Recognition of a mechanical linkage based on occlusion-robust object tracking

摘要

In this paper, we propose a vision-based technique for recognition of a mechanical linkage. The aim is to realize a robot system, which can autonomously operate an object with a mechanical linkage. First, using stereo vision, the system observes the operated mechanical linkage. It recognizes the position and posture of its parts over a sequence of time frames using a geometric model-based approach. Next, the system estimates the position and direction of the rotation axis from the relative trajectory of these parts. Robustness of the vision system with regard to occlusion is needed because there is a great deal of overlap between the parts of the linkage, depending on its operating state. Moreover, the rotation axis has to be estimated from a sequence of positions and postures in the presence of noise. We present the following two techniques to solve these problems: 1) Occlusion robust object tracking based on prediction. 2) Parameter estimation of the mechanical rotation linkage from the noisy relative trajectories of its parts. Good experimental results were achieved by adapting these methods to some objects pairs with a linkage mechanism.