Occlusion Boundary Detection For Depth Image Based On Learning Methods

Occlusion phenomenon is everywhere,and a series of occlusion problem has brought lots of inconvenience in real life such as aspects of scene reconstruction,object recognition,target tracking,stereo matching and visual measurement.In recent years,the occlusion problem gradually separated from the visual tasks which are getting more attention from the researchers.How to apply the theory and knowledge to find out more accurate occlusion information is the key study object to solve occlusion problem.According to the research status at home and abroad,occlusion boundary detection method of visual target is studied by using mathematics,machine learning and depth and spatial information.Firstly,introducing the knowledge of the depth image and explaining the reason why occlusion happened,and then briefly described the methods of classification and normalization.Secondly,an occlusion boundary detection approach is proposed for depth image based on unsupervised clustering。A feature named weighted longest line segment and its computing method are proposed based on the spatial and depth information according to visual target of depth image.A novel nonlinear normalization method is proposed to normalization the occlusion related features according to the feature distribution of occlusion boundary points and their neighbor points.Each pixel’s combined feature vector,which is composed of occlusion related features,is inputted into the unsupervised clustering classifier to judge whether the pixel is occlusion boundary point.Thirdly,a new occlusion boundary detection approach is proposed for depth image based on supervised and K-means segmentation.A series of labeled samples are created according to effective center value transform and K-means segmentation,then inputting the labeled samples to detection classifier training classifier,and detecting the occlusion boundary by classifier.Finally,the feasibility and effectiveness of our method are reported with the proposed occlusion boundary detection.Comparison the experimental results to analyze them.