Ground Control Configuration Analysis for Small Area UAV Imagery Based Mapping

Transcription

1

2 Ground Control Configuration Analysis for Small Area UAV Imagery Based Mapping ASPRS IGTF 2017, Baltimore, MD March 15 th, 2017 Presenter: David Day, CP, GISP Wes Weaver Keystone Aerial Surveys, Inc.

3 Summary of activities 1. Set out to test as many sensors over the same location as possible during the same day 2. Location as close as possible to Northeast Philadelphia Headquarters for reduction of costs

4 Manned Sensors Used High resolution, multispectral digital image sensor Vexcel Ultracam Falcon Prime Focal Length: mm Pixel Size: 6 μm Image format: x pixels Teledyne Optech Galaxy Dynamic Field of View (0-60 degrees) Multiple pulse capabilities Record 8 range and intensity measurements per pulse Unique Swath Tracking mode maintains constant-width flight lines for consistent data density in variable terrain. Ultracam Falcon Prime Galaxy

5 Manned Aircraft Used Cessna 210 single engine aircraft for Ultracam Falcon Operating Ceiling of 26,000 Cruise speed of 190 kts Single camera port Cessna 310 twin engine aircraft for Galaxy Operating Ceiling of 30,000 Cruise speed of 205 kts Single camera port

6 UAS Used Altavian Nova F6500 (fixed wing) Hand-launched Endurance: 90 min Cruise Speed: 35mph Altitude: 200ft AGL/10, 000ft MSL Wing Span: 108" Weight: 15lbs Maximum Take Off

7 UAS & Sensors Used SteadiDrone Mavrik X8 (Rotorcraft) 8 Propellers Dimensions: 960mm x 970mm x 495mm Max Weight with payload: 18 kg Max Altitude: 4000m Estimated Endurance: 10 min Altavian Canon MP22 for RGB and CIR Effective Pixels: 22MP Image Max Size: 5184x3456 Focal Length: 20mm (effective 35mm) Sony A7r on Mavrik Effective Pixels: 36.4MP Image Max Size: 7360x4912 Sony/Zeiss FE 35mm f/2.8

8 Planned Location and Flights AGA Farms in Perkasie, PA 10 acres AOI on 100 acre farm Fully compatible with 333 rules working under at the time

9

10

11 Planned Location and Flights Falcon Prime: 2cm and 4cm 18 ppm2 Altavian RGB: 400ft & 200 ft Altavian CIR: 400ft & 200 ft Mavrik RGB: 200ft & 400 ft (with and without IMU)

12 Planned Location and Flights UltraCam Falcon 2cm GSD Galaxy Flight Plan

13 Planned Location and Flights Altavian MP22 400ft Plan Sony A7r 200ft Plan

14 Timeline Ground Control Placed: June 28 th Galaxy flown: June 29 th 7:44 am local time Altavian: 10:00 10:10 am UltraCam Falcon: 10:33 10:39 am Mavrik: 10:42 10:52 am UltraCam Falcon: 10:46-10:52am Altavian: 11:06 11:20 am Mavrik: 11:45 11:54 am Altavian: 12:14 12:36 pm

15 Capture Statistics Platform Sensor Bands GSD Altitude Lines Images Time Cessna 210 UC FalconP RGBI 4cm 2575 ft min Cessna 210 UC FalconP RGBI 2cm 1375 ft min F6500 Cannon RGB 2.76cm 400 ft min F6500 Cannon RGB 1.38cm 200 ft min F6500 Cannon CIR 2.76cm 400 ft min F6500 Cannon CIR 1.38cm 200 ft min Mavrik Sony RGB 1.47cm 400 ft min Mavrik Sony RGB 0.72cm 200 ft min The Galaxy captured the area in 5 strips in 9 minutes of production flight time. The point density averaged 17.8 points per square meter using a PRF of 300 khz and a scan frequency of 92 Hz at an altitude of 3000 feet AGL.

16 Results of AT RMSE on Check Points DX in cm DY in cm DZ in cm Flight 0 GCP 2GCP 3GCP 0 GCP 2GCP 3GCP 0 GCP 2GCP 3GCP Sony 200ft Sony 400ft Cannon 200ft Cannon 400ft UCFp 1387ft UCFp 2560ft GCPs used resulted in 15 Check Points 2 GCPs used resulted in 13 Check Points 3 GCPs used resulted in 12 Check Points

17 Results of AT RMSE on Check Points DX in cm DY in cm DZ in cm Flight 5GCP 7GCP 10GCP 5GCP 7GCP 10GCP 5GCP 7GCP 10GCP Sony 200ft Sony 400ft Cannon 200ft Cannon 400ft UCFp 1387ft UCFp 2560ft GCPs used resulted in 10 Check Points 7 GCPs used resulted in 8 Check Points 10 GCPs used resulted in 5 Check Points

18 Results of AT All measurements in meters. The UAV residuals are worse for 10 control points with all X and one Y. From 3 GCPs onward for most are at or below the expected accuracy of the control.

