Context

CARS enables 3D reconstruction using photogrammetry. The tool is integrated into the ground segment of the CO3D mission. This study aims to improve the denoising of point clouds produced by CARS for 3D scene reconstruction. This is a follow-up of a 2022 study. Three areas for improvement were investigated, one of which was integrated into the PANDORA tool.

CS Group responsabilities for Point cloud clearing in epipolar geometry using trusted intervals are as follows:

Research and software development

Main Picture

The features are as follows:

The goal is to remove noise while preserving important details, to improve the DSM output.
Three improvements were tested: 1 - Optimizing the cost volume by SGM regularization, 2 - Bilateral filtering at the disparity map level, and 3 - Selecting secondary peaks from the cost profile when calculating the disparity map.
Method 2 yielded good results and has been integrated into PANDORA. The other methods remain to be further developed.

Project implementation

The project objectives are as follows:

Improving CARS point cloud denoising methods
Improving CARS output MNS
3D scene reconstruction by photogrammetry

The processes for carrying out the project are:

Agile

Technical characteristics

The solution key points are as follows:

Gitlab

The main technologies used in this project are:

Domain	Technology(ies)
Operating System(s)	Linux
Programming language(s)	Python
Production software (IDE, DEVOPS etc.)	Git, CARS, PANDORA