Binocular Vision Based Pose Measurement Using an Improved ICP Algorithm

Abstract

Aiming at the problem of pose measurement of non-cooperative targets in space, an improved iterative closest point algorithm(ICP) is proposed in this paper. The traditional ICP pose measurement algorithm optimizes the point cloud registration by iterating the nearest point continuously. The improved ICP algorithm considers the Euclidean distance between two points and the angle between the normal vectors as the criterion for feature points screening. The improved ICP algorithm filters out some points with large distance and angle to reduces the measurement error. Binocular images are simulated by Blender software and point cloud data are recovered by stereo correction and stereo matching. The experimental results of the traditional ICP algorithm and the improved algorithm show that the angle error of the traditional pose measurement algorithm is about 0.5° and the distance error is 0.08m, while the angle error of the improved algorithm is 0.2° and the distance error is 0.06m. The improved ICP pose measurement algorithm can get better pose measurement accuracy than the traditional ICP algorithm.