19 Results of AT

20 Control Layout Statistics Percentage of RMSE Improvement by GCP Count Flight 3 to 5 x/y 3 to 5 z 5 to 7 x/y 5 to 7 z 5 to 10 x/y 5 to 10 z Sony 200ft -6% 81% 48% -4% 23% 4% Sony 400ft 10% 72% -5% -13% 14% 54% Cannon 200ft 25% 66% 16% 26% 8% 11% Cannon 400ft -14% 57% -14% 10% 12% 10% Mean 4% 69% 11% 5% 14% 20%

21 UltraCam Image Sharpness There was concern that the imagery from the UltraCam Falcon taken in the Cessna 210 could have resulted in image blurring due to heat/exhaust. Although there were no visible signs of blur or fuzziness in the imagery, a standard sharpening was performed on the imagery using the Vexcel UltraMap suite. Standard vs Sharpened Imagery The two UltraCam flights were then run through the same AT as the unsharpened imagery. The results below show that sharpening had a slightly negative impact (generally 2-3 millimeters) and indicates that the imagery was not negatively affected by aircraft heat or exhaust Sharpened RMSE - Standard RMSE No Sharpened is: Better Worse Change No GCP 2GCP 3GCP 5GCP 7GCP 10GCP UCFp 1387ft Dx Dy Dz UCFp 2560ft Dx Dy Dz

22 Study using MEMS IMU with L1 GPS In January of 2017, the area was flown again using new control, a new UAV with the same Sony A7R and Vectornav VN-200

23 Study using MEMS IMU with L1 GPS Due to time and budget constraints, only 5 GCPs were placed with the same methodology Sony A7R flown with DJI M600 at 200ft Raw EO values from the sensor were compared to Controlled AT EO results Raw EO error (raw EO minus controlled adjusted EO) Delta X Delta Y Delta Z Delta Omega Delta Phi Delta Kappa Mean Mean (abs) Min (abs) Max (abs) RMSE

24 Study using IMU with L1 GPS Results of January Flight with M600 UAV at 200 ft. were poor, reasons for this The antenna offsets from the camera to antenna are further on the M600 than with the Marvrik and without being entered, caused GPS error and possibly poor IMU data through the Kalman filter A different FMS was used to trigger the camera that did not rotate the aircraft or camera for each line, so GPS shift/drift cannot be modeled and removed The small number of control points does not lend to a rigorous solution Error on Checkpoints in meters Further study and flights are being conducted to determine methodology to achieve better results RMSE DX DY DZ 0 GCP 5 Check GCP 4 CHECK GCP 3 CHECK GCP 2 CHECK DX in cm DY in cm DZ in cm Flight 0 GCP 2GCP 3GCP 0 GCP 2GCP 3GCP 0 GCP 2GCP 3GCP Sony 200ft June Sony 200ft January

25 Conclusions The Canon 22MP camera mounted on the fixed wing aircraft performed better than the Sony A7R 36MP camera on the rotorcraft in lower control situations. However, with more control, the Sony sensor produced accuracies comparable to the UCFp. Despite the high quality Zeiss 35mm lens on the Sony A7R, the adjustments performed by the software would suggest an instability in the lens. In the case of the Zeiss lens, the situation of less control caused the errors to manifest without the ability to be systematically corrected. For the UltraCam imagery, it is clear that applying sharpening techniques to imagery that is already crisp results in no appreciable improvement for automated aerial triangulation results.

26 Conclusions Control point accuracy is increasingly important as pixel sizes allow for better results. It is increasingly possible to exceed the limits of GPS control The largest improvement in terms of RMSE in Z was from 3 to 5 GCPS (69%) while increasing from 5 to 7 resulted in less impressive improvements in x/y (11%) and z (5%). Using a MEMS IMU with L1 GPS does not show improved results as compared to the navigation GPS and telemetry data on board the UAV The use of a MEMS IMU for systems unable to export location data is an excellent an reliable solution

27 Special Thanks Special thanks must go to Wes Weaver of Keystone Aerial Surveys, Inc. for his commitment to running the same datasets through so many iterations!

28 Thank You! Contact Us David Day: Web: Facebook: