BEYOND POINTS AND PIXELS Stereographic Compilation from UAS

Transcript

1. BEYOND POINTS AND PIXELS Stereographic Compilation from UAS
2. CLIENT NAME HERE RFP or solicitation name here ORRIN THOMAS Sr. Software Engineer Cardinal Systems orrin@cardinalsystems.net CHRISTIAN STALLINGS, CP Research & Development Manager McKim & Creed cstallings@mckimcreed.com RESEARCH TEAM
3. CLIENT NAME HERE RFP or solicitation name here THE SUCCESS OF DRONES AND MAPPING  Often times for many projects, image derived point clouds are all you need.  Comparatively to conventional survey, this approach has changed the way certain projects are approached by lowering the cost while increasing the speed, data quality, and safety
4. CLIENT NAME HERE RFP or solicitation name here IDPC VS. TERRESTRIAL LIDAR  IDPC’s can represent rolling bare earth terrain quite accurately but does not represent steep vertical relief well  A & E = curb and gutter  B & F = roof vent  C & D = church steeple  D & F = roof gutter line
5. CLIENT NAME HERE RFP or solicitation name here POINTS AND PIXELS  Up to this point UAS tools for mapping have focused on points and pixels  The fact is that most mapping products delivered by surveyors don’t deal in either  Most mapping products are VECTOR based
8. STEREOGRAPHIC COMPILATION  Stereo Comp takes UAS mapping to the limits of what is possible with a simple RGB sensor  It is an intuitive process that our eyes do naturally similar to a 3D movie
9. STEREO VERSUS CONVERGENT Stereo  Faster  Able to measure more features  Better photo interpretation Convergent  Potentially more accurate  Doesn't require special hardware  Doesn’t require particular image geometry
10. CLIENT NAME HERE RFP or solicitation name here EXPERIMENT SETUP Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Motion Blur / Pixel Flight 1 NS 1.7 800 6.9 200 y 0.22 Flight 2 NS 1 800 4.3 125 y 0.24 Flight 3 NS 2.5 800 10.3 300 y 0.23 Flight 4 NS 3.3 800 13.8 400 y 0.23 Flight 5 NS 1.7 800 6.9 200 y 0.22 Flight 6 EW 1.7 800 6.9 200 y 0.22 Flight 7 EW 1.7 800 9.2 200 y 0.30 Flight 8 EW 1.7 800 13.8 200 y 0.45 Flight 9 EW 1.7 800 4.6 200 y 0.15 Flight 10 EW 1.7 800 6.9 200 y 0.22 Flight 11 NS 1.7 800 6.9 200 y 0.22 Flight 12 NS 1.7 1000 6.9 200 y 0.18 Flight 13 NS 1.7 500 6.9 200 y 0.36 Flight 14 NS 1.7 2000 6.9 200 y 0.09 Flight 15 NS 1.7 800 6.9 200 y 0.22 Flight 16 EW 1.7 800 6.9 200 y 0.22 Flight 17 EW 1.7 800 6.9 200 N 0.22 Flight 18 EW 1.7 800 6.9 200 Y 0.22 Flight 19 NS Varied 800 6.9 200 Y 0.22 Flight 20 NS Varied 800 6.9 200 y 0.22 Flight 21 EW 1.7 800 6.9 200 Y 0.22 Flight 22 NS 1.7 800 6.9 200 y 0.22
11. CLIENT NAME HERE RFP or solicitation name here CONTROL DATA • The camera was pre-calibrated on site • A Trimble R-10 was used to establish 2 ground control points • A Trimble S5 Robotic Total Station was used to collect 63 photo id shots with a closure of 1:50,000 • A Leica C10 was used to collect additional data on above ground featured and fit the other data at 0.011 ft. RMSEz
12. CLIENT NAME HERE RFP or solicitation name here TERRESTRIAL SCAN CONT.  The Terrestrial Scan was then used to extract additional above ground points that could be measured in the UAS data  18 challenge points that were on edges or tips of objects were selected  32 points that were on flat roof surfaces were also collected
13. CLIENT NAME HERE RFP or solicitation name here CONTROL LAYOUT  4 points selected were selected at the corners of the test site  The 63 total station points were paired down to 20 points randomly to reduce the amount of manual measurements to a reasonable amount (500 per flight vs. 2k)  18 “Challenge Points” were selected from the Terrestrial scan  32 flat roof points were selected from the Terrestrial scan
14. CLIENT NAME HERE RFP or solicitation name here MEASURING ACCURACY (STEREO)  Each control and checkpoint was measured in each stereo model that it appeared (~5-25 models per point) for a total of ~500 measurements per flight  A tool was created to drive the compiler to each point and randomly move the curser in the XYZ direction forcing the compiler to move the curser back to the point to measure it
