Investigations Into the Accuracy of the Uav System Dji Matrice 300 Rtk with the Sensors Zenmuse p1 and l1 in the Hamburg Test Field

INVESTIGATIONS INTO THE ACCURACY OF THE UAV SYSTEM DJI MATRICE 300 RTK WITH THE SENSORS ZENMUSE P1 AND L1 IN THE HAMBURG TEST FIELD T. Ke s en 1 *, J. Wol 1, M. Linds aed 1 1 Ha enCi y Uni e si y Hambu g, Pho og amme y & Lase Scanning Lab, Henning-Vosche au-Pla z 1, 20457 Hambu g, Ge many - (Thomas.Ke s en, Joshua.Wol , Ma en.Linds aed )@hcu-hambu g.de Commission I, WG 10 KEY WORDS: accu acy, bundle block adjus men , g ound con ol poin s, e e ence da a, RTK-GNSS, es ield, UAV/UAS. ABSTRACT: The de elopmen o inc easingly powe ul Unmanned Ae ial Vehicles (UAV) is p og essing con inuously, so ha hese sys ems equipped wi h high- esolu ion senso s can be used o a a ie y o di e en applica ions. Wi h he Ma ice 300 RTK, Da-Jiang Inno a ions Science and Technology Co. L d (DJI) has launched a sys em ha can use he high- esolu ion came a Zenmuse P1 o he lase scanne Zenmuse L1 as a eco ding senso , among o he senso s. In o de o in es iga e he geome ic quali y o hese wo senso s, Ha enCi y Uni e si y Hambu g, in coope a ion wi h LGV Hambu g, NLWKN in No den and he Ge man A chaeological Ins i u e in Bonn, lew o e he 3D es ield in he Inselpa k in Hambu g-Wilhelmsbu g on 5 Augus 2021 wi h he P1 came a and he L1 lase scanne . Using he Ma ice 300 RTK as ca ie pla o m, he es ield was eco ded in a ious con igu a ions a al i udes be ween 50 m and 90 m abo e g ound. P io o he UAV ligh campaign, 44 ma ked g ound con ol poin s (GCP) we e signalised in he es ield, which had al eady been su eyed by LGV in 2020 using geode ic measu emen me hods o achie e a coo dina e accu acy o ±5 mm o each GCP. The esul s o ae ial iangula ions as well as 3D poin clouds gene a ed om image da a and lase scanning a e compa ed wi h e e ence da a in o de o demons a e he accu acy po en ial o hese measu emen sys ems in his pape . * Co esponding au ho 1. INTRODUCTION Unmanned ae ial ehicles (UAVs) a e inc easingly used in a ious disciplines o lexible su eys o small o medium- sized su ey a eas. The use o UAV sys ems equipped wi h Real-Time Kinema ic (RTK) GNSS inc eases he a ac i eness o hese sys ems o many asks, as hey o e a posi ioning accu acy o 2-3 cm in he na ional coo dina e sys em wi h hese senso s (Ge ke and P zybilla, 2016; P zybilla e al., 2020; Ke s en and Linds aed , 2022). As a consequence, a signi ican educ ion o con ol poin s is possible, making he use o RTK- GNSS based pla o ms mo e lexible and e icien o many applica ions. In ecen yea s, UAV sys ems wi h RTK-GNSS ha e inc easingly es ablished hemsel es as wo kho ses o applica ions in UAV pho og amme y. Wi h he DJI Ma ice 300 RTK, a sys em is now a ailable ha has high posi ioning accu acy and can be equipped wi h a high- esolu ion came a o lase scanne , among o he senso s. This makes i possible o eco d a wide a ie y o objec s such as u ban scenes, coas al zones, ag icul u al a eas o o es a eas. Resul s on he geome ic quali y o ae ial iangula ions o di e en UAV based came a sys ems ha e al eady been published (P zybilla e al. 2019; Ke s en e al. 2020). Ge ke and P zybilla (2016) p esen ed i s esul s on he in luence o on- boa d RTK-GNSS and c oss- ligh s o a UAV sys em, while P zybilla e al. (2020) published i s esul s o RTK-based UAV pho og amme y using ou DJI Phan om 4 RTK sys ems lown in c oss- ligh s a di e en al i ude on he si e o he Zolle n collie y UAV es ield in Do mund. Fu he accu acy es s ha e been ca ied ou by Zhao e al. (2020) and Zhao (2021). In ecen yea s, unmanned ae ial sys ems wi h RTK- GNSS a e s a e-o - he-a in UAV pho og amme ic applica ions. In o de o in es iga e he geome ic accu acy po en ial o hese wo senso s P1 and L1 on-boa d he UAV sys em Ma ice 300 RTK, Ha enCi y Uni e si y Hambu g, in coope a ion wi h he S a e O ice o Geoin o ma ion and Su eying (LGV) Hambu g, he Lowe Saxony S a e O ice o Wa e Managemen , Coas al and Na u e Conse a ion (NLWKN) in No den, Ge many and he Ge man A chaeological Ins i u e (DAI) in Bonn, ca ied ou ae ial ligh s o e he 3D es ield in he Inselpa k o Hambu g-Wilhelmsbu g on Augus 5 h, 2021. The UAV ligh s we e conduc ed in a ious ligh con igu a ions and a ligh