Transcript Document
VIIth International Scientific and Technical Conference From Imagery to Map: Digital Photogrammetric Technologies Bundle Adjustment Method in PHOTOMOD® 4.3 Andrej Yu. Sechin, Scientific director, Racurs September 17-20, 2007, Nessebar, Bulgaria Selecting a method of adjustment Tests on “synthetic” projects Over 100 parameters Tested on blocks up to 3000 images Tests in Racurs Production Department More than 50 real blocks adjusted. Scales 1:10000-1:40000. From 20 to 1000 images per project New parameter: Accuracy Block of 2000 images XYZ deviations ~ 0.40m Comparing different programs Two synthetic test blocks 1000 images, 30000 points, 40 GCPs, H=2000m, f=150mm,Scale – 1:13000 Deviations – 10 mkm (image), 20 cm(XYZ) 2000 images, 60000 points, 75 GCPs, H=2000m, f=150mm,Scale – 1:13000 Deviations – 10 mkm (image), 40 cm(XYZ) Computers used PHOTOMOD/Aerosys Photocom(Antipov) Orima Celeron 2.8 Ghz Pentium 4 3 Ghz Core2Duo E6400 Comparing different programs I. 1000 Images synthetic block (40 GCPs) Method PHOTOMOD(Models) PHOTOMOD(Bundles) Photocom(Antipov) Aerosys Orima(*) CPU Time 0’18’’ 2’43’’ 4’4’’ >4h 1’14’’ ErrX 9.1 7.0 6.6 7.4 6.6 ErrY 9.8 8.6 8.7 10.0 8.2 ErrZ 24.3 16.7 16.3 53.3 16.1 II. 2000 Images synthetic block (75 GCPs) Method PHOTOMOD(Models) PHOTOMOD(Bundles) Photocom(Antipov) Orima(*) CPU Time 0’48’’ 10’23’’ 20’12’’ 4’54’’ ErrX 10.6 8.3 7.4 7.2 ErrY 11.7 8.6 7.5 7.0 ErrZ 28.7 19.9 19.2 18.6 When bundles is to be used IMU/GPS Data When bundles is to be used “Big” water areas on the images Example project: 1:12000 image scale Bundles: GCPs RMS residual 17 cm – XY 16 cm – Z Models: RMS residual 2.6 times as big General Considerations/Experience Independent models vs. Bundle Adjustment • For regular projects with no blunders the accuracy is similar • Each method allows to detect specific blunders • Independent models is better for controlling GCPs • For final adjustment it is better to use “bundles” Usage scheme PHOTOMOD Solver Models/detecting tie blunders Bundles/detecting tie blunders Models/detecting GCP blunders Bundles (increased accuracy) final adjustment Future Developments •Self calibration •Robust blunder detection •Speedup, using sub-blocks for adjustment •Various accuracy estimations Thank You for Attention More Information at the Master Class Everybody is Invited