Development and analysis of Laser Scanner data

TREBALL DE FI DE CARRERA TÍTLE: Development and analysis of Laser Scanner data DEGREE: Technical Telecommunication Engineering Speciality: Telematics AUTHOR: Emma Riera Wilson DIRECTOR: Michele Crosetto SUPERVISOR: Jaume Piera Fernández DATE: 18th February 2008 TÍTLE: Development and analysis of Laser Scanner data DEGREE: Technical Telecommunication Engineering Speciality: Telematics AUTHOR: Emma Riera Wilson DIRECTOR: Michele Crosetto SUPERVISOR: Jaume Piera Fernández DATE: 18 th February 2008 Overview Terrestrial Laser Scanning offers an alternative to traditional survey techniques and consists of high speed data capture. This technique is very recent and is still being studied. The aim of this project is to study the capabilities of Terrestrial Laser Scanners for Land Deformation Monitoring. This document firstly reviews the basic functioning of Terrestrial Laser Scanning technique and device. Furthermore it describes in detail the experimental work carried out at the Institute of Geomatics (IG). TÍTOL: Elaboració i anàlisi de dades de Laser Scanner DEGREE: Enginyeria Tècnica de Telecominicació Especialitat: Telemàtica AUTOR: Emma Riera Wilson DIRECTOR: Michele Crosetto SUPERVISOR: Jaume Piera Fernández DATA: 18 de febrer de 2008 Resum Terrestrial Laser Scanner ofereix una alternativa per tècniques tradicionals de medició que consisteix en medició de dades a gran velocitat. Aquesta tècnica és molt jove i encara s’està estudiant. L’objectiu d’aquest projecte és estudiar les capabilitats de Terrestrial Laser Scanner per a Deformacions en el terreny. Aquest document primer estudia el funcionament bàsic de la tècnica i l’aparell de Terrestrial Laser Scanner. Després descriu amb detall el treball experimental fet a l’Institud de Geomàtica (IG). ÍNDEX INTRODUCTION ............................................................................................... 1 CHAPTER 1: WORKING PRINCIPLES ............................................................ 2 1.1.Instrument description .......................................................................................................... 2 1.2.TLS classification ................................................................................................................. 5 1.2.1.Distance measurement principle .......................................................................... 5 1.2.1.1.Time of Flight ....................................................................................... 5 1.2.1.2.Phase Difference .................................................................................. 6 1.2.1.3Optical Triangulation .............................................................................. 7 1.2.2.Deflection of laser beam ....................................................................................... 9 1.2.2.1.Camera view ................................................................. .......................9 1.2.2.2.Panoramic view ..................................................................................... 9 1.2.2.3.Hybrid view ......................................................................................... 10 1.3.Classification by technical properties ................................................................................. 10 CHAPTER 2: DATA PROCESSING................................................................ 11 2.1.Data processing ................................................................................................................ 11 2.1.1.Data acquisition ................................................................................................. 11 2.1.2.Data treatment ................................................................................................... 12 2.1.3.Data visualization ............................................................................................... 13 2.2.Influence of the characteristics of the object (Reflectivity) ................................................. 13 2.3.Influence of environmental conditions ................................................................................ 15 2.4.TLS applications ................................................................................................................. 16 CHAPTER 3: DATA ANALYSIS .................................................................... 20 3.1.Available experimental data..................................................................................................20 3.1.1.Castellfollit village ................................................................................................ 20 3.1.2.Experiment at the Institute of Geomatics ............................................................ 21 3.1.3.Formigal village ................................................................................................... 23 3.2.Data Analysis.........................................................................................................................24 3.3.Matching ............................................................................................................................. 24 3.3.1.Least Squares 3D surface matching .................................................................. 24 3.3.2.Matching analysis ............................................................................................... 25 3.3.3.Precision analysis ............................................................................................... 29 3.4.Residuals vs Photograph ..................................................................................................... 33 3.5.Incidence Angle vs Residuals .............................................................................................. 36 3.6.Analysis of Standard Deviation measurements................................................................. 40 3.6.1.Use of Standard Deviation value ....................................................................... 44 3.7.Study of the influence of the characteristics of the object. ............................................... 49 CONCLUSIONS ............................................................................................. 53 BIBLIOGRAPHY ............................................................................................. 54 ANNEX ............................................................................................................ 