al i udes be ween 50 m and 90 m abo e g ound. Fo accu acy in es iga ions, he image o ien a ions and came a calib a ions o he di e en UAV image ligh s we e calcula ed by ae ial iangula ion using he so wa e Agiso Me ashape. The accu acies o ae ial iangula ion we e analysed using di e en g ound con ol and check poin con igu a ions. The accu acy po en ial o he lase scanne was analysed using geode ic check poin s and e e ence da a (p o iles and selec ed a eas) o a e es ial lase scanne . Addi ionally he lase poin clouds we e compa ed wi h image- based poin clouds o P1 and wi h o icial da a o ai bo ne lase scanning p o ided by LGV. The ollowing ques ions, among o he s, a e answe ed:  Wha accu acies (ae ial iangula ion and e ain models) a e achie ed by he UAV ligh s o he wo eco ding sys ems in hese in es iga ions? The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 339  Which ae ial ligh con igu a ions p o ide he bes esul s compa ed o e e ence?  Is i possible o educe he numbe o GCP wi h co esponding lowe accu acy equi emen s o p ojec s when using accu a e RTK-GNSS obse a ions o UAV ligh s? 2. THE UAV TEST FIELD IN WILHELMSBURG INSELPARK In he Inselpa k in Hambu g's Wilhelmsbu g dis ic , which hos ed he In e na ional Ga den Show in 2013, he LGV Hambu g se up a es ield o UAV sys ems consis ing o 45 g ound con ol poin s (GCP) on an a ea o 150 m × 300 m. The GCP coo dina es we e de e mined using a ious geode ic measu emen me hods and he heigh s we e addi ionally de e mined by le elling. The LGV speci ies a coo dina e accu acy o ± 5 mm o each GCP coo dina e. As can be seen in Figu e 1, he GCP a e e enly dis ibu ed o e his app oxima ely 4.5 ha a ea o he Inselpa k. P io o he su ey on Augus 5 h, 2021, 44 GCP we e signalised on g ass, asphal and sand using a ge boa ds made o wa e p oo plas ic wi h dimensions o 50 cm × 50 cm (Fig. 1, igh ). Figu e 1: G ound con ol poin dis ibu ion in he UAV es ield Inselpa k Hambu g-Wilhelmsbu g (le ) and a ge s on di e en su aces ( igh ) - g ass, asphal , sand and s one. 3. THE UAV SYSTEM USED The DJI Ma ice 300 RTK (Figu e 2) is a 6.3 kg quadcop e om he Chinese manu ac u e DJI Technology, which can be ope a ed a al i udes o up o 5000 m wi h a maximum ligh ime o 55 minu es. Equipped wi h he Au oma ic Dependen Su eillance - B oadcas (ADS-B) an i-collision sys em, he UAV can achie e a posi ioning accu acy o 1.0-1.5 cm + 1 ppm using RTK-GNSS. In con as o many compa able sys ems, he M300 RTK does no ha e a ixed senso , ins ead he pla o m can be equipped wi h a ious senso s such as he came a DJI Zenmuse P1 o he (ai bo ne) lase scanne DJI Zenmuse L1 o ae ial ligh s. The M300 RTK is powe ed by wo TB60 ba e ies. Fo longe missions, he ba e ies can be eplaced one a e he o he du ing ope a ion a e landing wi hou disconnec ing he senso sys em om he powe supply. 3.1 The DJI Zenmuse P1 Came a The DJI Zenmuse P1 came a (Figu e 3 le ) is o e ed by DJI o he Ma ice 300 RTK. This is a 45 megapixel (pixel size 4.4 μm) digi al came a equipped wi h a ull- ame (35.9 mm × 24 mm) CMOS senso ha can be ope a ed wi h a ious lenses o e ed wi h di e en ocal leng hs. In he con ex o hese in es iga ions, a lens wi h a ocal leng h o 35 mm was used, which has a ield o iew (FOV) o 63.5° and can ake pho os in an ape u e ange (F-S ops) o F2.8 o F16. Figu e 2: Top - DJI Ma ice 300 RTK wi h Zenmuse P1 came a (le ) and L1 lase scanne ( igh ), bo om - Zenmuse P1 came a (le ) and L1 lase scanne ( igh ). 3.2 The DJI Zenmuse L1 Ai bo ne Lase Scanne In addi ion o he P1 came a, he Ma ice 300 RTK can op ionally be used wi h he DJI Zenmuse L1 ai bo ne lase scanning senso (Figu e 3 igh ), which is he i s lase scanne om DJI. This scanne , which is equipped wi h a LiDAR module om he manu ac u e Li ox, has a ange o 450 m wi h a FOV o 70° (LIVOX 2022). In ligh planning, a choice can be made be ween single- e u n o mul iple- e u n mode. In addi ion, wo di e en scanning modes a e a ailable, which esul in di e en poin pa e ns o speci ic equi emen s o objec s o be scanned, and which enable scanning o up o 240,000 poin s pe second. Howe e , he L1 senso also manages up o h ee e u ns pe lase sho , so ha he poin a e can be up o 480,000 poin s pe second when scanning ege a ion, o example, wi h