Research On Dynamic Occlusion Avoidance Approach Based On The Occlusion Region Model And Object Motion Estimation

Occlusion phenomenon can be seen everywhere in human's daily life.The occurrence of occlusion phenomenon can lead to visual task errors or fail.With rapid development of artificial intelligence,computer vision,automation and other related fields,How to avoid the occlusion of move object has received extensive attention from experts and scholars.Therefore,Dynamic occlusion avoidance has gradually become a challenging problem in the visual field such as target tracking,3D reconstruction,object recognition,scene rendering,motion estimation,etc.Considering that the depth image acquired by the Kinect depth image acquisition tool can quickly and accurately acquire the three-dimensional information of the visual object.This paper proposes a novel method based on motion visual object depth image to use the occlusion information to achieve dynamic occlusion evasion.Firstly,An approach to quickly construct the optimal observation azimuth model of the occlusion region is proposed.In this approach,the key points on the boundary of the depth image in this paper are first used to extract the key points on the occlusion boundary of the visual object depth image.Then use it to build key line segments to quickly model the occlusion region.Finally,the optimal observation orientation model of the occlusion area is constructed by the key line segment and the occlusion region modeling result.Secondly,motion estimation is performed on motion visual object using image matching algorithm.In order to improve the accuracy of visual object motion estimation,a hybrid curvature feature was been proposed.The quantity of feature points extracted in the process of image matching is increased by calculating the mixed curvature matrix corresponding with depth image,which is conducive to estimating motion of visual object accuratelyThirdly,The Newton's method is used to calculate the next best view position of the camera based on the optimal observation azimuth model of the occlusion region and the motion estimation equation of the visual object.And then the rational planning of the camera observation orientation makes dynamic avoidance of the visual target occlusion region.Finally,Simulation experiments and real experiments are carried out respectively to demonstrate the feasibility and effectiveness of the proposed approach.At the same time,the experimental results were analyzed and summarized.