Monocular Visual Odometry Approach for Trajectory Estimation of a Moving Object Using Ground Plane Geometry

摘要

Trajectory estimation of a moving object using monocular visual odometry approach is an important research direction in computer vision. The moving object can be a car or a robot or anything which is carrying camera need to be tracked for autonomous navigation. Scale estimation of the translational motion is tedious in monocular visual odometry system. This paper proposes a new trajectory estimation algorithm in which Nister's 5-point algorithm is used as the baseline algorithm. We have adapted the method discussed in [1] for the scale estimation of trajectory with the additional input of camera height. Estimated scale is nothing but the ratio between the known camera height and the estimated camera height. The camera height shall be found by decomposing the Homography matrix. The proposed algorithm is tested with KITTI standard benchmark dataset [2]. Experimental results on KITTI dataset shows significant improvement in the trajectory estimation of moving object.