56 Introduction 1 INTRODUCTION The aim of this document is to study the capabilities of TLS technology to be used for Land Deformation Monitoring. TLS technology is based on Light Distance and Ranging (LiDAR) and is also referred to as ground-based LiDAR or tripod LiDAR. “LIDAR (Light Detection and Ranging; or Laser Imaging Detection and Ranging) is a technology that determines the distance to an object or surface using laser pulses. Like the similar radar technology, which uses radio waves instead of light, the range to an object is determined by measuring the time delay between transmission of a pulse and detection of the reflected signal. LIDAR technology has application in geology, seismology, remote sensing and atmospheric physics” [14]. The primary capability of TLS is the generation of high resolution 3-dimensional maps and images of objects over scales of metres to kilometres with centimetre or under centimetre precision. This allows for high accuracy mapping as well as the determination of surface changes over time via repeated measurements. The document is divided into three chapters: -WORKING PRINCIPLES: Introduction to the main functioning and types of TLS devices. -DATA: Deals with the collection and treatment of data obtained from TLS devices and it also explains the influence depending on the characteristics of the object. -EXPERIMENTAL PART: This is the most important part of the document. In this chapter different experiments are carried out to study the data obtained from TLS devices, especially for Land Deformation Monitoring. Working principles 2 CHAPTER 1 : WORKING PRINCIPLES Terrestrial Laser Scanning (TLS) offers an alternative to traditional survey techniques and consists of high speed data capture. Terrestrial Laser Scanning works like ordinary radars, except these systems send out narrow pulses or beams of light rather than broad radio waves. A receiver system processes the returning light to calculate the point cloud corresponding to the measured area. TLS systems are an active noncontact range-finding device. The basic principle of these types of devices is to project an optical signal onto an object and then process the reflected or scattered signal to determine the distance. For each reflected signal, scanners collect four measurements: • Two angles (horizontal and vertical) • Distance • Intensity With the two angles and the distance the TLS obtains a polar coordinate for every measured point in the target. Once these polar coordinates have been measured they are immediately transformed into a local 3D Cartesian system (x,y,z). Finally, these internal coordinates are treated and transformed into an external coordinate system by specific software. 1.1. Instrument Description TLSs are high precision systems, capable of working in most real world environments under a variety of conditions. In order to work properly and to have an accurate result, ground-based Lidars must be placed in a way that will guarantee no movements will occur during the measurements. The most reliable way is to mount the Lidar on a tripod, ensuring constant position and orientation in short and middle time-length measurements. Whereas, for long time monitoring a more stable and solid platform is recommended. TLSs use tacheometric measurements [2], which consist of the combination of measured distances to the target and the angles from the device. The principle can be seen in image 1.1: lmage 1.1 - Tacheometric Laser Scanning principle Working principles 3 The laser scanner range finder only detects the distance of one point at a time in its view direction. Thus, the scanner scans its entire field of view (FoV), one point at a time, by changing the laser’s view direction to scan different points. The view direction of the laser scanner can be changed by either rotating the device itself, or by using a system of rotating mirrors. The latter method is the most used, because mirrors are much lighter and can be rotated much faster and with a greater accuracy. The laser Scanner calculates an oblique distance (s’) and two orthogonal angles (w1 and w2) for every measured point in the target surface. With these parameters the TLS defines the position of every point in the space in a polar coordinate system. The measured angles correspond to the horizontal direction (w2) and the vertical direction (w1). These polar coordinates are transformed into an internally defined Cartesian system (x,y,z). The reference of this internal coordinate system is the TLS geometric rotation centre, this point is considered as the zero point (0,0,0). Image 1.2 is a picture of a type of TLS: Image 1.2 - Terrestrial Laser Scanner. RIEGL LMS-Z420i [15] The horizontal scan, also called “frame scan”, is the part responsible for measuring the angle in the horizontal direction (w2). Image 1.2 shows the rotating optical head of the instrument with two vertical windows. It is mounted on a cylinder which connects it to the base tripod. The horizontal scan can have a FoV from 40º up to 360º, depending on the rotation of the complete optical head. Inside the windows of the rotating head there are two mirrors which are responsible for scanning in the vertical direction, also called “line scan”. The mirrors in the rotating head are polygonal with a number of reflective surfaces. The mirrors can be rotating or oscillating depending on the size of the angles it is measuring: Working principles 4 • For high scanning rates and a large vertical scan angle, the polygonal mirror rotates continuously at an adjustable speed. • For slow scanning rates and small scanning angles, the polygonal mirrors oscillate linearly up and down. The vertical scan can have a FoV from 40º up to 310º. The FoV is limited to 310º because of the support that holds the Lidar. Regarding the part that calculates the distance to the target, there are different possible mechanisms that can be used. These mechanisms are explained in the next point. Once the angles and the distance for each point in the point cloud have been measured, they are monitored and recorded with high precision and then transmitted to the storage device. As explained before, these polar coordinates are then transformed into an internal Cartesian coordinate system (x,y,z), which can be later georeferenced into an external co-ordinate system (X,Y,Z) as is shown in image 1.3: Image 1.3 - Co-ordinate Systems [9] In addition, the intensity (i) of the returning distance signal for each point is also stored. The intensity is a measurement of the received energy for each point. The intensity is often called the fourth dimension and it is very useful for the visualization of the target, especially in dense and complex point clouds. To sum up, the Laser scanner creates a point cloud where each point is determined by position (x,y,z) and intensity (i). To add information, digital images can be taken during the data capture. A digital camera may be fixed to the TLS or may be used independently. Photographs give information about the colour and the texture of the object. This information is very useful in measurements where there is, for example, a lot of vegetation, which introduces noise. Colour information can help remove the mentioned noise. Data Processing 11 CHAPTER 2. DATA PROCESSING The following section deals with the collection and treatment of data from TLS instruments. It goes on to explain the effect of measuring different types of target surfaces while scanning, commenting on the effects of texture, colour and orientation based on information extracted from several studies carried out by others. 