: 1Hz  Positioning mode: static  Solution: forward + reverse (combined)  Elevation mask: 10 degrees  SNR mask: 35 dB-Hz  Dynamics estimation: disabled  Tide corrections: disabled  Ionospheric model: as broadcasted  Troposphere model: Saastamoinen  Ephemeris: as broadcasted  GPS ambiguity resolution: fix and hold  GLONASS ambiguity resolution: disabled  Ambiguity resolution threshold value: 3  SBAR reference position: 41º58’48.38852’’ 2º10’27.45507’’ 938.003 m  Datum transformation: none  Geoid: none  Antenna phase centre variation files: none The resulting ETRS89 geodesic coordinates (latitude, longitude and ellipsoidal height) are: Table 6 DMP coordinates to be used in the photogrammetric coordinates. 2.2.9 Photogrammetric processing The software used to process the aerial pictures and merge the result with the coordinates of the GCPs is Agisoft PhotoScan Professional version 1.4.2. (APS) GCP ID Serial Number Latitude Longitude Height Astrx30 261017-1-0006 41.808317973 2.162578191 918.33 Astrx32 261017-1-0056 41.811337521 2.164821013 922.72 Astrx34 261017-1-0021 41.812091129 2.163437401 878.13 D01 261017-1-0208 41.809616481 2.163097989 920.72 D09 261017-1-0201 41.808290308 2.164522570 919.40 D14 261017-1-0121 41.810795306 2.166601352 932.69 D15 261017-1-0055 41.811607963 2.167262006 936.42 D17 261017-1-0051 41.813209189 2.167793222 927.24 D20 261017-1-0048 41.812929946 2.168843918 930.31 D43 261017-1-0013 41.808950181 2.160389115 867.31 34 STAMP Project This software package is known to be computer efficient because it uses multi core CPUs as well as Nvidia GPU CUDA acceleration. This kind of optimization is very relevant when facing the processing of 518 raw images in a single work station. The software also provides a good combination of flexibility and automation which is very convenient to perform the different comparisons that are required to assess STAMP suitability. It is worth noting that typically in a photogrammetric work the two desired final deliverables are the Orthomosaic and the Digital Surface Model (DSM) but in this thesis we will focus on the resulting accuracy of the nonlinear least-squares algorithm used in the bundle adjustment of the aerial triangulation which is the first step required in the photogrammetric workflow. APS can perform calculations and provide products like textured 3D models or Digital Terrain Models (DTM) among others that will not be represented in this thesis. Figure 13 workflow chart depicts the steps within APS used to compute the GCP coordinates and errors as well as to export the typical outputs expected from a photogrammetric project even though those are not essential to fulfil the objectives of this thesis. Technical analysis 35 Figure 13 APS flowchart Import imagery in DNG format with geotags Align photos with pre selection enabled Build dense cloud with colour Computed photo coordinates Build Digital Elevation Model (DEM) Build Orthomosaic Import GCP coordinates Identify the GCPs on the Orthomosaic Refine automated GCPs on the pictures Optimize camera alignment Computed GCP coordinates Build dense cloud Dense cloud with point colour Build Digital Elevation Model (DEM) Digital Surface Model Build Orthomosaic Orthomosaic 36 STAMP Project Two different processings were performed within APS: 1. The aerial triangulation bundle adjustment is based on the GCPs from Setat survey and the Septentrio survey together to compute an optimized camera alignment. STAMP GCPs are adjusted as “check points” as an output of the aerial triangulation and only serve the purpose of evaluating the errors. 2. The aerial triangulation bundle adjustment is based on the 10 STAMP GCPs. Setat GCPs and the Septentrio GCPs are adjusted as “check points” as an output of the aerial triangulation and only serve the purpose of evaluating the errors. 