Visual Odometry

This project report explains the Visual Odometry pipeline for a Stereo Camera setup, which has been implemented in C++ environment using OpenCV libraries. We have used various KITTI dataset image sequences of stereo camera and documented the results. Our results show that the Visual odometry works well given that features are tracked over the straight path or paths which doesn’t have any sharp turns. In case of sharp turns, due to the error accumulation, the drift increases between the actual and estimated trajectory. This can be solved using loop closure which is considered as the back end of SLAM process. The scope of this project is limited to only till Visual odometry implementation i.e., front end of SLAM process.

Experimental Demonstration