2.1. Data processing It is a well known fact to people in the field that there are different types of TLS which are designed and developed for specific tasks. However, data processing is treated in a very similar way for all TLSs [2]. In general, each surveying task can be divided into three major steps: • Data acquisition • Data treatment • Data visualization 2.1.1. Data acquisition The first step to process data is data acquisition, which consists of measuring the scene with a TLS to obtain the data. The procedure to obtain the mentioned data is also divided into several steps. Firstly, the area of interest is scanned once or several times. To ensure a full record of the target and to avoid holes in the obtained data, scans from different points of view are recommended. Also various scans in the same position can be used to contrast that the data has been scanned correctly, verifying the precision of this data. Once data has been scanned a point cloud is obtained, where each point represents a 3D coordinate (xyz) of the situation of the object and the intensity of the reflected signal for each point. Finally, all the different scans (point clouds) of the same area are fused in a global one. This procedure involves the matching of the different point clouds. Matching is a procedure that consists on finding each correspondent point in two different acquisitions to match both acquisitions. There are two techniques for matching: • Point based Technique: this is the most precise method, because it uses known shared points from the different scans to carry out the matching. • Area based Technique: this method uses natural structures and is enabled by the existence of matching algorithms like the Iterative Closest Point (ICP) method and its many variants. In its most basic form, the Data Processing 12 reference point cloud is modelled by triangulation. The other point cloud is transformed to best fit the surface, such that the sum of distances between its points and the surface model is minimised. These algorithms perform best when there is a large overlap and the surface is rough and not smooth. The problem with planar surfaces is that the surface is very homogeneous in all its points and this can cause errors during matching. Data acquisition is a very fast and easy procedure for present-day systems. The main reasons are that data capture is automated and modern TLS allow a wide Field of view that permits modelling the whole scene from only a few locations. 2.1.2. Data treatment Once the raw data have been collected they must be treated. Before treating data the quality of the measurement must be looked at. The manufacturers give detailed technical information on their systems, but if a certain quality of measurement is needed, the precision of the 3D-coordinates must be looked at. The judgment of the quality of the measurements cannot be realized looking at single points (Point based Technique). With TLS the quality measurement has to be realized using groups of points (Area based Technique). A Point based technique cannot be used by TLS because Laser Scanners do not have a telescope for collimation, then it is difficult to measure the exact same point in each observation. The point will be calculated in each observation inside the area of the laser beam diameter. So, the smaller the laser beam diameter, the higher the precision. As mentioned before, an area based technique must be used to solve the problem. This technique extracts groups of points from the laser scanned data. The scanned point clouds can contain perturbations. These perturbations are caused by wrong points or data voids which can be misinterpreted while examining the different point clouds. These “wrong points” can be caused by: • Reflexions from temporary obstructions like vehicles, animals or people passing the scene during the scan. • Total reflexions from the object that can cause either data voids (no information) or virtual objects. Data voids should be minimised during data acquisition through the selection of appropriate scanning positions and minimising temporary obstructions to the scanner during operation. Data voids due to the occlusion to the line of sight can normally only be eliminated by using multiple scans. To help eliminate wrong points due to total reflexions a combination of scanned data with digital images is recommended. As mentioned before, the information of colour and texture from digital photographs helps to separate good data from noise or wrong points. Data Processing 13 The treatment of data is still a time consuming process. The current bottleneck in TLS data process is the extraction of desired information from the point cloud. This is due to the need of human work for it. 2.1.3. Data visualization One of the advantages of using point clouds is the visualization of the measured scenes or objects. This view can usually be generated from any direction and with any zoom rate. This visualization can be carried out in different ways: • Point clouds in a 3D-projection, with a colour or greyscale-coded intensity representation, are often used as a first visual check of the acquired data. • Point clouds can also be combined with derived geometrical elements. • Ortho-Photos are realized by the fusion of digital images and the registered point clouds. 2.2. Influence of the characteristics of the object (Reflectivity) Distance measurement precision is one of the most important features of TLS. It is predominated by both distance measurement systems (TOF, triangulation and phase measurement) and by technical details of the instrument in question. Each system has its own measurement accuracy. An influence of surface properties like colour, material and shape can be observed. The accuracy of distance measurements depends mainly on the intensity of the reflected laser light and therefore directly on the reflectivity of the object surface. The reflectivity depends on the angle of incidence and the surface properties. The influence of environmental conditions is commented in section 2.3. Study of colour and surface parameters made by [5] indicate that they are directly connected to the reflectivity factor. This study was carried out in a prepared indoor scenario, so it is rather a specific effect, we do not know exactly its impact in general. This reflectivity factor is presented in percentage values and is equal to the following relation of energy: Reflectivity factor= received energy from the target / emitted energy from the TLS * 100 The main rule is that the bigger the reflectivity factor, the smaller the Standard Deviation value of the point cloud. In [5] a selection of coloured sheets were scanned with Mensi GS100 in order to find out the reflectivity factor value depending on the colour. The results gave a scale that classifies the colours and the obtained reflectivity factor. A scale from 0 to 100, where 0 corresponds to the colour black and 100 to the colour white. The scale can be observed in image 2.1: Data Processing 14 Image 2.1 – Reflectivity factor scale As can be seen, dark colours have a lower reflectivity factor than light colours, so higher Standard Deviations will be registered with the former and lower distances with the latter. The effect can be visualized in image 2.2: Image 2.2 – Example of reflectivity factor [5]. Image 