2.2.9.1 Setat and Septentrio GCPs camera optimization The outcome of this calculation allows us to see how diverges the different postprocessed L1 only GNSS solutions from the photogrammetric solution based only on a heavily redundant network of GCPs measured with high end equipment. The expectation is to easily identify any wrong GNSS carrier phase integer cycle fixings in the low cost L1 only solutions of the DMPs because the other GCPs will act as a truth. Commercial study 43 3.2 Analysis and segmentation The professional drone market is typically segmented in the verticals listed in Figure 17 Figure 17 UAV market verticals STAMP is mainly oriented at the “survey and terrain mapping” vertical but it may also have use in cinematography to create 3D models to be used in movies, infrastructure monitoring, public safety to reconstruct car accidents, precision agriculture to level the terrain to avoid pond formation, and construction stock pile volume management. 3.3 The Competition There seems to be an increase in the number of professional drone systems that come equipped with survey grade dual frequency RTK GNSS receivers to provide high accuracy positioning to the drone capturing aerial pictures with the objective to perform direct georeferencing of the images captured by the drone. Those high end systems come at a very high premium cost and require the end user to setup a local base station with a radio link continuously operating that must be kept alive for the whole duration of the flight as otherwise the pictures may not be georeferenced correctly. The high end receiver approach also has the inconvenience of draining the drone battery and adding weight / complexity to the system. RTK receivers do not represent a real competence to STAMP because there is no real need to obtain the pictures with a precise positional stamp right in the field and the complexity of the RTK setup unnecessarily increases the system setup time 23% 11% 10% 9% 7% 6% 5% 5% 5% 4%3% 11% Photography and / or video Surveying and terrain mapping Cinematography Real State Asset or infrastructure inspection & monitoring Testing and research and development Education or training Public safety or first responders Agriculture or farming 44 STAMP Project The consequences of a possible drone crash may cause catastrophic failures in very expensive equipment making a repair a prohibitive endeavour. The pricing for two relevant competitors is:  The Trimble UX5-HP UAS (RTK capable) costs around $40000 not including GNSS post processing software nor the GNSS base station nor the photogrammetric software  The Sensefly eBee costs between $25000 and $27000 depending on the options but it does includes the Pix4D photogrammetric software license which is valued in 6500€ but it does not include the GNSS post processing software nor it includes GNSS base station. During the development of the STAMP system we have not identified any direct competitor using drones as GCPs however we have located a single indirect competitor (see Table 10) Company Name: Propeller Aerobotics Pty. Ltd. (or PropellerAero for short) Registration date: 08-May-2014 Employees: 24 based in Australia, 5 in the US and 3 in Canada Twitter followers: 932 Linked in followers: 987 Table 10 Competitor basic company details. The competing product is a battery powered GCP (see Figure 18 AeroPoint GCP) with an L1 only GNSS receiver integrated in the centre of the target with logging capability (see Table 11) Product Name: AeroPoints Release date: August 2016 Patents: Australian patent number AU2015903258 Patent file date: 13-August-2015 Price: 6000 US dollars per a group of ten AeroPoints Table 11 Competitor product basic information. Commercial study 45 Figure 18 AeroPoint GCP The bundle includes one year of post processing in PropellerAero own cloud based SaaS platform and every additional year costs $600. We believe that their product does have some drawbacks that are worth mentioning: 1. AeroPoints must be manually spread over the survey area. 2. Each AeroPoint weight is 1.55 kg and measures 54.4x54.4x3.2 cm so the weight of 10 would be around 15.5 kg and the size 54.4x54.4x32 cm that suggests that a single operator is unlikely to carry them all in his backpack in addition to the drone, the batteries, the remote control and tablet. It looks like AeroPoint GCPs have been designed to be carried in a vehicle, in fact the corporative web page only lists applications where the Aeropoints are being spread by a wheeled vehicle. 3. The raw data from the AeroPoints is not exportable in open standard formats like RINEX, in fact the end users are forced to use a mobile app to directly upload the files from AeroPoint to PropellerAero servers. This means that the end users will never be able to handle the files exported by the AeroPoints. This implies that end users are forced to pay the $600 a year if they want to obtain the coordinates with the embedded GNSS receiver. 4. There is no official representation outside Australia and US, it may be difficult / expensive to repair. 5. Spain is being served from the Netherlands, so PropellerAero presence in southern Europe is scarce at best. 3.4 Comparative product analysis In the end of degree project [4] there is a profitability comparison between a Trimble UX5 UAV and a land surveying of the same area performed by high end GNSS receiver located. The area in which the test was performed is mostly open fields therefore it is the ideal location to execute a photogrammetric survey: 46 STAMP Project Trimble UX5 UAV Trimble R10 GNSS receiver 3120 hectares per year 21213 € 10923 € 4320 hectares per year 49247 € 12581 € Table 12 Industrial benefit over the course of a year As per Table 12 from company revenue perspective the UAV survey provides almost a 200% improvement over a GNSS survey in the small test field and almost a 400% improvement in a test field that is just 140% larger. The conclusion is that in any job where the features to survey are visible from the sky the survey work should be tackled with UAV instead of classic techniques, at least from profitability point of view. 3.5 Revenue Streams STAMP chosen drone Manufacturer's Suggested Retail Price (MSRP) would be set as per the following estimative tables. The minimum hardware required to supply a fully working STAMP system would be: Table 13 Entry level STAMP system hardware list The hardware required to provide a top end STAMP system would be: 30/1/2018 Brand Item description MSRP Unit Price Number MSRP Total Margin Rokubun's cost Margin DJI Mavic Pro Platinum + battery + remote controller 1 300.00 € 1 1 300.00 € 5.00% 1 235.00 € 65.00 € DJI Mavic Pro Platinum extra battery 100.00 € 3 300.00 € 5.00% 285.00 € 15.00 € DJI Mavic Pro Platinum 2 propellers 10.00 € 1 10.00 € 5.00% 9.50 € 0.50 € DJI Standalone Mavic Pro + battery (no remote no charger) 900.00 € 5 4 500.00 € 5.00% 4 275.00 € 225.00 € DJI Mavic Pro extra battery 100.00 € 1 100.00 € 5.00% 95.00 € 5.00 € DJI Mavic Pro 2 propellers 10.00 € 1 10.00 € 5.00% 9.50 € 0.50 € DJI Mavic Pro 4 way battery bays 45.00 € 2 90.00 € 5.00% 85.50 € 4.50 € DJI Mavic Pro 50w battery charger 22.00 € 2 44.00 € 5.00% 41.80 € 2.20 € DJI AC battery charger cable 5.00 € 2 10.00 € 5.00% 9.50 € 0.50 € DJI Propeller protectors 180.00 € 6 1 080.00 € 5.00% 1 026.00 € 54.00 € DJI High brightnes CrystalSky 5.50 inch tablet 469.00 € 1 469.00 € 5.00% 445.55 € 23.45 € DJI High brightnes CrystalSky 7.85 inch tablet 699.00 € 0 - € 5.00% - € - € DJI Ultrabright CrystalSky 7.85 inch tablet 1 149.00 € 0 - € 5.00% - € - € DJI CrystalSky remote holder 90.00 € 1 90.00 € 5.00% 85.50 € 4.50 € DJI SanDisk Extreme microSD 64GB 34.00 € 1 34.00 € 5.00% 32.30 € 1.70 € PGYTECH Landing gear extension for Mavic Pro 25.00 € 6 150.00 € 0.00% 150.00 € - € MicroRaptor Pro Large Duffel Bag / Backpack - With Paded Divider 225.00 € 1 225.00 € 0.00% 225.00 € - € MightySkins DJI Mavic Pro Quadcopter Drone Skin 30.00 € 5 46.00 € 0.00% 46.00 € - € Rokubun SPA 450.00 € 6 2 700.00 € 66.67% 900.00 € 1 800.00 € 11 158.00 € 19.73% 8 956.15 € 2 201.85 € Providers Entry Configuration Commercial study 47 Table 14 Complete STAMP system hardware list. In between the two extreme configurations presented the end user can decide the options to include in his system based on the needs regarding accuracy, extension of the areas to be mapped and funds availability. STAMP is designed to be upgradeable