Error Analysis and System Optimization of Pose Measurement with Monocular Vision Based on Point Feature Model

Abstract

The visual measurement technology is prerequisite to implement high accuracy pose measurement tasks for space manipulator. Firstly, a Newton Downhill-Least Square (ND-LS) algorithm is presented to optimize the object’s pose in this paper. Secondly, the pose measurement error caused by the focal ratio, the principal point, image coordinates and world coordinates is analyzed separately. Finally, the performance of suggested algorithm and the results of pose error analysis are evaluated respectively by re-projection error (RPE) and mean absolute of pose error (MAPE). Experimental results validate that the ND-LS algorithm is superior than Newton method or Least Square (LS) method in pose optimization. In general, solved by ND-LS algorithm, the RPE reduces nearly 3 times contrasted to initial value, and it also reduces to half of that obtained by both Newton method and LS method. On the other hand, the results of pose error analysis indicates that the error influenced by the principal point approximately equals to the one caused by image coordinates, about 2 and 15 times higher than what caused by both focal ratio and world coordinates, respectively, to orientation error with MAPE. Moreover, the relationships above are about 2 and 20 times separately as for the position error.