15. CLIENT NAME HERE RFP or solicitation name here MEASURING ACCURACY (IDPC)  Image derived Point Clouds (IDPC) were created using Agisoft Photoscan.  Using LP360, a control report was run on the Z delta between the IDPC and the control point.  Only the Z axis was tested on the IDPC datasets
16. CLIENT NAME HERE RFP or solicitation name here DOES ALTITUDE AFFECT ACCURACY? 2 control flights were conducted at the beginning and end of test block 1 Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Motion Blur / Pixel Flight 1 NS 1.7 800 6.9 200 y 0.22 Flight 2 NS 1 800 4.3 125 y 0.24 Flight 3 NS 2.5 800 10.3 300 y 0.23 Flight 4 NS 3.3 800 13.8 400 y 0.23 Flight 5 NS 1.7 800 6.9 200 y 0.22 Flight Speed was adjusted to maintain a similar motion blur for each flight
17. CLIENT NAME HERE RFP or solicitation name here TEST: GSD/ ACCURACY  In traditional photogrammetry, GSD and mapping accuracy are generally correlated strongly.  In other studies, they have not been well correlated with UAS IDPC
18. CLIENT NAME HERE RFP or solicitation name here GSD / ACCURACY CONT. The relationship between vertical RMSE in GSD units and the altitude. The relationship between vertical RMSE in cm and the altitude 0 1 2 3 4 5 6 7 0 100 200 300 400 500 VERTICAL RMSE (CM) FLIGHT ALTITUDE 0 1 2 3 4 0 100 200 300 400 500 VERTICAL RMSE (GSD) FLIGHT ALTITUDE PT Check Points Stereo Check Points IDPC Check Points
19. GSD / ACCURACY CONT.  The vertical RMSE / Altitude linear correlation squared were: • 0.8146 for Stereo measurements • 0.7909 for Photo Triangulation measurements • 0.1177 for Image derived point cloud measurements  IDPC’s showed a higher level of noise and a weak correlation to the GSD resulting in a more randomized result.
20. SHUTTER MODE (MECHANICAL SHUTTER) 3 flights were conducted to test the affect of the mechanical shutter (two with and one without) Mechanical shutters (more specifically leaf shutters) are increasingly common on UAS camera systems The vast majority of all camera systems today are CMOS based vs. CCD which digital mapping cameras traditionally use. Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Motion Blur / Pixel Flight 16 EW 1.7 800 6.9 200 y 0.22 Flight 17 EW 1.7 800 6.9 200 N 0.22 Flight 18 EW 1.7 800 6.9 200 Y 0.22
21. SHUTTER MODE (MECHANICAL SHUTTER) RESULTS 0 0.01 0.02 0.03 0.04 0.05 0.9 0.95 1 1.05 1.1 1.15 1.2 3-AXIS RMSE (M) Accuracy and Shutter Type Leaf Shutter Curtain Shutter The two flights with the leaf shutter had similar RMSE (0.02 m). The flight with the curtain shutter was 0.026 m worse (12 times the difference between the two similar flights). This difference is statistically significant (p<0.001, two-sample T-Test). It is fascinating that such a small systematic error is so significant.
22. DOES FLIGHT SPEED AFFECT ACCURACY? Flight speeds ranging from 4.6 – 13.8 m/s at 200 ft were tested This corresponds to a predicted motion blur of 0.15 pixels – to 0.45 pixels Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Motion Blur / Pixel Flight 6 EW 1.7 800 6.9 200 y 0.22 Flight 7 EW 1.7 800 9.2 200 y 0.30 Flight 8 EW 1.7 800 13.8 200 y 0.45 Flight 9 EW 1.7 800 4.6 200 y 0.15 Flight 10 EW 1.7 800 6.9 200 y 0.22
23. FLIGHT SPEED TESTING CONCLUSION  Flight speed was poorly correlated with accuracy  Although the motion blur was nearly 1\2 a pixel at the higher speed, we did not see a strong correlation that would be expected suggesting other potential sources of error R² = 0.0574 0.0270 0.0280 0.0290 0.0300 0.0310 0.0320 0 2 4 6 8 3-AXIS RMSE (M) FLIGHT SPEED (M/S) Accuracy and Flight Speed
24. SHUTTER SPEED TESTING Shutter speeds were tested between 1/500 – 1/2000 which is the fastest this camera will shoot while utilizing the mechanical shutter; The Motion blur based on shutter speed was between 0.09 and 0.36 pixels. Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Motion Blur / Pixel Flight 11 NS 1.7 800 6.9 200 y 0.22 Flight 12 NS 1.7 1000 6.9 200 y 0.18 Flight 13 NS 1.7 500 6.9 200 y 0.36 Flight 14 NS 1.7 2000 6.9 200 y 0.09 Flight 15 NS 1.7 800 6.9 200 y 0.22
25. SHUTTER SPEED TESTING  Correlations were again inconclusive.  Based on motion blur, we would expect a higher correlation suggesting unaccounted for variables. R² = 0.4407 0.0000 0.0050 0.0100 0.0150 0.0200 0.0250 0.0300 0 500 1000 1500 2000 2500 3-AXIS RMSE (M) SHUTTER SPEED (S-1) Accuracy and Shutter Speed
