2018 年 84 巻 7 号 p. 652-657
Robots in dynamic environments are required to acquire distance information autonomously to work as a substitute for humans. One method for achieving this is the stereo vision method, which uses two or more lenses with a separate image sensor. However, it is not always possible to measure distances in real environments using this approach, as the camera cannot track an object that is occluded by an environmental obstacle. This paper presents a high-speed and stable visual tracking control approach for a stereo vision robot that can maintain continuous operation even under such occluded conditions. The proposed approach adopts a particle filter to estimate the position of a moving object in the horizontal plane, and the stereo vision robot is controlled to continue to track the object in the center of the camera image according to the estimation. In evaluation experiments, our approach demonstrated sufficient performance in visual tracking control for a periodic moving object in the absence of occlusions, and hence is almost equivalent to an image-based approach. Furthermore, even when the camera was intermittently occluded, the approach exhibited small tracking errors as long as the duration of the occlusion was less than 7.5% of the moving period of the object. The method is hence sufficiently sophisticated for visual tracking control in actual environments.