so if the end user decides at a later date that he needs a 7 drone system instead of the basic 5 drone system he just needs to purchase the two missing drones. Because Rokubun is a company born to develop GNSS navigation algorithms, the intent of STAMP is to open another access point for customers to use the Rokubun’s post processing engine. STAMP was conceived as a sustainable business that should provide a moderate commercial benefit because the main revenues are expected to come from the PaaS post-processing engine. Rokubun’s PaaS post processing engine still is in test phase so the final pricing has not yet been decided but most likely Rokubun will charge customers by the processing mode selected, which determines the result final accuracy. The more accurate the result obtained from PaaS is, the more expensive the processing should be and the other way around. This pricing strategy ensures: 1. Power users of the PaaS do not pay more than 100€ per processing without having to worry about monthly fees. 2. The payment is small enough so that it does not become an entry barrier, especially when compared to $3000~$6000 software packages. 3. It would take 120 days to reach 3000 € mark making Rokubun’s PaaS much more budget friendly than commercial software packages. 30/1/2018 Brand Item description MSRP Unit Price Number MSRP Total Margin Rokubun's cost Margin DJI Mavic Pro Platinum + battery + remote controller 1 300.00 € 1 1 300.00 € 5.00% 1 235.00 € 65.00 € DJI Mavic Pro Platinum extra battery 100.00 € 4 400.00 € 5.00% 380.00 € 20.00 € DJI Mavic Pro Platinum 2 propellers 10.00 € 1 10.00 € 5.00% 9.50 € 0.50 € DJI Standalone Mavic Pro + battery (no remote no charger) 900.00 € 10 9 000.00 € 5.00% 8 550.00 € 450.00 € DJI Mavic Pro extra battery 100.00 € 1 100.00 € 5.00% 95.00 € 5.00 € DJI Mavic Pro 2 propellers 10.00 € 1 10.00 € 5.00% 9.50 € 0.50 € DJI Mavic Pro 4 way battery bays 45.00 € 4 180.00 € 5.00% 171.00 € 9.00 € DJI Mavic Pro 50w battery charger 22.00 € 4 88.00 € 5.00% 83.60 € 4.40 € DJI AC battery charger cable 5.00 € 4 20.00 € 5.00% 19.00 € 1.00 € DJI Propeller protectors 180.00 € 11 1 980.00 € 5.00% 1 881.00 € 99.00 € DJI High brightnes CrystalSky 5.50 inch tablet 469.00 € 0 - € 5.00% - € - € DJI High brightnes CrystalSky 7.85 inch tablet 699.00 € 0 - € 5.00% - € - € DJI Ultrabright CrystalSky 7.85 inch tablet 1 149.00 € 1 1 149.00 € 5.00% 1 091.55 € 57.45 € DJI CrystalSky remote holder 90.00 € 1 90.00 € 5.00% 85.50 € 4.50 € DJI SanDisk Extreme microSD 64GB 34.00 € 1 34.00 € 5.00% 32.30 € 1.70 € PGYTECH Landing gear extension for Mavic Pro 25.00 € 11 275.00 € 0.00% 275.00 € - € MicroRaptor Pro Large Duffel Bag / Backpack - With Paded Divider 225.00 € 1 225.00 € 0.00% 225.00 € - € MightySkins DJI Mavic Pro Quadcopter Drone Skin 30.00 € 10 311.00 € 0.00% 311.00 € - € Rokubun SPA 450.00 € 11 4 950.00 € 66.67% 1 650.00 € 3 300.00 € 20 122.00 € 19.97% 16 103.95 € 4 018.05 € Providers Highest specification system 48 STAMP Project 4. Sporadic users like small land surveying companies pay a predictable, easy to compute figure helping them to forecast their operational costs. 5. As per our market survey the proposed price for Rokubun’s target market, the L1 fixed integer ambiguity resolution (45€) is the highest that we believe the market is ready to accept. 6. The PaaS positioning solutions that can be obtained from other sources at a low cost is also provided at low cost from the PaaS even though long observations may lead to a reasonably accurate result (like PPP). The PaaS payment would be completely automated. All what the customer would be required to do is to select one of the price points listed in Figure 19. Figure 19 PaaS pricing price evolution. After selecting the desired accuracy the uploaded raw data file it would be parsed to determine the first and last observation times. Three different situations may arise as per L1 SPP L1L2 SPP L1 DGNSS L1L2 DGNSS L1 PPP L1L2 PPP L1 only float L1L2 float L1 only fixed L1L2 fixed Pricing €1.00 €2.00 €3.00 €4.00 €5.00 €6.00 €30.00 €40.00 €45.00 €50.00 €- €5.00 €10.00 €15.00 €20.00 €25.00 €30.00 €35.00 €40.00 €45.00 €50.00 PaaS price per every 16 field collection hours Commercial study 49 Raw data falls within payed period? Higher accuracy requested ? User pays: Yes No Nothing because he already paid for higher accuracy. Yes Difference between what has been paid and the new accuracy No Irrelevant Pays full price as per previous graph. Table 15 Raw data file temporal range fitness. Because customer accounts at Rokubun’s PaaS would contain the encrypted banking details all the usage charges would be instantaneously executed through a secure e-banking platform While it is in Rokubun’s best interest to ease as much as possible the transactions it is also important to ensure that there are no mistakes from the end user side so the end user would have to confirm each processing to ensure that is aware of the costs involved. 3.6 Supply chain The supply chain of STAMP is relatively simple because it only involves products manufactured in house and products manufactured or stoked by DJI. The Argonaut / SPA receivers are manufactured by Rokubun and Rokubun can adapt the production rate as well as stoking at will. 𝑢−𝑏𝑙𝑜𝑥 𝑀𝑎𝑥𝑡𝑒𝑛𝑎 ⋮ 𝑒𝑙𝑒𝑐𝑡𝑟𝑜𝑛𝑖𝑐 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝑛}→𝑒𝑙𝑒𝑐𝑡𝑟𝑜𝑛𝑖𝑐𝑠 𝑂𝐸𝑀 𝑚𝑎𝑛𝑢𝑓𝑎𝑐𝑡𝑢𝑟𝑒𝑟 →𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑠𝑡𝑜𝑟𝑎𝑔𝑒→𝑑𝑒𝑎𝑙𝑒𝑟𝑠→𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟𝑠↔𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑃𝑎𝑎𝑠 𝑤𝑒𝑏 𝑓𝑟𝑜𝑛𝑡𝑒𝑛𝑑 Figure 20 Argonaut and SPA supply chain PaaS is mostly software (except the servers) and does not really require of a supply chain understanding the term in the classical sense. 𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑠𝑜𝑢𝑟𝑐𝑒 𝑐𝑜𝑑𝑒 →𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑛𝑒𝑡𝑤𝑜𝑟𝑘 𝑟𝑒𝑝𝑜𝑠𝑖𝑡𝑜𝑟𝑦→𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑃𝑎𝑎𝑆 𝑐𝑜𝑚𝑝𝑢𝑡𝑎𝑡𝑖𝑜𝑛𝑎𝑙 𝑠𝑒𝑟𝑣𝑒𝑟 ↔𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑃𝑎𝑎𝑠 𝑊𝑒𝑏 𝐹𝑟𝑜𝑛𝑡𝑒𝑛𝑑↔𝑒𝑛𝑑 𝑢𝑠𝑒𝑟 𝑟𝑎𝑤 𝑑𝑎𝑡𝑎 𝑓𝑖𝑙𝑒𝑠 Figure 21 PaaS supply chain STAMP supply chain, while not controlled by Rokubun, it is mainly driven by a very large manufacturer that has a strong presence in the market, the risk of 50 STAMP Project shortages is very mild, the only exception being when a new product is released but this is not the case of the DMP/DMPP. 𝐷𝐽𝐼 𝑀𝑎𝑣𝑖𝑐 𝑃𝑟𝑜 𝐷𝐽𝐼 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 1 ⋮ 𝐷𝐽𝐼 𝑐𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝑛 𝑃𝐺𝑌𝑇𝐸𝐶𝐻 𝑀𝑖𝑐𝑟𝑜𝑅𝑎𝑝𝑡𝑜𝑟 ⋮ } →𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑜𝑓𝑓𝑖𝑐𝑒 ℎ𝑎𝑟𝑑𝑤𝑎𝑟𝑒 𝑠𝑡𝑜𝑟𝑎𝑔𝑒 +𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑆𝑇𝐴𝑀𝑃 𝑚𝑜𝑏𝑖𝑙𝑒 𝑎𝑝𝑝.→𝑑𝑒𝑎𝑙𝑒𝑟𝑠→𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟𝑠↔𝑅𝑜𝑘𝑢𝑏𝑢𝑛 𝑃𝑎𝑎𝑆 𝑤𝑒𝑏 𝑓𝑟𝑜𝑛𝑡𝑒𝑛𝑑 Figure 22 STAMP supply chain 3.7 Key Partners The main partnership that Rokubun needs to successfully initialize the STAMP project is with Dà-Jiāng Innovations Science and Technology Co., Ltd (Chinese: 大疆创新科技有限公司; doing business as DJI) headquartered in Shenzhen, Guangdong. This partnership is central to Rokubun’s STAMP system because the factory default firmware loaded in DJI COTS drones and more specifically on the Mavics does not allow certain operations that are required to successfully deploy slave drones. As of today, there is only a pending “show stopper” regarding the STAMP project and that is DJI drone firmware and API capabilities: 1. As per DJI manuals and API documentation on their web pages it clear that the firmware and API do not support more than one single drone paired with the radio control at the same time. This means that it would not be possible to fly multiple drones at the same time. At the light of the current legal texts this is a desirable limitation. 2. Only one single drone can be paired with the remote control at any given time this means that a pilot would be unable to retrieve the slave drones manually denying the benefits of the STAMP system. 