26. CHALLENGE POINT ACCURACY COMPARISONS In order to have a more manageable dataset for comparison, only 4 flights were compared to each other. These flights were all our NS control flights that had the same altitude, shutter speed, and ground speed. Mission Flight Direction GSD (CM) Shutter Speed Flight speed (MPH) Altitude (feet) Mechanical Shutter Flight 1 NS 1.7 800 6.9 200 y Flight 5 NS 1.7 800 6.9 200 y Flight 11 NS 1.7 800 6.9 200 y Flight 15 NS 1.7 800 6.9 200 y
27. CHALLENGE POINT ACCURACY COMPARISONS 0 2 4 6 8 10 12 Z ERROR (M) IDPC VS. STEREO "CHALLENGE POINT" ACCURACIES IDPC Stereo
28. SOLAR CITY ROOF SITES  Rooftop survey of three sites for a new solar panel installation.  It was important to know location and shape of all features including height of each object to determine if they would obstruct the view of each panel.  A detailed inventory of every existing rooftop features, the type of feature, and the height and location of those features was needed. Many of these items were as small as outlet boxes which required a high-fidelity approach
29. SOLAR CITY ROOF SITES  All 3 locations were located throughout the San Antonio area (flown with LAANC approval)  All 3 sites were flown on the same day as well as establishing ground control  Based on the amount of field time for these 3 sites, 6 sites could easily be flown and surveyed in a single day by a 2 man crew.
30. SOLAR CITY ROOF SITES  Compilers extracted 3D vector work for the entire roof sites and easily discerned small objects such as outlet boxes.  CAD drawings were created for each roof survey utilizing this method and verified compared to traditional survey shots at an accuracy of less than 1\10th of a foot  Approach cut time almost in half while increasing safety and reducing cost. 12 7 58% 41% Total Survey Days Total UAS Days Savings (Time) Savings (Cost)
31. SOLAR CITY ROOF SITES CONT.  Because some roofs lacked parapet walls, a cherry picker and harness system would have been required near the roof edge  Survey personnel was placed on the roof using centrally located access to place checkpoints for additional verification  Based on our new research, this will not be necessary for future projects and checkpoints can be placed on the ground as well.
32. WHAT CONCLUSIONS CAN WE DRAW ABOUT TESTED VARIABLES?  Can you measure beyond the vertical domain of the GCPs?  YES! (Following these procedures)  Does Flight altitude affect your accuracy?  Yes for Stereo Comp, to a lesser degree for IDPC  Does Flight speed affect your accuracy?  Difficult to determine but minimal at best  Does Shutter speed affect your accuracy?  Difficult to determine but minimal at best  Does Oblique angles affect your accuracy?  Difficult to determine but your accuracy may decrease at low oblique angles  Does Mechanical shutter affect your accuracy?  YES! More than any other variable Can stereo compilation economically fill in features missing or distorted in imagery-derived point clouds? YES! Test projects saved 40%-70% over conventional surveying methods.
33. IS STEREO COMPILATION FROM UAS ACCURATE ENOUGH?  Additional work is required to further unlock the full accuracy potential of this approach however our research has shown that it is suitable for many applications and significantly increases the capability of camera based UAS systems  Based on an averaged 3cm RMSEx and RMSEy tested accuracy the data would meet/exceed the Legacy ASPRS Map Standard of 1990 for 1:120 (1"=10’) for Class 1 planimetric data and 1:76 (1”=6.3’) based on National Mapping Accuracy Standards (NMAS)  Given an averaged vertical accuracy 4cm RMSEz the equivalent FGDC National Standard for Spatial Data Accuracy (NSSDA), at 95% confidence level = 7.84 CM suitable for a vertical minimum contour interval of 0.39 ft.
34. WHAT IS THE PRACTICALITY OF THIS APPROACH?  Stereo compilation can achieve high levels of accuracy on features that IDPC’s cannot  Stereo compilation is a cost effective and can often be completed quicker than conventional survey  Stereo compilation may not be a replacement for LiDAR in all cases but in many it can work just as well with significantly lower cost hardware  Stereo compilation can help keep field personnel out of harms way  There is still untapped potential in camera based UAS systems in mapping and survey