Research On Occlusion Boundary Detection Approaches Of Depth Image Based On Neural Network

The occlusion phenomenon is everywhere in real life.The existence of occlusion will seriously affect the tasks in most computer vision areas,such as image classification,object tracking,3D reconstruction,content understanding,etc.Therefore,occlusion detection has become an urgent problem to be solved in the field of computer vision.On the problem of occlusion,the depth image of the visual object is taken as the research object.According to the existing occlusion boundary detection approaches of depth image,the unsupervised learning and the deep learning theory are used to study the visual object's occlusion boundary detection approaches of the depth image.First,the depth image,occlusion phenomenon,occlusion boundary,the occlusion boundary of deep image and some existing occlusion related features are described.On this basis,the basic knowledge of Self-Organizing Maps and convolutional neural network are introduced.Secondly,an approach for occlusion boundary detection of deep image is proposed based on Self-Organizing Maps is proposed.First,a new feature named the eight neighborhood is proposed by analyzing the existing occlusion related features of depth image,and the approach of feature calculation is given.Secondly,the unsupervised SOM network is modified based on the proposed feature combined with the supervised learning.Thirdly,semi supervised SOM network is trained by labeled samples,and let it achieve two classification.Lastly,every pixel of the depth image is divided into occlusion boundary point or non occlusion boundary point according to that network,and finally the occlusion boundary detection of depth image is determined.Thirdly,an approach for occlusion boundary detection of deep image is proposed based on convolutional neural network.The structure of the existing convolutional neural network is modified with the feature fusion idea,and the training data set is used to train the modified convolutional neural network.A well trained convolutional neural network is used to realize the end-to-end occlusion boundary detection of depth image.Finally,the two proposed approaches for occlusion boundary detection of depth image are experimentally verified,and compared with the existing occlusion detection approaches of depth image,proving the effectiveness and applicability of the proposed approaches.