Visual Odometry Analysis on the ISRO’s Chandrayaan-3 Pragyan Rover Images

Visual odometry proves critical for accurate motion estimation in planetary exploration, mainly in terrains where wheel odometry fails due to slip. This paper discusses the application of visual odometry using stereo images from ISRO’s Chandrayaan-3 Pragyan rover to improve rover localization and navigation. This approach uses the SIFT algorithm for feature extraction and matching, combined with stereo geometry and pinhole models for 3D reconstruction. Camera calibration was performed to reconstruct the 3D features properly. The results from the camera model match well with the ground truth. However, the accuracy decreased with increased distance from the camera, which points to the importance of precise calibration. Maximum Likelihood Estimation (MLE) was incorporated to further tune the results. In general, this paper demonstrates the effectiveness of visual odometry in planetary exploration, offering ways to improve rover navigation in complex environments. Future works on this will focus on fusing VO data with lidar, chassis kinematics, and wheel odometry to further enhance robustness in motion estimation.