Research On Self-occlusion Detection And Next Best View Of Vision Target

In the vision fields such as robot navigation, automatic inspection and manipulation, etc, it is very difficult to construct a three dimension description of a scene from a single image. Hence, the vision system always needs to obtain multiple images from several different views. Therefore, the problem is how to determine the best view orientation and position of sensor, so that from this best view we can observe the greatest amount of unknown information which can't observe from the previous view. This problem is the notable next best view problem. Based on the range image of visual target, this paper comprehensively uses the theories and methods about computer vision, image processing, differential geometry, etc, and do studies on how to detect self-occlusion and determine the next best view problem of visual target.Firstly, the concept of range image, the representing methods of range image and surface normal computation methods are elaborated.Secondly, the depth value changing rule among the adjacent pixel in the region which exists self-occlusion is analyzed according to range data. Then, founded on in-depth analysis and studies of the image threshold segmentation technology, through analyzing the relationship between self-occlusion and image threshold segmentation, the self-occlusion detection is realized by using global threshold based on simple image statistic.Thirdly, according to the detected self-occlusion information, a method for determining the next best view using self-occlusion information is proposed. Using surface normal at the self-occlusion points and ray tracing algorithm, this paper determines the next best view orientation and position of the sensor.Finally, the self-occlusion algorithm and the next best view method are verified feasible and effective by the experimental results.