Robust camera pose estimation combining 2D/3D points and lines tracking

This paper presents a new robust camera pose estimation algorithm based on real-time 3D model tracking. We propose to combine point and line features in order to handle partial occlusion and increase the accuracy. A non linear optimization method is used to estimate the camera pose parameters. Robustness is obtained by integrating a M-estimator into the optimisation process. Furthermore, a crucial condition for pose estimation problem is the consistency of 2D/3D correspondences between image and model features. We propose here to implement a natural point and line robust trackers in order to find corresponding features in the sequence images. Our method has been evaluated on several video sequences. The results show the robustness and the efficiency of our algorithm compared to other tracking approaches.