2.2 represents the front and side views of the point cloud obtained from measuring a traffic sign. The side view must be visualized from the right hand side because it is where the TLS was located while measuring. On this target we can only visualize three colours, black, red and white. As was mentioned before, black has the lowest reflectivity factor value, so it should be the colour with the highest Standard Deviation. In the image we can observe that black points are the furthest away from the TLS whereas white points cannot be visualized because white is a very reflective surface. Highly reflective objects may saturate some laser detectors, while the return signal from low-reflectivity objects may occasionally be too weak to register as valid. Regarding the texture of the target, matt and rough surfaces have got a lower reflectivity factor than shiny and smooth ones. Another parameter that should be considered is the angle of incidence. The angle of incidence is directly connected to the shape of the target. In general, the bigger the angle of incidence, the smaller the returning energy to the instrument. This is why it is important to choose the best possible position of the TLS in relation to the target. The best position is to have the surface of the target perpendicular to the TLS beam, because then there is a good reflectivity and a small incidence angle is obtained. The general rule is that the more parallel the surface of the target from the TLS the bigger the angle of incidence. This is also why rough surfaces have got a lower reflectivity factor than smooth Data Processing 15 ones. This is due to the heterogeneous nature of rough surfaces which will obviously have less points positioned perpendicularly to the TLS than the smooth homogeneous surfaces. In areas where the surface has a round shape serious errors can occur. This effect is demonstrated in image 2.3 where a sphere was scanned. Image 2.3 –Photograph of a sphere (left). Point cloud of a sphere as front view (middle) and as a view from above (right) [5]. Image 2.3 represents the front view and view from above of the point cloud obtained from measuring a sphere. As can be observed, there is a higher dispersion of the points the nearer it gets to the outline. This is due to the incidence angle, because the nearer to the outline of the round object, the bigger the incidence angle, causing a higher dispersion of the data. The measurement accuracy or standard deviation can be improved by increasing the number of distance measurement shots. By increasing the shots redundancy is obtained. However, multiple measurements can also cause negative effects. For rough and for dark surfaces a multiple measurement leads to a reduction of the dispersion. In contrast, for smooth surfaces and for a 0º angle of incidence, multiple measurements frequently lead to a wrong result of measurement and to a high dispersion of the points. 2.3. Influence of environmental conditions The effect of environmental conditions [13] is a possible effect that must be looked into because most surveys carried out by TLS are realized in outdoor environments. Data Processing 16 On the one hand, TLSs are not affected by the following environmental conditions: • Illumination: TLSs use an “active illumination” technique that, unlike photography, does not depend on ambient illumination, so TLSs can work during the day or at night. • Temperature or temperature variation. • Background noise. On the other hand, there are some conditions that affect TLSsmeasurements: • Sunlight and reflections: a strong sunlight reflection off a highly reflective target may saturate a receiver, producing an invalid or less accurate reading. However, laser measurements are not usually affected by other reflections • Dust and vapour: laser measurements can also be weakened by interacting with dust and vapour particles, which scatter the laser beam and the signal returning from the target. • Radiation: TLS devices determine baseline radiation levels to ensure that it does not interfere with the measurements. 2.4. TLS applications TLS is a young and recent instrument; therefore the number of applications is still rising. TLSs can be used for the following applications [13]: • Architecture and cultural heritage Structural surveying is an important step in urban design, from construction through to historical preservation. Using traditional methods, structural surveying is time-intensive and costly undertaking. Laser systems provide the means to reduce the time and cost significantly. Since the introduction of laser scanning technology there has been a misconception that the technology is not suitable for structural surveys due to accuracy limitations. But really, the strength of a laser scanner is not in its single shot precision but in its ability to sample the target area. Image 2.4 – Point cloud of a building. Data Processing 17 • Mine Planning and Tunnel surveying Data collection for current methods of open mine volume calculations and tunnel surveying are long and often dangerous. Single point collection with a total station and pole man along various slopes and faces is labour intensive, costly, and most important, dangerous. Therefore, measurements can be collected much more effectively and with no risk using Terrestrial Laser Scanner technology. Image 2.5 - Point cloud of a tunnel section • Topographical surveys In the past, topographical surveys were dominated by total stations, theodolites and, more recently, GPS systems. However, these traditional methods record a limited amount of measured points compared to Terrestrial Laser Scanner technologies. Additionally, these methods are labour-intensive, costly and inconvenient. In many cases, measurements with other devices rather than TLS have a lack of key information or vital points that make it more difficult to obtain good measurement results. Image 2.6 – Point cloud of downstream gate structures. Data Processing 18 • Object monitoring Modern surveying instruments, such as Total Stations and GPS receivers, have made this task easier and less time-consuming than traditional methods, but still do not offer the required point spacing to deliver a comprehensive surface analysis. Three dimensional laser imaging is the latest technology to be used in this essential monitoring process. Instead of using several dozen “target” points to monitor movement, users can only rely on a “point cloud” consisting of hundreds of thousands or even millions of points to represent the structure being measured. • Forensic Survey: Accident Reconstruction In the wake of an accident, experts are employed to aid in the reconstruction of the events that led to it. These experts provide vital functions, ranging from the identification of contributing factors to present in a court of law. Accurate documentation of the "crash environment" is vital for these experts. To determine the most probable scenario leading to the accident, physical elements such as skid marks, road settings ,etcetera.. are analysed. Weather and lighting conditions (which can contribute to an accident) also need to be included