Research On The Method And Application Of Foreign Objects Occlusion Detection In Surveillance Video

With the wide application of safe city intelligent monitoring system,the number of surveillance cameras is booming.Thus,there is a more urgent need for solution to whether a camera is in regular work,and whether a monitoring area is effective.However,cameras are easily sheltered by the foreign matter,due to the complex and varying environment of camera monitoring region.To deal with the problem,this paper studies utility pole detection and leaves detection in security monitoring videos.The paper presents from two aspects as follows:For utility pole sheltering in monitor video,an improved LSD straight line detection algorithm is proposed combining with several common straight line detection methods.The proposed algorithm exacts straight lines first,for the pole edge looking like straight line.Next,for every rectangle region of two straight lines,competing standard deviations of H and S components in the HSV color space.At the meanwhile,dividing the region into two up and down symmetric regions,and computing structure similarity of the two regions.Finally,the utility pole detection is achieved by standard deviations and structure similarity.To detect sheltered leaves in monitor video,a convolutional neural network(CNN)based detection algorithm is proposed,aiming at automatically determining whether the video suffers from leaf occlusion.First,the distribution area of the leaf in the image is gotten by using the convolution neural network.Then,the depth information in the leaf area is also computed by CNN,with leaf area being detected at the first step.Finally,the severity of leaves block can be obtained by combining the distribution area information and the depth estimation information of the leaf,which are merging into a CNN framework.At last,the effectiveness of the proposed approaches is verified using experiments.And in particular,it proves the proposed approaches satisfy the needs of the utility pole detection and the leaf occlusion detection in real-world traffic surveillance video,and the feasibility and application value of the proposed methods.