Sparse Stereo Visual Odometry with Local Non-Linear Least-Squares Optimization for Navigation of Autonomous Vehicles

In this thesis, the author presents a Sparse Stereo Visual Odometry system for navigation of autonomous vehicles. The proposed system has the capability to estimate the camera's pose based on its surrounding environment. In contrast to other Visual Odometry systems with Bundle Adjustment optimization, the system proposed in here differs in four main aspects: (1) it utilizes both stereo frames to track features between frames; (2) it does not require a bootstrap step to initialize the algorithm; (3) it performs a local optimization at every increment frame instead of perform a windowed optimization; and (4) it consider the both stereo images inside the optimization instead of just one side of the stereo system. The system was tested on the Karlsruhe Institute of Technology (KITTI) Vision Benchmark Suit, as well as with a set of video sequences recorded with commercial stereo cameras on the roads of the city of Ottawa, Ontario.