in the assessment, but these conditions can hinder traditional data collection methods. Terrestrial Laser scanning technology is currently being deployed by police organizations and surveyors for a variety of applications, including accident reconstruction. Since an accident can happen at any time, rain, snow and darkness, traditional data collection methods such as photogrammetry can be limited. TLS devices, that are rated to operate in inclement weather as well as independent of ambient lighting, are ideally suited for quick, accurate and efficient on-site data collection in virtually any circumstance. Image 2.7 – Point cloud of the scene of an accident • Land deformation monitoring Data Processing 19 This document talks about Land deformation monitoring. Land deformation monitoring is a type of object monitoring. The purpose of Land deformation monitoring is the detection of physical changes in a surface land or target in relation to the environment that surrounds it. To visualize these changes it is necessary to measure different scans of the target after a time and then compare the resulting scans. Land deformation monitoring is a very recent application for TLS technology; this is why there are still many dark points in this field. Data Analysis 20 CHAPTER 3 : DATA ANALYSIS After analysing the main functioning and types of TLS devices we will start the Experimental part of this document. The objective of this chapter is to explore the capabilites of TLS instruments for land deformation monitoring applications. 3.1. Available experimental data To carry out this analysis, we used data obtained from three different environments: • Castellfollit village • The experiment at the Campus • Formigal village 3.1.1. Castellfollit village The data of the experiment carried out in Castellfollit village was given to the Institute of Geomatics by “The research group of Natural Hazards of the University of Barcelona”. The given data was used to gain experience and study the TLS behaviour on different types of surfaces. Castellfollit is a village located in the province of Girona. The village, image 3.1, is situated at the top of a cliff. The data obtained from this environment is very useful due to the wide variety of surfaces, colours and shapes present. Since the aim of the experiment is to test the accuracy of the TLS instrument, these characteristics are interesting when comparing the different acquisitions obtained. The TLS device used to capture the data from this scenario was an Optech ILRIS-3D. Image 3.1 – Photograph of Castellfollit village Data Analysis 27 The residuals obtained from the matching can be seen in the following image: Image 3.7 – Residuals of two matched scans from Castellfollit. In this case, before matching these point clouds, the main sources of noise were eliminated previously. The residuals are around the order of a millimetre and with a standard deviation in the order of a centimetre which is more or less the expected precision of the instrument. These scans were taken at approximately 400m distance. . The second example of matching analysis was carried out over the Formigal test site, the two matched scans were two scans taken from approximately the same view point, about 550 metres distance, but with a difference of three months in time. In the following images both scans can be visualized: Image 3.8 – Photograph Formigal taken in July 2006 Data Analysis 28 Image 3.9 – Photograph Formigal taken in October 2006 We can observe that there has been a visible change in these three months, mainly because grass has grown all along the front part. The expected values cannot be previously calculated because of the mentioned differences. Table 3.3. Estimated results Formigal Value Deviation Scale 1.000000 0.000000 Omega(gon) 398.585 0.00008 Phi(gon) 1.90 0.00005 Kappa(gon) 399.472 0.00007 XO (m) -1.960 0.0004 YO (m) 1.945 0.0001 ZO (m) -0.089 0.0005 The following image shows the residuals, without noise, after matching both Scans from the Formigal scenario: Image 3.10 – Residuals Formigal Data Analysis 29 The results of this matching are worse than the previous ones, this is mainly due to the characteristics of the area which is not optimal for the angle of incidence viewpoint, for the type of terrain and for the weather conditions during the acquisitions. As was pointed out before, a difference on the land can be seen, turning out to be noise while matching both scans. Finally, we carried out the exact same test with the IG experiment. In this case, we only used a subsets of the wall with a part of a stair, avoiding this way any possible source of noise,such as vegetation. Image 3.11 –Photograph IG Scenario Image 3.12 – Residuals IG We can observe the residuals are much better than on the previous scenarios. This is due to the higher resolution and the optimum incidence angle from the surface to the device, in spite of the fact that smooth surfaces are not normally the best kind of surfaces while matching. Studying the results we detected a strange value on the surface of the wall, marked with a black square. We do not know what it is due to, because the wall is homogenous all along the surface, as can be seen in the photograph. This strange value has probably been caused by sistematic errors due to a bad instrument calibration. 3.3.3. Precision analysis In this section, one example of precision analysis is presented. Two repeated scans were matched using the LS3D method in order to evaluate the precision of the measurements from the residuals. The scans to be matched were two scans from Castellfollit which were taken from the exact same position and with a temporal baseline of minutes. Because they were taken from the exact same position there is a total overlap, i.e. the Data Analysis 30 two scans coincide. The two coinciding images are illustrated in the same image under these lines: Image 3.13 – Image of Castellfollit Scans The expected transformation parameters should be close to 0, because of the scans being taken on the same day and from the exact same position. The estimated results and their standard deviation obtained from the registration are shown in the following table: Table 3.4. Estimated results Castellfollit repeated Valor Desviació Scale 1.0000 0.00000 Omega(gon) 0.0068 0.00011 Phi(gon) -399.9955 0.00015 Kappa(gon) 0.00227 0.00008 XO (m) 0.0062 0.00014 YO (m) -0.0079 0.00005 ZO (m) -0.0048 0.00018 Nºiteracions 4 The residuals obtained from matching both scans have got the appearance illustrated in the following image: Data Analysis 31 Image 3.14 – Residuals scan 1 and 2 The matching of these two Scans can be used to verify the precision of the data registered from the TLS. The reason for this is that the scans were taken from the exact same position, without moving the TLS, so the results should be really close to 0. As expected, the obtained results are approximately 0 with a difference between both scans in the order of a millimetre and a standard deviation in the order of a centimetre. Altough the scans were taken from the exact same position, there can always occur a movement of the vegetation