wo o h ee e u ns (Singh, 2020). These wo scanning modes a e e e ed o by DJI as epe i i e and non- epe i i e (Fig. 4). Acco ding o he manu ac u e , he L1 senso achie es a sys em accu acy o 10 cm in a i ude and 5 cm in al i ude a a lying heigh o 50 m abo e g ound. Un o una ely, i is no clea om he manu ac u e 's echnical speci ica ion whe he he sys em accu acy e e s o posi ioning o 3D poin de e mina ion. The p ecision o he dis ance measu emen (RMS 1σ) o he lase scanne is speci ied as 3 cm a a dis ance o 100 m (DJI 2022). Figu e 3: Non- epe i i e ci cula scanning (le ) and epe i i e line scanning ( igh ) wi h he L1 lase scanne (LIVOX 2022). The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 340 3.3 Ae ial ligh con igu a ions Fi s , he es ield was eco ded by wo image ligh s wi h he Ma ice 300 RTK/Zenmuse P1 sys em (Table 1). These wo ligh s ook place a an al i ude o 70 m and 90 m abo e g ound. Du ing he i s ligh , a combina ion o nadi and oblique images (backwa ds and sideways) was aken, while du ing he second ligh a he highe al i ude only nadi images we e aken. This esul ed in a G ound Sampling Dis ance (GSD) o 8.8 mm o he nadi images and 10.2 mm (image cen e) o he oblique images (oblique) a an angle o 60° o he i s ligh , while he second nadi ligh had a GSD o 11.3 mm. Fo bo h ligh s, he exposu e ime was se o 1/1000 sec, while he F-S op a ied be ween 4 and 7.1 and he ligh sensi i i y o he senso be ween ISO 400 and 640 o an op imal exposed image. Pa ame e Ae ial ligh 1 Ae ial ligh 2 Flying heigh 70 m 90 m Reco ding angle Nadi and Oblique Nadi GSD (cen e) 8.8 mm / 10.2 mm 11.3 mm O e lapping 80 % / 80 % 80 % / 80 % Flying ime 39 min 25 s 9 min 18 s Pho os 2215 408 Table 1. Ae ial ligh s wi h he Zenmuse P1 came a. Subsequen ly, h ee ligh s o e he es ield we e ca ied ou wi h he Zenmuse L1 lase scanne (Table 2). The di e en scanning modes we e compa ed and he in luence o inc easing he ligh al i ude om 50 m o 90 m was in es iga ed. The s ip o e lap was se o 60% o all ligh s. In addi ion, mul iple e u n echo mode was used on all ligh s o in es iga e he abili y o lase scanning o pene a e ege a ion. A e he s a o he UAV ligh , he lase scanne and he ine ial measu emen uni we e calib a ed in he ai by a eco ding p ocedu e implemen ed by he manu ac u e be o e he ac ual da a acquisi ion s a ed. Du ing he ligh and scanning ope a ion, he 3D poin cloud was al eady colou ed in eal ime by he RGB alues o he Zenmuse X4S came a (20 megapixels) in eg a ed in he lase scanne and displayed on he DJI En e p ise sma emo e con ol, which has an ul a-b igh 5.5-inch 1080p display o con olling he UAV sys em du ing ligh . Pa ame e Fligh 3 Fligh 4 Fligh 5 Flying heigh 50 m 90 m 50 m Poin densi y 399 p s/m² 209 p s/m² 445 p s/m² Scanning mode epe i i e epe i i e non- epe i i . Echo mode mul iple- e u n O e lapping 60 % 60 % 60 % Flying ime 13 min 4 s 8 min 19 s 13 min 4 s Table 2. Ae ial ligh s wi h he Zenmuse L1 lase scanne . 4. DATA EVALUATION AND RESULTS The eco ded ae ial image blocks we e e alua ed in he so wa e Agiso Me ashape V1.7 using he signalised 44 GCP. The ae ial iangula ions o bo h image ligh con igu a ions we e calcula ed wi h di e en GCP con igu a ions in o de o assess he quali y o he esul s based on di e en a ian s simila o (Ke s en e al., 2020). In Agiso Me ashape, he image poin measu emen s we e pe o med au oma ically and he GCP measu emen s semi-au oma ically. In he subsequen bundle block adjus men s, he so wa e calcula ed he image o ien a ion and came a calib a ion pa ame e s o each GCP e sion. In he nex s ep, 3D poin clouds we e gene a ed by dense image ma ching o he pho o blocks o UAV ligh s 1 and 2 using he o ien a ion pa ame e s o he e sion wi h all 44 GCP. The da a om he Zenmuse L1 lase scanne can (cu en ly) only be analysed wi h he DJI Te a so wa e. The impo ed poin clouds o he h ee ligh s we e each op imised by s ip adjus men and inally expo ed in LAS o ma in he UTM coo dina e sys em (EPSG 4647) and wi h ellipsoidal heigh s, jus like he poin clouds gene a ed in he pho os. The highes quali y le el was selec ed o he da a p ocessing. The quali y o he 3D poin clouds gene a ed om he acqui ed da a o he i e UAV ligh s was in es iga ed using 44 checks poin s (ChP) and by compa ing di e en p o iles and e e ence su aces acqui ed wi h a FARO Focus3D X330 e es ial lase scanne . The e e ence da a we e scanned a ound he building, which is isible in Figs. 8 and 10, in 29 scans ( esolu ion 1/5 and quali y 3x). When egis e ing he scans in he FARO® SCENE so wa e, an a e age poin e o o 4.2 mm was achie ed. Compa able geome ic accu acy in es iga ions o image-based 3D poin clouds ha e al eady been ca ied ou o a ious UAV sys ems in he es ield a he Zolle n collie y in Do mund (P zybilla e al., 2019). 