3. The current DJI Mavic firmware does not include a SubSecTime field in the JPEG EXIF tag. As a consequence the time stamp of the pictures taken by the STAMP master drone temporal resolution is only at second level, it does not reach millisecond. This would deny the possibility to install a Rokubun GNSS receiver in the master drone because it would not be possible to accurately correlate the aerial pictures with the GNSS data. 4. DJI drones, to the best of our knowledge, are not capable to perform flight plans that include landing at supplied coordinates, powering off the motors (to save battery while on the ground) and automatically take off after a pre-set time. This missing feature is important when sending STAMP slave drones to areas where there is no possibility to have a viable radio-link due to obstructions caused by terrain, vegetation and similar obstructions. Commercial study 51 Rokubun only requires from DJI a the few adaptations on their COTS Mavic drone firmware and API listed above, Some of those software adaptations have deep implications on the behaviour of the drones allowing them to perform manoeuvres that may cross the line of legality. Up until today DJI has been selling generic drones into the consumer and prosumer market that due to the quality, availability, robustness, size, weight and low cost have been adapted to professional applications like videography and photogrammetry. DJI started their drone company by segmenting their market horizontally however in the recent years there has been a change of strategy and DJI is starting to target solutions to specific market verticals like agriculture. This specialization in DJI products can be perceived as a company that is providing hardware to the Original Equipment Manufacturer (OEM) market but at the same time may also be interested in releasing products to the same end users that Rokubun would. As of today DJI is not a STAMP competitor because it has never released any product specifically tailored to photogrammetry however the after mentioned precedent makes difficult for us to evaluate DJI’s future intentions. DJI has already warranted Rokubun an industrial discount over 5% on all the products we need for the STAMP but those discounts oscillate on the total lump sum of the material requested. 3.8 Risk Mitigation The following risk analysis table summarises several key risks related to STAMP project which are identified, classified and a mitigation strategy is presented. 52 STAMP Project Table 16 Risk mitigation table Table 16 demonstrates that there are show stopper issues that do require addressing before attempting to build a STAMP prototype. Because DJI is the STAMP key supplier of drones and those are limited by what can be done with their firmware, before any further attempt to continue with the project it should be clarified what is DJI willingness to help Rokubun by introducing the required changes. 3.9 Cost Structure & Financial Projection In this section, a review of what are the most important costs to operate the business, what are the most expensive key activities and resources, what are the fixed costs, what are your variable costs and what are the sources of income projected for the following three years is presented. Risk Impact (e.g. Cost, Schedule, Technical) Likelihood of occurrence Severity of consequences Risk Magnitude Precaution Measures Mitigation plan Cost of risk mitigation Multiple Mavic controller pairing firmware modification Technical 5 Show stoper Very high None can be taken A different, more customizable drone low Absence of milliseconds in the JPEG EXIF Technical 5 Show stoper Low Preliminary study on using second resolution time stamp We suspect that the impact can be severly mitigated by using the API low Flight plan aplicability after landing Technical 4 Limited Moderate Pending on DJI response. None, the targeted market size impact is likely to be small low Currency fluctuations may impact the cost of the STAMP Cost 2 Low Low None, stocking is not an option for Rokubun Item price will be updated on every customer enquiry low Some components may become unavailable Schedule 2 Low Low None, stocking is not an option for Rokubun Issue long lead times and use other vendors (at a higher cost) moderate