Abstract
For environment recognition with the camera, it is preferable to reduce the number of images to be processed in order to reduce the processing time. In complicated environments, it is difficult to locate objects and measure their correct positions. We propose an active vision system to solve these issues. An initial position of the camera is known. The camera is installed in a location which an occlusion problem doesn't generate. When the external environment is static, the parallel stereo camera with few amounts of movement is performed from an initial position, and the rough coordinates of each objects are calculated. A model of object trajectories is designed by considering the position and the state of the camera. We adjust the next position of the camera in accordance with this model; thus, we can avoid the problem of occlusion of the objects by calculating the coordinates of the next position of the camera accurately. Finally the accurate coordinates of the objects are calculated by considering the state of the camera and the obtained from the images information. The efficiency of the proposed active vision system is verified by simulations and experiments.