4.1 Compa ison o he Resul s o he Ae ial T iangula ion Fo de ailed accu acy in es iga ions, di e en GCP e sions wi h di e en numbe s o spa ially well dis ibu ed g ound con ol poin s (all GCPs, 12, 5 and 1 GCP) we e calcula ed in bundle block adjus men s, whe eby all GCP no aken in o accoun we e hen used as check poin s. In all bundle adjus men s, he posi ioning coo dina es o he ex e io o ien a ion showed an RMSE (Roo Mean Squa e E o ) in he ange o 11 o 16 mm, while he de ia ions o he heigh coo dina es we e calcula ed a app ox. 11 mm. The a e aged s anda d de ia ions o he RTK-GNSS measu emen s o he image posi ions o bo h image ligh s we e 15 mm in a i ude and 29 mm in heigh . Howe e , he indi idual alues o he RTK-GNSS measu emen s pe image posi ion we e in oduced in o he bundle adjus men as a p io i s anda d de ia ion. The esul s o UAV image ligh 1 wi h nadi and oblique images (2215 pho os) a e summa ised in Figu e 5. The GCP ha e been measu ed on a e age in 155 pho os. The a p io i s anda d de ia ion o each con ol poin coo dina e was se o 5 mm in each adjus men e sion. In he bundle adjus men wi hou GCP o wi h a single con ol poin , he de ia ions a he 43 and 44 checks poin s a e o X = 15 mm and Y = 11 mm, whe eby he de ia ions a he heigh Z a e highe by a ac o o 2.8 wi h up o 42 mm ( igh wo columns in Fig. 5). E en in he adjus men wi h all GCP, he RMSE o he check poin s is 19 mm in he heigh coo dina e, while he XY coo dina es a e a a e age de ia ions o 10 mm and 5 mm espec i ely. The ewe GCP a e used in he adjus men , he signi ican ly highe he RMSE alues in he heigh coo dina e become. Due o he e y high edundancy caused by obse a ions in 2215 ae ial images, a signi ican ly be e esul was expec ed, which was hen achie ed wi h image ligh 2 (Fig. 6). Causes o he la ge heigh de ia ions in he GCP and ChP could be he geome y o he ligh con igu a ion, he na ow FOV o he lens as well as he eco ding p ocedu e wi h he pi o ing o he came a on he le e a m (gimbal) and he associa ed change in ocusing o oblique images compa ed o nadi images, which hus also in luences he came a calib a ion. DJI de ines he ec o o he le e a m om he GNSS an enna cen e o he p ojec ion The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 341 cen e o he came a, which should ha e only mino co ec ion e ec s on he esul . The ep ojec ion e o , a geome ic e o co esponding o he dis ance in he image be ween a p ojec ed and a measu ed image poin , was 0.4 pixels o image ligh 1 and 0.3 pixels o image ligh 2. The image poin measu emen accu acy o he signalised GCP was de e mined o be 0.2 pixels o bo h image blocks. The esul s o UAV ligh 2 including only 408 nadi images a e summa ised in Figu e 6. Each GCP was measu ed on a e age in 23 pho os. As an a p io i s anda d de ia ion, 5 mm was chosen o all h ee coo dina es o he GCP in he adjus men s, which co esponds o he accu acy achie ed by he geode ic GCP de e mina ion. This assump ion o he s anda d de ia ion was con i med by he adjus men using all GCP (Fig. 6 le column). E en wi h dec easing numbe o con ol poin s, he de ia ions (RMSE) a he check poin s emain a 10 mm o be e . I can also be seen ha using only a single GCP s abilises he esul o he adjus men in he posi ion and heigh o he check poin s ( igh columns in Fig. 6). F om his i is concluded ha despi e he accu a e RTK-GNSS measu emen s o he image posi ions du ing he ae ial ligh , a leas one GCP should be placed in he objec a ea o achie e an accep able esul o he ae ial iangula ion, especially a al i ude. The impo ance and in luence o g ound con ol poin s o ae ial pho o iangula ion, especially o ae ial ligh s wi hou RTK-GNSS, is shown by (Linds aed and Ke s en, 2018) o a ious p ojec s. Uni [ m ] F1-P1 F2-P1 F3-L1 F4-L1 F5-L1 Max. de . + -0 , 019 0 , 013 0 , 039 0 , 041 0 , 035 Max. de - -0,063 -0,029 -0,032 -0,036 -0,030 A . de -0,039 -0,000 -0,000 -0,002 0,006 S d. de . 