Fast Algorithms of Multi-Object Recognition and High Precision Localization for Pose Estimation

Abstract

The recognition and localization of cooperative objects are very important for spacecraft pose estimation towards autonomous rendezvous and docking (RVD). In this paper, an adaptive threshold segmentation algorithm is proposed base on weighted maximum gray value for multi-object detection, and eight-neighborhood region growing is employed for multi-object recognition. In order to achieve high-accurate localization, a sub-pixel object positioning approach is put forward by combination bilinear interpolation with median filtering, which employs bilinear interpolation to calculate sub-pixel centroid for reducing algorithm systematic errors, and applies median filter to reduce random errors produced by image noises. The experimental results show that the proposed algorithms are feasible and effective with high positioning accuracy of 0.01 pixels, and have outstanding advantages of anti-disturbance and real-time performance, thus can satisfy the practical requirements in the visual measurement and pose estimation of cooperative objects for the RVD in space exploration.