Some trees do not show up in orchid DSM

I recently photoed a lychee orchard and tried to make a 3D map. However, some trees are not shown as high points in the DSM, although orthomosaic images look completely normal. Does anyone know what factors could cause this issue?

Below are the summary of the process, and screenshot of DSM and orthomosaic.

Thank you very much.

Hi Chengyuan Xu, 

There is a red warning sign next to the camera optimization in the quality report, that indicates a 24.86% difference in the camera parameters: 

Based on the EXIF ID “FC330_3.6_4000x3000” you used the Phantom 4’s camera. The latter is a well known model in the Pix4D camera database, so large differences in the optimization are suspicious and generally mean that the software ended up in the wrong optimum for the camera parameters. 

The initial camera parameters can be trusted, as it is a well known camera model from the database. Using the “All Prior” setting in the processing options tells the software to reduce the scope of the optimization close to these initial parameters: 

Also, this camera has a linear rolling shutter. It could be that there is a rolling shutter effect influencing the results. This can be taken into account in the processing, see this article: https://support.pix4d.com/hc/en-us/articles/208460436 

Let us know how this went :slight_smile:

Hi Pieranglo:

I used applied rolling shuttle adjustment and the parameter calibration is improved. However, it does not resolve the DSM issue. DSM still exists.

When checking the Point Clouds, there appear to be many ‘black holes’ where the trees are not reconstructed

Those black holes are filled as ‘green ground cover’ in triangle mesh.

Plus there is no tie points over the tree rows.

I also tried All Prior in calibration with rolling shuttle adjustment. The result is similar and the image is even a bit distorted, so it is not likely to solve the problem.

The camera calibration issue actually happens a lot when images are acquired with DJI Ground Station Pro App (large error in calibration and often distorted orthomosaic images). The situation is far better when using third party mapping App such as DroneDeploy. There must be some issues with the camera setting DJO GSP.

Thank you very much.

Stephen

I don’t know for what purpose you would like to use the DSM, but if the goal is to have all trees reconstructed, I would try one of these: 

  • Change the Ground Sampling Distance (GSD) from about 1cm to 10cm. This should improve the reconstruction of the trees, because the geometry will be easier so more matches can be found between images. 
  • Take more overlap in the images. You could try to use Pix4Dcapture to have a good control on the overlap settings. This should also improve the matches found in the images. 

I’d be interested to see if that improved the results :)