The Research On Occlusion Boundary Detection And Next Best View Approaches In Single Depth Image

Occlusion phenomenon can be seen everywhere in our daily life,Therefore,Makes the occlusion problem become an important and difficult problem in the field of computer vision research.In robot navigation,automatic assembly,automatic drive and other research fields are affected by the occlusion problem.The occlusion phenomenon bring great disturbance in the research work.So,it became an urgently the problem needs to solve in the field of computer vision research.in order to solve the occlusion problems,this paper using the depth information of the visual object,aiming at the shortcoming of existing problem about the research of occlusion problem at home and abroad,making full use of related theoretical knowledge such as machine learning,mathematical modeling.This paper deeply researched the occlusion detection based on depth image and the way of determine the next best view of the camera.Firstly,introduces the related basic concepts of depth image and the way to collection it,the generation and classification of occlusion phenomenon,the definition and classification of the next best view,this paper also introduces the relevant theoretical knowledge of the random forests and the algorithm of FCM clustering in machine learning.Secondly,through analysis the pixel's depth value in depth image,and the deeply research of the space position relations in occlusion part of the visual object in 3D space when it occurs occlusion phenomenon.The average angle feature is proposed,and the extraction method is also provided.On the basis,an occlusion detection method based on the unsupervised FCM clustering algorithm is proposed,to achieve the detection the occlusion boundaries of the visual object in depth image.Thirdly,in the basis of the occlusion detection,then the external surface of occlusion region is constructed and modeled.The observation direction,observation center point and area information for the external surface of the occlusion region are solved.The next best view is detemined by obtained information to achieve the occlusion avoidance.Finally,the methods of detection occlusion and determine the next best view are experimented,the experimental results are analyzed and compared with other methods.The results demonstrate the approach is feasible and effective.