caused by the breeze or the internal instrumentation error of the device. These scans were taken at approximately 400m distance. The precision is consistent with the used device’s range accuracy (7 mm at 100 m). We also carried out several analyses to visualize the behaviour of the residuals over different areas from the scans above: • Houses • Rocks with vegetation and without vegetation The first area to be studied is the tower of the church, as can be seen in the following image: Image 3.15 – Photograph of the tower from Castellfollit Data Analysis 32 The image shows the characteristics of the target. It consists of a solid tower with two visible walls at different incidence angles, which make the two chosen regions (A,B). The areas with windows and flags were not considered. For both regions we calculated the statistics which are shown in the following table: Table 3.5. Statistics for regions A and B Region Min (m) Max (m) Mean (m) Stdev (m) A -0.04604 0.09108 0.00907 0.01094 B -0.14362 0.04911 -0.00357 0.01279 The obtained results can be considered as good, they are in the order of a millimetre. Section A and B have got the exact same characteristics, but Section B obtained better results because it has got a smaller incidence angle than section A. The second area consists of several buildings made of brick with nonparallel walls but with a more or less constant incidence angle and reflectance. In this area several small windows appear. The area can be visualized in image 3.16. Image 3.16 – Photograph of the group of building. The statistics of the area are shown in table 3.6. Table 3.6. Statistics for building region. These results are very similar to the obtained for the previous regions, even though this section is further away from the Lidar, so the results should be a little worse. The results are probably as good because in this section we included windows which makes it easier while matching than with smooth surfaces with no structures, like the sections above. Min (m) Max (m) Mean (m) Stdev (m) -0.13900 0.08300 -0.00390 0.01200 Data Analysis 33 The next sections to be studied are some parts of the rocks which are shown in the following images: Image 3.17 – Photographs of the rock areas. On the one hand, A and B are the areas of the rocks closest to the Laser Scanner device. In region A there is a lot of vegetation and it has got quite a big incidence angle, whereas region B has little vegetation and the incidence angle varies within the area. On the other hand, the areas C and D are the furthest away from the Laser Scanner Device. Both areas have no vegetation, area C hasquite a constant incidence angle and D has got a big distance variation. The statistics for the four regions can be seen in the following table: Table 3.7. Statistics for regions A and B Region Min (m) Max (m) Mean (m) Stdev (m) A -0.13037 0.1794 0.00175 0.0241 B -0.16255 0.1717 0.00057 0.01773 C -0.09023 0.14234 0.00066 0.01683 D 0.044012 0.131496 0.005007 0.014913 To finish, the results for these last sections are better than for the sections of the houses because they have got a smaller incidence angle. We can observe that the rocks with no vegetation, B and C, obtained much better results than the section with vegetation, A. This is consistent, because vegetation always causes noise, resulting in worse results. 3.4. Residuals vs Photograph The objective of this section is to find out some conclusions comparing the original photograph and the obtained residuals from the registration. We carried out several analyses to visualize the behaviour of the residuals over different areas (houses, rocks,etc.). To carry out the test the matched scans from the sections above were used. Data Analysis 34 Image 3.18 – Photograph Castellfollit scan Image 3.19 – Residuals from matching Castellfollit Scans C E B D A Data Analysis 35 Image 3.20 – Photograph Scans 3-4 Image 3.21 - Residuals Scans 3-4 In the images above a comparison between the obtained residual of matching scans and their corresponding photograph can be seen. We have chosen five different areas, which are: A: Village area B: Areas with vegetation C: Areas with trees D: Surfaces of the village with same orientation E: Area with water The conclusions obtained from this test are that depending on the type of surface or shape that is being calculated a big difference in the results of the residuals can be found. The part of the field scan where the village is reflected (A) is the surface which obtained the best residuals whereas surfaces composed of trees or vegetation (B) gave the worst residuals. The reason for this is that the houses are made of a solid non-variable surface, surface which is not affected by external conditions such as wind, so it is logical that when matching, the readings coincide almost exactly. In contrast, when dealing with surfaces with a degree of movement, like vegetation, it is very difficult for the C B A Data Analysis 36 TLS to get the same reading twice. In conclusion, readings of trees and vegetation result in noise, so previous to matching these areas should be discarded. In the scans, a strong correlation can be visualized on the surfaces marked D. This correlation, which is studied in the following section, could be due to the orientation of the targets with respect to the TLS. Finally, E seems to be the reflexion of the rocks in the water, as if the beam had landed on the water and the reflexed beam had bounced off the rocks and returned to the TLS. 3.5. Incidence angle vs Residual The goal of this test is to study the relation between the incidence angle of the targets and the precision by using the residuals. To find out this relation, we calculate the incidence angle for every point in the point cloud and use the residuals obtained when matching the images from the TLS in the LS3D matching method explained before. The incidence angle is calculated for every point in the resulting point cloud from the TLS instrument. The procedure used to calculate the incidence angle is documented below and the used programme can be seen in Annex [1] . The reference of the coordinate system is the TLS geometric rotation centre, which is considered as the zero point (0,0,0). To calculate the angle of incidence we need two vectors, as is shown in image 3.22: Image 3.22 – Needed vectors to calculate the incidence angle of point P. Where P is the point to calculate, O the center of the TLS and n the normal vector. The PO vector is calculated directly from two known points, the zero point and P. PO = P – O = (X,Y,Z) – (0,0,0) = (X-0,Y-0,Z-0) = (X,Y,Z) (3.2) Data Analysis 43 Image 3.35 - Incidence Angle Image 3.36 - Standard Deviation In this case, to make the comparison between the Standard Deviation and the Incidence angle we made a graphic for different Incidence Angle ranges. We grouped the Incidence Angle in intervals of 10, then we calculated the statistics of the Standard Deviation of the points with the mentioned incidence Angle: : Inc Angle vs St. Deviation (Formigal) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 10 20 30 40 50 60 70 80 90 Incidence Angle (º) St. Deviation (m) Image 3.37 – Inc Angle vs Standard deviation. X corresponds to Incidence Angle intervals. Y corresponds to the Standard Deviation. In both cases we can observe that the bigger the angle of incidence, the higher the Standard deviation of the calculated data from the TLS. The same reason as for the above case causes this, the bigger the incidence angle, the less energy the TLS receives from the target, so the higher the possibility of error. We can visualize that on the outlines of the objects standard deviations are very high compared to the rest of the object. Data Analysis 44 3.6.1. Use of Standard Deviation value In this section we will try and find out the possibility of using the Standard deviation to locate same areas in two different scans taken at a different baseline of time or from a different point of view. We think that if the scans are taken at different baselines of time but more or less from the same position, we will probably be able to detect and locate the different areas from the Standard deviation. Whereas we are not as sure that it will be possible if the scans are taken from different view points, due to the possibility of the targets changing the shape, so the Standard deviation can also change. If it is possible to locate this areas, they will be very useful as initial values for co-registration. First we will study this with the data obtained from Formigal, because we have got two scans taken at different periods of time, one in July and the other in October. In the following image, both scans can be visualized: Image 3.38 – Standard Deviation Formigal in July 2006 Image 3.39 - Standard Deviation Formigal in October 2006 As can be observed, both images were taken from a very similar point of view but with three months difference. This is a short period of time, but as was Data Analysis 45 mentioned in previous sections, there was a big change on the area. We have chosen two different subsets of data which we think could make it possible to locate different targets in the scans. The first subset corresponds to the lowest standard deviation values, from 0 to 5 cms. Here we show the resulting images of both scans: Image 3.40 – 0-5cms Standard Deviation in July 2006 Image 3.41 – 0-5cms Standard Deviation in October 2006 In the images the similar located targets in both scans have been marked. They correspond to some rocks, a place where there was like a hole and the fence from the road. With these located targets it could be possible to find several points as reference points or areas to start with the matching. The second subset corresponds to the highest standard deviation values. This values can gives us both, the outlines or noise data or some other reference targets. In the following images the scans can be visualized: Image 42– max Standard Deviation Formigal in July 2006 Data Analysis 46 Image 3.43 – max Standard Deviation Formigal in October 2006 We can visualize in both scans all the poals from a fence. In this case there are no wrong points because there a are no trees or bushes that would result as noise. This is why we will carry out this last experiment with Castellfollit, which has got a lot of trees ans vegetation: Image 3.44 - Castellfollit Scan1 In image 3.44, we can observe that there are very high values that correspond to bad points. These bad points will be due to noise. To eliminate this noise, we will look for a high range of Standard deviation, because these bad points should have quite a different value compared to the wright ones. In the following image the values correspond to the ones from 1m to the maximum (275m): Data Analysis 47 Image 3.45 - Noise in Castellfollit We have been able to eliminate most of the bad points by simply eliminating the highest values of Standard Deviation. In this case it is possible to do this because there is not a very big distance between the targets in the image. In the purple square there are some good points, the outline and windows of the houses that shouldn’t have been eliminated..To solve this we just need to select the area and not apply the mask. The data in the red square has been eliminated directly, because looking at the digital photograph we can observe it corresponds to the forest behind the mountain, so it is data that will only provoque bad results In the following image we can visualize the resulting scan without vegetation: Image 3.46 – Castellfollit Scan without noise Data Analysis 48 We can observe in the image, that the smallest vegetation, like little bushes in between the rocks haven’t been eliminated. In some occasions, it would be better to eliminate this data, but because it is part of the rocks and the scenario it is not necessary. After using scans taken from the same view point, we will try and locate similar parts in two scans taken from different view points. The scans we are going to use for this section are the following: Image 3.47 - Scan4 Image 3.48 - Scan5 In the folliwing images the scans can be visualized after applying the Standard Deviation: Image 3.49 – Standard Deviation Image 3.50 – Standard Deviation Data Analysis 49 In this case we will create a mask that will include the range from 0 to 2 m. Here we show the resulting images of both scans: Image 3.51 - Scan4 Image 3.52 - Scan5 The similar targets in both scans have been marked with different squares. In this case they correspond to some houses in the village and a little cottage in the fields. These points could be used as reference points to start the matching. In conclusion, the standard deviation can be used to find initial reference points for both cases, scans taken from different viewpoints, and scans taken from the same viewpoint but a different baseline in time. It can also be used quite effectively to eliminate noise caused by trees or vegetation. To conclude, this is not a good method to be able to distinguish objects automatically, because manual interaction has to be carried out. 3.7. Study of the Influence of the characteristics of the object As mentioned before, the goal of this test is to study the quality of the measurement realized with the TLS. To study this quality, we will also focus on the dispersion of the point clouds depending on the properties of the material. For this test we will use some subsets from the obtained data in the experiment at the institute. In the following image a photograph of the ten artificial used pannels can be seen: Data Analysis 50 Image 3.53 – Different Pannel surfaces from the IG experiment The number of pannel in the image goes from top left starting in 1. First of all we will verify that the standard deviation corresponds with the shape of the pannel, not taking into account the material in this case, just the shape. The theory tells us that the rougher the surface, the higher the standard deviation should be. Here we can find a graph comparing the Standard deviation for the ten pannels: Standard Deviation (Pannels) 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 1 2 3 4 5 6 7 8 9 10 Pannels St.Deviation (m) Image 3.54 – Standard Deviation of the distance measured by the TLS Pannels Data Analysis 