0,008 0,008 0,015 0,019 0,015 Table 3. De ia ions (Z) o 3D poin clouds a 44 check poin s o ligh 1-5 and P1 and L1. 4.2 Poin -based compa ison Fo poin -by-poin compa isons, he sho es dis ance (in e ical di ec ion) be ween he check poin s (ChP) and he dense poin cloud is calcula ed. Due o he high poin densi y (see Tab. 4) and he la a ge signs, i is assumed ha he Z- coo dina e a ound he cen e o he a ge sign is he same. The dis ibu ion o GCP o he s udy a ea is shown in Fig. 1. Tab. 3 summa ises he mean, maximum (posi i e) and minimum (nega i e) de ia ions (Z in m) in he de i ed poin clouds o he di e en UAV ligh s o 44 check poin s. The dense poin cloud Figu e 5. Resul s o bundle block adjus men s wi h di e en con ol and check poin e sions o UAV ligh 1 using Zenmuse P1 came a. Figu e 6. Resul s o bundle block adjus men s wi h di e en con ol and check poin e sions o UAV ligh 2 using Zenmuse P1 came a. The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 342 was c ea ed in Me ashape wi h he esolu ion "medium" om he image da a o ligh s 1 and 2, while he poin clouds o ligh s 3-5 we e acqui ed di ec ly om he lase scanne and p ocessed in he DJI Te a so wa e. The esul s show ha he ae ial ligh wi h he combina ion o nadi and oblique images has a sys ema ic heigh o se o 39 mm, which also occu s in he ae ial iangula ion esul s due o he de ia ions (RMSE) a he check poin s in he same ange. This esul is also documen ed in Figu e 7 (le ) by he ed colou ing o he check poin s. In con as , only small local sys ema ic e ec s a e isible in Fig. 7 ( igh ), which, howe e , esul in small de ia ions a he check poin s. The smalles de ia ions a he check poin s we e achie ed wi h he nadi images ( ligh 2), as he maximum nega i e de ia ion anges om -29 mm o a maximum posi i e de ia ion o 13 mm and hus has a span o 42 mm (Tab. 3). Fo he h ee da a se s o he lase scanne , an equal le el o accu acy is achie ed in each da a se , which di e s only sligh ly om he good esul o image ligh 2. Figu e 7. Colou -coded de ia ions in al i ude a he check poin s - ae ial ligh 1 ( op) and ligh 2 (bo om) each wi h he Zenmuse P1 came a. 4.3 Line-based compa ison In he line-by-line compa isons be ween p o iles om he poin clouds o he i e UAV ligh s and e e ence da a, objec a eas wi h heigh di e ences we e selec ed in he s udy a ea scanned wi h he e es ial scanne (Fig. 8), such as s ai s (p o iles 1-3) and a house açade wi h oo s uc u e (p o ile 4). The quali y o he poin clouds was isually analysed he e using p o iles 2 (s ai s) and 4 (house wall) as examples (Fig. 9). In he isual compa ison be ween he gene a ed p o iles and he e e ence da a o he e es ial scanne , he measu emen noise in he poin clouds o he L1 lase scanne can be seen on he one hand and he qui e good ep oduc ion o he s ai s in he poin clouds o he UAV image ligh s on he o he hand (Fig. 9 le ). The compa ison o he esul s shows a e y simila esul o p o iles 1 and 3 as o p o ile 2. As expec ed, he poin cloud o image ligh 1 showed a e y good i o he house wall below he oo o e hang due o he oblique images in p o ile 4 (Fig. 9 igh ), while he o he poin clouds a e smoo hed in he a ea o he oo o e hang. Especially in p o ile 4, he ad an age o oblique images can be demons a ed i e ical s uc u es in dense poin clouds should be measu ed. Fo he compa ison o he p o iles, ai bo ne lase scanning da a om 2020 was also used, which was acqui ed on behal o he LGV Hambu g using a RIEGL VQ-780II lase scanne wi h a poin spacing o app ox. 10 cm as he esul . In his da a se , he s ai s a e also sligh ly smoo hed, bu due o he small numbe o poin s and p esumably good il e ing including smoo hing, measu emen noise is no ob iously isible. Figu e 8. Selec ed p o iles o compa ison wi h e e ence da a o e es ial scanne . Figu e 9. Compa ison o p o ile 2 c ea ed om di e en poin clouds wi h e e ence da a om e es ial lase scanning. The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 343 4.4 A ea-based compa isons Fo he a eal compa isons wi h he a ailable e e ence da a, he di e en poin clouds