Estimating and Correcting Bias in Stereo Visual Odometry

Stereo visual odometry (VO) is a common technique for estimating a camera's motion; features are tracked across frames and the pose change is subsequently inferred. This method can play a particularly important role in environments where the global positioning system (GPS) is not available (e.g., Mars rovers). Recently, some authors have noticed a bias in VO position estimates that grows with distance travelled; this can cause the resulting estimate to become highly inaccurate. In this thesis, two effects are identified at play in stereo VO bias: first, the inherent bias in the maximum-likelihood estimation framework, and second, the disparity threshold used to discard far-away and erroneous observations. To estimate the bias, the sigma-point method (with modification) combined with the concept of bootstrap bias estimation is proposed. This novel method achieves similar accuracy to Monte Carlo experiments, but at a fraction of the computational cost. The approach is validated through simulations.