Research On Self-Occlusion Detection Approaches Of Vision Target Based On Depth Image Information

In the vision research fields such as target recognition, stereo matching, visual tracking and scene reconstruction, etc, self-occlusion phenomenon has become one of main influence factors for the executive effect of various tasks. How to find an accurate and reasonable approach to test the self-occlusion phenomenon of vision target is the pivotal and necessary link to solve all kinds of self-occlusion problems. Based on visual target depth image information, this paper comprehensivly uses the theories and methods about computer vision, image processing, differential geometry, etc, and do studies on the self-occlusion phenomenon detection methods of visual target.Firstly, the concept of depth image, the methods of obtaining depth image, and several common principle methods of depth image pretreatment techniques are elaborated.Secondly, starting with the analysis of relationship between self-occlusion detection and the two aspects which are the depth image information and the surface mean curvature change, this paper gives in-depth analysis and discussion about how to detect the self-occlusion phenomenon of vision target. An algorithm for detecting self-occlusion is proposed based on mean curvature and depth image information in this paper. By analyzing the change feature of mean curvature value in depth image and combining with two thresholds judgment, the vision target self-occlusion detection is realized.Thirdly, founded on in-depth analysis and studies of the image threshold segmentation technology, a self-occlusion detection algorithm based on optimal threshold segmentation iteration and depth image information is proposed in this paper. The optimal threshold segmentation iteration method of gray image processing field is applied to depth image processing field. The self-occlusion detection is realized by using the depth difference image information of vision target and obtaining proper threshold.Finally, the feasibility and effectiveness of the two proposed methods are verified through experiments. Combining with the analysis of algorithms and the experimental results, two algorithms are analyzed and compared.