Robust ∞ convex optimisation for monocular visual odometry trajectory estimation

The most important applications of many computer vision systems are based on robust features extraction, matching and tracking. Due to their extraction techniques, image features locations accuracy is heavily dependent on the variation in intensity within their neighbourhoods, from which their uncertainties are estimated. In the present work, a robust L∞optimisation solution for monocular motion estimation systems has been presented. The uncertainty estimation techniques based on SIFT derivative approach and its propagation through the eight-point algorithm, singular value decomposition SVD and the triangulation algorithm have proved an improvement to the global motion estimation. Using monocular systems makes the motion estimation challenging due to the absolute scale ambiguity caused by projective effects. For this, we propose robust tools to estimate both the trajectory of a moving object and the unknown absolute scale ratio between consecutive image pairs. Experimental evaluations showed that robust convex optimisation with the L∞norm under uncertain data and the Robust Least Squares clearly outperform classical methods based on Least Squares and Levenberg-Marquardt algorithms.