51 In the graph we can observe that Pannel 9 has got the highest standar deviation, this is consitent because is has got the rougher surface, so the difference of values between the points in the point cloud are higher. Whereas for pannel number 1 we obtain the lowest value, due to the smooth surface. For the other pannels de values are very close because they have all got similar shapes. Now we will go on to study the effect of the different materials of the pannels in the TLS measurements. To do this we will use subsets of the different pannels where the surface is smooth, not selecting the different created shapes. An example is shown with pannel 3, the parts marked in red are the ones that have been chosen: Image 3.55 – Smooth subsets chosen for pannel 3. We need to do it this way because to compare different materials it is necessary to have a more or less similar surface (shape) for all the pannels. In the following graph the results can be observed: Standard Deviation (Pannels) 0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 1 2 3 4 5 6 7 8 9 10 Pannels St.Deviation(m) Data Analysis 52 Image 3.56 – Standard Deviation of the distance measured by the TLS of smooth subsets of Pannels We can see that there are hardly any differences between the different Standard Deviation of the ten pannels, the differences are of a few millimeters. However, the three pannels with the shaving material are the ones with highest values. This is due to the rough and irregular surface, then, it is normal to obtain these values. Opposite, we can visualize that pannel 8 has got the lowest value, this is because it corresponds to the smoothest surface, so the standard deviation is the lowest. Annex 59 matx[auxf][auxc]=x; maty[auxf][auxc]=y; matz[auxf][auxc]=z; auxc++; } auxf++; } auxf=0; while(auxf<nlin) { auxc=0; while(auxc<ncol) { kk=0; //Create PO vector po[0]=matx[auxf][auxc]; po[1]=maty[auxf][auxc]; po[2]=matz[auxf][auxc]; //If bad points (0,0,0) we force 0 angle if((po[0]==0.000)&&(po[1]==0.000)&&(po[2]==0.000)) { angle=0; fprintf(pout,"%lf\n",angle); auxc++; kk=1; } //For the outline points we don’t calculate vector, we force 0 angle if ((auxf==0)||(auxf==nlin-1)) { for(auxc=0;auxc<ncol;auxc++) { angle=0; fprintf(pout,"%lf\n",angle); } kk=1; } if((auxc==0)||(auxc==ncol-1)) { angle=0; fprintf(pout,"%lf\n",angle); Annex 60 auxc++; kk=1; } if(kk==0) { //Create vector 1 ok1=0; if(nvent>(ncol-1)-auxc) { v1[0]=0; v1[1]=0; v1[2]=0; for(auxv=1;auxv<nvent;auxv++) { if(matx[auxf][auxc-auxv]!=0.0000) { v11[0]=matx[auxf][auxc]- matx[auxf][auxc-auxv]; v11[1]=maty[auxf][auxc]- maty[auxf][auxc-auxv]; v11[2]=matz[auxf][auxc]- matz[auxf][auxc-auxv]; v1[0]=(v1[0]+v11[0]); v1[1]=(v1[1]+v11[1]); v1[2]=(v1[2]+v11[2]); ok1++; } } } else { v1[0]=0; v1[1]=0; v1[2]=0; for(auxv=1;auxv<nvent;auxv++) { if(matx[auxf][auxc+auxv]!=0.0000) { v11[0]=matx[auxf][auxc]- matx[auxf][auxc+auxv]; v11[1]=maty[auxf][auxc]- maty[auxf][auxc+auxv]; v11[2]=matz[auxf][auxc]- matz[auxf][auxc+auxv]; Annex 61 v1[0]=(v1[0]+v11[0]); v1[1]=(v1[1]+v11[1]); v1[2]=(v1[2]+v11[2]); ok1++; } } } //Create vector 2 ok2=0; if(nvent>(nlin-3)-auxf) { v2[0]=0; v2[1]=0; v2[2]=0; for(auxv=1;auxv<=nvent;auxv++) { if(matx[auxf-auxv][auxc-auxv]!=0.0000) { v21[0]=matx[auxf][auxc]- matx[auxf-auxv][auxc-auxv]; v21[1]=maty[auxf][auxc]- maty[auxf-auxv][auxc-auxv]; v21[2]=matz[auxf][auxc]- matz[auxf-auxv][auxc-auxv]; v2[0]=(v2[0]+v21[0]); v2[1]=(v2[1]+v21[1]); v2[2]=(v2[2]+v21[2]); ok2++; } } } else { v2[0]=0; v2[1]=0; v2[2]=0; for(auxv=1;auxv<=nvent;auxv++) { if(matx[auxf+auxv][auxc+auxv]!=0.0000) { v21[0]=matx[auxf][auxc]- matx[auxf+auxv][auxc+auxv]; Annex 62 v21[1]=maty[auxf][auxc]- maty[auxf+auxv][auxc+auxv]; v21[2]=matz[auxf][auxc]- matz[auxf+auxv][auxc+auxv]; v2[0]=(v2[0]+v21[0]); v2[1]=(v2[1]+v21[1]); v2[2]=(v2[2]+v21[2]); } } } if((ok1=!0)&&(ok2=!0)) { v1[0]=v1[0]/ok1; v1[1]=v1[1]/ok1; v1[2]=v1[2]/ok1; v2[0]=v2[0]/ok2; v2[1]=v2[1]/ok2; v2[2]=v2[2]/ok2; // “Producto vectorial” between vector 1 and vector 2 to calculate normal vector norm[0]=v1[1]*v2[2]-v1[2]*v2[1]; norm[1]=v1[2]*v2[0]-v1[0]*v2[2]; norm[2]=v1[0]*v2[1]-v1[1]*v2[0]; // “Producto escalar” between PO and Normal vector, the incidente angle. a=po[0]*norm[0]+po[1]*norm[1]+po[2]*norm[2]; b=sqrt((po[0]*po[0])+(po[1]*po[1])+(po[2]*po[2])); c=sqrt((norm[0]*norm[0])+(norm[1]*norm[1])+(norm[2]*norm[2])); cos=a/(b*c); angle=acos(cos); angle=angle*(180/3.1416); if(angle>90) angle=(180-angle); fprintf(pout,"%lf\n",angle); auxc++; } Annex 63 else { angle=0; fprintf(pout,"%lf\n",angle); auxc++; } } } auxf++; } printf(" Programma terminato!\n"); fclose(pdam); fclose(pout); return; [2] Programme to calculate th} Standard Deviation: # include <stdio.h> # include <stdlib.h> # include <math.h> # include <cmath> # include <malloc.h> void main(void) { FILE *pdam,*pout; int aperto=0, aperto1=0,auxf,auxc,ncol,nvent,ntot,nlin,i=0,j,f=0,kk,ja=0; char inp_name[50], inp_name1[50]; double x,y,z,desv,a=1,b=0,med,suma,r=0; double **mat,*vect,**mat2; //Open input file while (aperto==0) { printf(" Inserte el nombre del fichero de entrada!\n"); scanf("%s", inp_name); pdam=fopen(inp_name,"rb"); if (pdam==NULL) { printf(" \n\tFichero no abierto pruebe de nuevo!\n"); } else { aperto=1; } } //Open output file while (aperto1==0) { Annex 64 printf(" Inserte el nombre del fichero de salida!\n"); scanf("%s", inp_name1); pout=fopen(inp_name1,"w"); if (pout==NULL) { printf(" \n\tFichero no abierto pruebe de nuevo!\n"); } else { aperto1=1; } } //Introduce number of lines and columns of the input file printf(" Inserire il numero di linee!\n"); scanf("%d", &nlin); printf(" Inserire il numero di colonne!\n"); scanf("%d", &ncol); printf(" Inserire de la ventana!\n"); scanf("%d", &nvent); mat=(double **)malloc(nlin*sizeof(double *)); for(j=0;j<nlin;j++) { mat[j]=(double *)malloc(ncol*sizeof(double)); } mat2=(double **)malloc(nlin*sizeof(double *)); for(j=0;j<nlin;j++) { mat2[j]=(double *)malloc(ncol*sizeof(double)); } vect=(double*)malloc((nlin*ncol)*sizeof(double)); auxf=0; while(auxf<nlin) { auxc=0; while (auxc<ncol) { fscanf(pdam,"%lf %lf %lf %d\n",&x,&y,&z,&i); r=sqrt((x*x)+(y*y)+(z*z)); mat[auxf][auxc]=r; auxc++; } auxf++; } Annex 65 auxf=0; while(auxf<nlin) { auxc=0; while (auxc<ncol) { mat2[auxf][auxc]=0; auxc++; } auxf++; } auxf=nvent; while(auxf<nlin-nvent) { auxc=nvent; while(auxc<ncol-nvent) { if(mat[auxf][auxc]!=0.000) { ntot=nvent*nvent; med=0; ja=0; for(i=auxf-(nvent/2);i<nvent+(auxf-(nvent/2));i++) { for(j=auxc-(nvent/2);j<nvent+(auxc- (nvent/2));j++) { if(mat[i][j]!=0.000) { suma=mat[i][j]; med=med+suma; ja++; } } } med=med/ja; b=0; for(i=auxf-(nvent/2);i<nvent+(auxf-(nvent/2));i++) { for(j=auxc-(nvent/2);j<nvent+(auxc- (nvent/2));j++) { if(mat[i][j]!=0.000) { a=mat[i][j]-med; a=a*a; Annex 66 b=b+a; } } } desv=sqrt(b/ja); mat2[auxf][auxc]=desv; } auxc++; } auxf++; } auxf=0; while(auxf<nlin) { auxc=0; while (auxc<ncol) { fprintf(pout,"%lf\n",mat2[auxf][auxc]); auxc++; } auxf++; } /*auxf=0; while(auxf<nlin) { auxc=0; while (auxc<ncol) { fprintf(pout,"%lf\n",mat[auxf][auxc]); auxc++; } auxf++; }*/ printf(" Programma terminato!\n"); fclose(pdam); fclose(pout); return; } Annex 67