om he i e UAV ligh s in h ee selec ed es a eas we e analysed. The es a eas o he a eal 3D compa isons a e shown in Fig. 10. The selec ed a eas ep esen su aces wi h a ying su ace s uc u es: A ea 1 (pa ing s ones, conc e e and sand), A ea 2 (smoo h pa ing s ones) and A ea 3 (wood, sand and lawn). Fo he es a eas (a eas 1 and 2), poin clouds om e es ial lase scanning wi h he FARO Focus3D X330 a e a ailable as e e ence da a (Fig. 11 and 12), while o a ea 3, compa isons we e only made be ween he poin clouds om image ligh 2 (nadi images) as he bes da a se o he image-based poin clouds wi h he h ee di e en poin clouds o lase scanning (Fig. 13). In addi ion, a compa ison was also made wi h he ai bo ne lase scanning da a om he Riegl scanne (Fig. 14). Figu e 10. O e iew o es a eas in he Wilhelmsbu g Inselpa k (ou lined in ed): A ea 1 (pa ing s ones, conc e e and sand), A ea 2 (smoo h pa ing s ones) and A ea 3 (wood, sand and lawn). Tables 4 and 5 summa ise he de ia ions (in Z) be ween he 3D poin clouds o all i e ligh s and he TLS e e ence da a o a ea 1 and 2, which we e calcula ed in CloudCompa e, as we e he p e ious compa isons. The ollowing esul s can be summa ized:  Fligh 4 wi h he lase scanne L1 has he lowes numbe o poin s pe m2 due o i s ligh al i ude o 90 m abo e g ound and, oge he wi h ligh 5, he highes maximum de ia ions o he la ges span as he amoun o he sum o maximum nega i e and posi i e de ia ion.  Fligh 2 wi h he Zenmuse P1 came a has he bes esul s in e ms o maximum de ia ion, span, a e age de ia ion and s anda d de ia ion. Howe e , he numbe o poin s pe m2 o bo h a eas is lowe han o he o he ligh s, also due o he ligh al i ude. Only ligh 4 wi h lase scanne L1 lown a 90 m abo e g ound has a lowe numbe o poin s pe m2.  The di e ences be ween he wo lase scanne ligh s 3 and 5 a e e y small, so ha one can conclude om hese esul s ha he e is no di e ence in he esul o he wo scan modes epe i i e and non- epe i i e in he a ailable da a se s.  The image-based 3D poin clouds o ligh s 1 and 2 p o ide be e esul s han he poin clouds o he ligh s wi h he lase scanne . Wi h he combina ion o nadi and oblique images combined wi h he signi ican ly highe numbe o pho os, he highes poin densi y pe m2 is achie ed.  Especially in a ea 2 wi h he smoo h pa ing s ones, he image-based poin clouds achie e signi ican ly be e esul s han hose o he lase scanne .  Wi h s anda d de ia ions o 5 mm o 40 mm om he e e ence, good esul s we e achie ed o he di e en gene a ed poin clouds (P1 and L1) in he poin -by-poin and a ea-by-a ea compa isons. Tes a ea 1 F1-P1 F2-P1 F3-L1 F4-L1 F5-L1 Max. de . + 0 . 243 0 . 208 0 . 296 0 . 500 0 . 302 Max. de . - -0.252 -0.209 -0.283 -0.255 -0.251 S p an 0.495 0.417 0.579 0.755 0.553 A . de 0.056 0.020 0.025 0.029 0.013 S d. de . 0.029 0.029 0.028 0.038 0.029 Poin s / m 2 745.5 578.4 656.4 325.8 681.2 Table 4. De ia ions (Z) o 3D poin clouds o P1 and L1 a es a ea 1 o ligh 1-5 (Uni [m]). Tes a ea 2 F1-P1 F2-P1 F3-L1 F4-L1 F5-L1 Max. de . + 0 . 153 0 . 016 0 . 099 0 . 115 0 . 124 Max. de . - -0 , 047 -0 , 034 -0 , 220 -0 , 224 -0 , 244 S p an 0.200 0.050 0.319 0.339 0.368 A . de 0.037 0.003 -0.036 -0.022 -0.012 S d. de . 0.005 0.005 0.017 0.030 0.012 Poin s/ m 2 644.6 501.3 589.9 314.5 629.0 Table 5. De ia ions (Z) o 3D poin clouds o P1 and L1 a es a ea 2 o ligh 1-5 (Uni [m]). The ollowing Fig. 11-14 isualises he colou -coded de ia ions o he 3D compa ison calcula ed in CloudCompa e be ween he es da a se o he espec i e 3D poin cloud and he e e ence o compa a i e da a. The colou -coded scale shows he de ia ions in he ange o ±2.5 cm in g een, while he posi i e maximum wi h +25 cm is shown in ed and he nega i e minimum wi h -25 cm in blue. The colou -coded isualisa ion o he de ia ions makes i easie o ecognise sys ema ics e ec s in he esul . In he le -hand g aphs o Figu es 11 and 12, sys ema ic de ia ions (yellow colou ing) o he TLS e e ence da a can be seen in he poin cloud gene a ed by pho os o ligh 1 o es a ea 1 and 2. In con as , o he poin clouds o ligh 2, as al eady isible in p o ile 2 (Fig. 9 le ), de ia ions can only be seen a he edges o he s ai s. The de ia ions a he edges o he s ai s a e somewha mo e p onounced in he poin cloud o ligh 3 wi h he lase scanne (see cen e in Fig. 11 igh ). In he su ace o he es a ea, he di e ences o he e e ence da a a e somewha la ge , whe eby e ec s om he s ip adjus men a e p obably also isible he e. Fig. 12 shows an example o he measu emen noise o he senso o ligh 5 (L1) wi h a sligh sys ema ic e ec a al i ude (yellow colou ing). The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 344 Since no e e ence da a we e a ailable o es a ea 3, compa isons we e only made be ween he poin clouds o ligh 2 (nadi images) as he bes da a se o image-based poin clouds and he h ee di e en poin clouds om he L1 lase scanne (Fig. 13). The colou ep esen a ion o he de ia ions be ween he poin clouds o ligh 2 and he lase scanne poin clouds also shows sligh sys ema ic e ec s in heigh (yellow colou ing in Fig. 13 le , eddish colou ing in he le pa o Fig. 13 cen e and blue colou ing in he le pa o Fig. 13 igh ). O e all, he heigh di e ences be ween he poin clouds a e wi hin he speci ied accu acy ange o he Zenmuse L1 senso (see chap e 3.2). Fo a isual compa ison o he UAV-based poin clouds, poin clouds acqui ed by ai bo ne lase scanning (ALS) wi h he RIEGL VQ-780II lase scanne could also be used. The da a was p o ided by LGV Hambu g om an ALS su ey in Ma ch 2020. These ALS da a canno se e as a e e ence due o he low poin densi y o 23 poin s pe m² and he p esumably poo e heigh accu acy, bu hey e eal sys ema ic e ec s in he UAV-based poin clouds. Fig. 14 isualises he esul s o he 3D compa isons. He e i is again clea ha he poin clouds o ligh 1 a e sys ema ically oo high o e all, while he poin clouds o ligh 2 and o he ligh s wi h he L1 i oge he su p isingly well. The e, he di e ences a e, among o he hings, due o he di e en eco ding da e, he ege a ion g ow h and he di e en accu acy anges. 5. CONCLUSION AND OUTLOOK This pape summa ises he i s esul s o he accu acy in es iga ions o he UAV sys em DJI Ma ice 300 RTK wi h he senso s Zenmuse P1 and L1 in he Hambu g es ield Figu e 11. Compa ison o poin clouds o TLS ( e e ence) o UAV ligh s 1, 2 and 3 on es a ea 1. Figu e 12. Compa ison o poin clouds o TLS ( e e ence) o UAV ligh s 1, 2 and 5 on es a ea 2. Figu e 13. Compa ison o poin clouds o ligh 2 (P1) o poin clouds o ligh s 3, 4 and 5 (each L1) on es a ea 3. Figu e 14. Compa ison o poin clouds o ai bo ne lase scanning wi h he RIEGL VQ-780II o poin clouds o UAV ligh s 1, 2 and 5 on es a ea 3. The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 345 Inselpa k. Fligh planning and con ol was e y easy using he DJI Pilo app, which is e y use - iendly and allows au oma ed ligh s. Compa ed o he DJI Phan om 4 P o, he ligh ime is wice as long due o he wo ba e ies on he ai c a pla o m. A sys em shu down is no necessa y when changing he ba e ies because bo h ba e ies can be changed one a e he o he . Due o he swi ched-on powe supply, he pa ame e s o he in e io o ien a ion p esumably also emain s able o he came a. The esul s o he ae ial iangula ions show ha o UAV p ojec s wi h somewha lowe accu acy equi emen s o checks poin s (XYZ = 3-5 cm), e.g. opog aphic applica ions, i is possible o compu e he bundle block adjus men e en wi hou GCP coo dina es, since he s anda d de ia ions o he ex e io o ien a ion pa ame e XYZ can nowadays each 1-2 cm in XY and 2-3 cm in heigh Z by RTK-GNSS measu emen s. Fo easons o eliabili y, a leas one bu p e e ably i e GCP should be used a he co ne and in he cen e o objec space. Fo he esul s o ae ial iangula ion, an accu acy o one GSD was expec ed, bu his was only achie ed in ae ial ligh 2 when he pho o block was o ien ed using a leas i e GCP. The ae ial iangula ion o he nadi images ( ligh 2) achie ed o e all signi ican ly be e esul s a he check poin s han he ligh 1 wi h he combina ion o nadi and oblique images, whe e he heigh componen showed de ia ions o up o 42 mm o all bundle block adjus men s. This combina ion o image sho s du ing he ae ial ligh (nadi -backwa d-sideways) p o ides e y good co e age o he e ain su ace, bu he je ky mo emen s o he came a and he ongoing e ocusing o he lens due o he changing shoo ing pe spec i es p obably p o ide uns able came a geome y. Howe e , his assump ion s ill has o be e i ied wi h he help o he image da a by spli ing he ae ial image con igu a ion o ligh 1 in o h ee blocks (nadi images, oblique images backwa ds and oblique images sideways) so ha h ee sepa a e came a calib a ions can be calcula ed. The examina ions o he 3D poin clouds showed a clea esul : Ae ial ligh 2 wi h nadi images p oduced he bes esul s in compa ison wi h he o he ligh s, while wi h he image da a o ligh 1 a sys ema ic heigh shi occu ed in he check poin s, in he p o iles and also in he a ea-by-a ea compa ison using e e ence da a, which was no o be expec ed in his way. The h ee poin clouds o he Zenmuse L1 lase scanne showed e y simila esul s, which a e e en sligh ly be e han he accu acy speci ica ions o he manu ac u e . A signi ican di e ence in he quali y o he poin clouds could no be ound in he wo scanning modes in he p esen s udy. In es iga ions in o he pe o mance o he lase scanne o applica ions in he de ec ion o ege a ion such as ees and bushes ha e no ye been ca ied ou wi h his da a se s. ACKNOWLEDGEMENTS We would like o hank Dipl.-Ing. Holge Di ks (NLWKN in No den) o p o iding he Ma ice 300 RTK wi h he Zenmuse P1 came a and o lying o e he es ield. We would also like o hank Dipl.-Ing. Ch is ian Ha l-Rei e (DAI in Bonn) o p o iding he Zenmuse L1 lase scanne and o his suppo du ing he UAV ligh s. The es si e a Inselpa k Hambu g- Wilhelmsbu g was se up, signalised and made a ailable by he s a o he S a e O ice o Geoin o ma ion and Su eying Hambu g unde he di ec ion o M.Sc. Ma in Helms, o which we would like o exp ess ou since e hanks. REFERENCES DJI, 2022: Technical speci ica ions Zenmuse L1. h ps://www.dji.com/de/zenmuse-l1/specs, las access 03.03.2022. Ge ke, M., P zybilla, H.-J., 2016: Accu acy Analysis o Pho og amme ic UAV Image Blocks: In luence o onboa d RTK-GNSS and C oss Fligh Pa e ns. Jou nal o Pho og amme y, Remo e Sensing and Geoin o ma ion, (1), 17- 30. h ps://doi.o g/10.1127/p g/2016/0284. Ke s en, T., Linds aed , M., 2022: UAV-basie e Bild lüge mi RTK-GNSS – b auchen wi da noch Passpunk e? UAV 2022 – Inno a ion und P axis. Sch i en eihe des DVW, Band 100, Bei äge zum 195. DVW-Semina am 28. und 29. Mä z 2022 (online), Wißne -Ve lag, Augsbu g, 39-58. Ke s en, T., Schlöme , F., P zybilla, H.-J., 2020: Ae o- iangula ion on UAV-Bildda en de Zeche Zolle n – Die E gebnisse e schiedene UAV-Sys eme und zweie So wa epake e im Ve gleich. UAV 2020 – The Sky is he Limi ? Sch i en eihe des DVW, Band 97, Bei äge zum 185. DVW-Semina am 17. und 18. Feb ua 2020 in Ha ingen, Wißne -Ve lag, Augsbu g, 121-140. Linds aed , M., Ke s en, T., 2018: Zu Bedeu ung on Passpunk en bei de Ae o iangula ion UAV-basie e Bild lüge. UAV 2018 - Ve messung mi unbemann en Flugsys emen, Sch i en eihe des DVW, Band 89, Bei äge zum 169. DVW-Semina am 19. und 20. Feb ua 2018 in Hambu g, Wißne -Ve lag, Augsbu g, 81-101. LIVOX, 2022: AVIA – A a glance. h ps://www.li ox ech.com/de/a ia, las access 03.03.2022. P zybilla, H.-J., Bäumke , M., Luhmann, T., Has ed , H., Eile s, M., 2020: In e ac ion be ween Di ec Geo e e encing, Con ol Poin Con igu a ion and Came a Sel -calib a ion o RTK- based UAV Pho og amme y. In . A ch. Pho og amm. Remo e Sens. Spa ial In . Sci., XLIII-B1-2020, 485-492, h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2020-485- 2020. P zybilla, H.-J., Linds aed , M., Ke s en, T., 2019: In es iga ions In o he Quali y o Image-based Poin Clouds om UAV Image y. In . A ch. Pho og amm. Remo e Sens. Spa ial In . Sci., XLII-2/W13, 539-545. Singh, I., 2020: Wha you need o know abou DJI’s i s -e e d one LiDAR payload. Top 7 Fea u es o he DJI L1, h ps://en e p ise-insigh s.dji.com/blog/zenmuse-l1- op-7- ea u es, las access 03.03.2022. Zhao, B., Li, J., Wang, L., Shi, Y., 2020: Posi ioning accu acy assessmen o a comme cial RTK UAS. P oc. SPIE 11414, Au onomous Ai and G ound Sensing Sys ems o Ag icul u al Op imiza ion and Pheno yping V, 1141409. h ps://doi.o g/10.1117/12.2557899. Zhao, S., 2021: A Comme cial PPK Solu ion o Phan om 4 RTK. GIM In e na ional, Business Guide, Issue 1, 30-32. The In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, Volume XLIII-B1-2022 XXIV ISPRS Cong ess (2022 edi ion), 6–11 June 2022, Nice, F ance This con ibu ion has been pee - e iewed. h ps://doi.o g/10.5194/isp s-a chi es-XLIII-B1-2022-339-2022 | © Au ho (s) 2022. CC BY 4.0 License. 346