Vision-Based Motion Estimation

Abstract

Vision-based motion estimation is the problem of recovering the rotation and translation (i.e., pose) of a moving camera or object from camera images. Algorithms that solve this problem are a key component of many control and robotic applications. Traditionally, the pose estimation problem is solved using image feature points, which must satisfy the rigid motion constraint. We explain how reformulation of the rigid motion constraint leads to different pose estimation algorithms. We discuss the advantages and drawbacks of these methods, benchmark their accuracy and runtime, and provide reference to open-source implementations.