Hello,
We have completed a flight with both LiDAR and RGB images. The flight was flown at 50% side lap at 70m AGL with a 20 MPEG camera. The area covered ranges from farmland, to densely vegetated woodlands. As suspected, some of the images have not calibrated as they are only capturing areas of woodland. We have tried a number of different AT settings to try and improve the number of calibrated images. As the image positions are known to be accurate and contain rotation information, the most successful settings are :
A. Positioning/georeferencing:
- Adjustment constraints: photo positioning metadata - Final rigid registration: none B. Main settings: - Key point density: High - Tags or QR Codes extraction: Disabled - Pair selection mode: Exhaustive - Color correction: machine learning - Splats: Disabled C. Estimation policies: - Tie points: Compute - Position: Compute - Rotation: Compute - Pre-calibration stage: Disabled - Focal length: Adjust - Principal point: Adjust - Radial distortion: Adjust - Tangential distortion: Adjust - Aspect ratio: Keep - Skew: Keep
We have also added additional data that all images are vertical views only and the minimum and max viewing distances are 20 & 100m respectively. Unfortunately this still means sections are uncalibrated. We have had more success in other photogrammetry packages (Pix4Dmatic) which calibrates more images. Do you have any other advice for other settings to use, possibly presents which may improve our result?
Regards, Jason.
Overlap is too small especially in woodland. For woodlands 85% overlap may be needed and fly higher(100m) so it covers more area and can get more tie points.Only option which can help in this situation is to use Pair selection mode - sequence if the flight is such. Also reuising and locking calibration from previous 3D flight could help. As last resort using preset "legacy engine"Pix4D may calibrate more but models are less accurate as it is more suitable for orthophotos.
Thanks Oto, I just thought I would post some of our findings in case anyone else has a similar situation. As the LiDAR sensor was dictating the flight parameters in-terms of overlap, we knew the image acquisition would not be optimal. Eventually we found that the most effective way of increasing the number of calibrated images (aside from using the stated AT settings) was to use the image down sampling option. While this does have an effect on the number of tie points generated, our goal here was to produce an orthomosaic. We found very little difference between down sampling between 50% and 25% so decided to run with 50%.
Yes completely forgot about downsampling it can also do the trick as virtually increases search area.
Answer Verified By: Jason Hagon