Validation of the QuestUAV PPK System

Transcription

1 Validation of the QuestUAV PPK System 3cm in xy, 400ft, no GCPs, 100Ha, 25 flights Nigel King 1, Kerstin Traut 2, Cameron Weeks 3 & Ruairi Hardman 4 1 Director QuestUAV, 2 Data Analyst QuestUAV, 3 Production Engineer QuestUAV & 4 Manager QuestUAV Glossary DSM GCP GIS GNSS GPS GSD OS PPK RTK UAV Digital Surface Model Ground Control Point Geo-Information System Global Navigation Satellite System Global Positioning System Ground Sampling Distance Ordnance Survey Post Processing Kinematic Real Time Kinematic Unmanned Aerial Vehicle Introduction The new QuestUAV Post-Processing Kinematic (PPK) drone combines high-resolution aerial photography with high-accuracy topographic data. A dual band GPS/GLONASS receiver on board the PPK drone allows cm-level position referencing of orthomosaics and 3D models without the need on physical Ground Control Points (GCPs). This saves hours of mission planning and setup time, physically measuring location points and walking the survey site for placement.

2 All GNSS positioning data are stored on board the QuestUAV PPK system, which eliminates the need for a real-time data link with a fixed reference station while still guaranteeing RTK cm-level position accuracy. Camera locations are corrected after the flight during the post-processing phase, when GNSS receiver data is combined with base station data to calculate the exact position of each camera exposure. The corrected image positions are directly added to the image EXIF and can be imported to photogrammetric software packages such as Pix4Dmapper or Agisoft Photoscan. Objectives The objective of this whitepaper is to validate the spatial accuracy (precision) of the new PPK enabled QuestUAV-200 series aircraft on the basis of independently measured verification points (GCPs) and repeated aerial surveys of a test site in the UK. Claims The authors claim the following accuracies using the QuestUAV PPK system without the use of ground control points on the test area, (subject to final validation by an independent, qualified survey authority). X Y Z QuestUAV PPK Accuracies 3 cm 3 cm 12 cm Study Area The test site is located on the North East coast of England and covers 100ha of agricultural land, coastal dunes and beach. By being very close to the North Sea weather conditions are windy and changeable, which allows to perform stress tests for both the QuestUAV aircraft and image data quality. Figure 1: Test site at Low Hauxley in Northeast England including the location of reference points (yellow dots).

3 Equipment The QuestUAV-200 PPK aircraft is equipped with a dual-frequency L1/L2 receiver tracking both GPS and GLONASS signals. 132 hardware channels allow to simultaneously track all visible GPS/GLONASS satellites. In parallel the receiver records the shutter events of the camera and logs the information during flight on an on-board SD card. The sensor on board the Q-200 PPK drone is the Sony A6000 camera with a 16mm wide angle lens providing a nominal Ground Sampling Distance (GSD) of 2.9cm at 400ft. The aircraft is launched with the standard QuestUAV tripod launch line and can be landed via parachute or belly landing. Figure 2: Parachute landing of the Q-200 PPK aircraft. Methodology The accuracy of the QuestUAV PPK system is validated by comparing the results of repeated aerial surveys with precisely measured verification points (GCPs). The verification workflow includes the following steps: 1 Reference Data 2 Reference Base Station 3 Aerial Surveys Marking verification points on the ground Measuring verification points by an independent surveying company Setup of a reference base station Repeated aerial surveys over the test site with the QuestUAV-200 PPK drone 4 Post-processing Post-processing of camera locations and image processing 5 Statistics Extraction of target points and statistics Reference Data A total of 40 target points were distributed over the 100 ha test site and measured by an independent surveying company. The majority of the points were placed on tarmac and permanent structures such as rocks and stable stone walls in order to avoid any movements for the duration of the accuracy study.

4 All points were measured independently in two separate runs with a double-frequency GPS/GLONASS receiver taking RTK corrections from a mobile base station. The base station was set up using the baselines of six reference stations of the British Ordnance Survey. The ground survey produced a horizontal accuracy of 4mm and a vertical accuracy of 10mm. By using yellow marker boards target points are clearly visible in the aerial images. The target points are the basis for assessing the spatial accuracy of orthomosaic and Digital Elevation Model (DEM). Figure 3: Surveyor measuring the reference points on the ground. Reference Base Station The PPK workflow does not need a real-time data link with a base station during a flight. However, GNSS information has to be recorded by a base station parallel to a flight in order to allow for positioning corrections later during the post-processing phase. In the UK, publicly available GNSS recordings taken by the Ordnance Survey are only available at a 30 second interval. As the QuestUAV PPK system requires one second correction data, the QuestUAV team set up their own base station close to the test site recording GNSS correction data during the flights. Aerial Survey The test site was flown with the QuestUAV-200 PPK aircraft repeatedly over a one month period while testing different flight and sensor settings, including: Flight altitudes: 300ft and 400ft Image overlap: 69m/60m/53m lag distance Flight path: Parallel lags/ perpendicular lags Camera settings: Auto/ aperture priority/ shutter priority/ manual Camera lenses: 16mm wide angle lens / 35mm long lens Figure 4: Ready for takeoff. Launch setup for the Q- 200 PPK aircraft.

5 A total of 25 flights were completed from August 8 to September 5 covering different weather and light conditions. Half of the flights were carried out in sunny weather while the other half was flown in overcast and rainy conditions. Wind speeds were high for the majority of the flights with up to 35kts at height. During the flight hours there was no significant geomagnetic activity recorded. Post-Processing After flight camera positions were corrected by combining GNSS recordings of the aircraft with the corresponding information of the reference base station on the ground, replacing the positions embedded in the images (EXIF data) with more precise cm-level RTK values. Photogrammetric processing was completed with the Agisoft PhotoScan software, including the generation of orthomosaic and digital elevation model and their export as GeoTiff files. Figure 5: Orthomosaic (left) and DSM (right) of the test site generated from a flight with the Q-200 PPK aircraft. Statistics Data analysis was performed inside the open-source Geo-Information System (GIS) software QGIS. For each flight orthomosaic and DSM were imported into QGIS and the 3D image coordinates of the target points extracted. The image coordinates were compared with the coordinates measured during the ground survey and their accuracies were determined by using the Root Mean Square Error (RMSE) as statistical measure. The RMSE is the standard error measure to estimate geo-spatial precision. It represents the sample standard deviation of the differences between the coordinates measured in flight and the observed coordinates during the ground survey.

6 Figure 6: Identifying target points in the orthomosaic (left) and comparing their position with the true location of the points as surveyed on the ground. Results During a workflow optimization process of 10 of the 25 flights, optimal camera settings and flight parameters were tested in different light and weather conditions. Best results were achieved with the Sony A6000 camera and the 16mm wide angle. We further improved results by using a manual configuration of the camera, fixed aperture and fixed ISO, dependent on light and weather conditions. After the optimization process the actual accuracy assessment was carried out on the basis of 15 flights and processed without any ground control, but using 40 check points to confirm accuracies. The table below summarizes the results of the validation: Table 1: Accuracy assessment on the basis of 15 flights with 40 check points each (ground reference). X (m) Y (m) Z (m) Mean Error Standard Deviation RMSE RMSE XY RMSE XYZ 0.130