Open fengyunxuan opened 3 years ago
the accuracy of the reconstruction depends a lot on the input image resolution per distance to the surface ratio; the higher this ratio the better the reconstruction should be normally. In this case flying at 155m above the surface sounds like a lot, the image resolution even if enough (only if you process at the native resolution, ex --resolution-level 0
) you start to encounter other issue, like noise in the image (by scaling down the image the noise is reduced too), noise in the estimated poses (which grows exponentially with the distance to the scene), and not lastly the normal to the surface to be reconstructed gets very oblique, which poses a problem when computing any photometric score.
@cdcseacave Hi, Thank you very much for providing the impressive open source code---OpenMVS. I benefited a lot from it, it's fun!But I used a group of drone photos to do 3D reconstruction, and found a problem that the facade of the reconstruction building was distorted and deformed. As shown below:
Data: DJ Phantom 4 RTK altitude: about 155m
textured 3d reconstruction model![1](https://user-images.githubusercontent.com/19601392/122152808-65015580-ce94-11eb-87b6-2ffca53140d4.png)
densifyPointCloud model(note: I use the PatchMatch method)![2](https://user-images.githubusercontent.com/19601392/122152900-895d3200-ce94-11eb-9e01-2f60fb8984c9.png)
Can you help me analyze the possible causes of this problem? I will be very grateful to you!