Research On Video Retrieval Technology In Security Field

With the rapid development of the Internet and the improvement of people’s safety awareness,a large number of monitoring devices have emerged in our daily life.Surveillance video retrieval has gradually become an important way to ensure personal safety and property safety.Surveillance video contains a lot of information,but with the continuous storage of information by surveillance devices,the amount of video data becomes larger and larger.How to retrieve surveillance video quickly and accurately is an important problem in security field.In the aspect of lens segmentation,since surveillance video should be segmtioned into video clips containing target motion,traditional video segmentation methods such as pixel difference method and histogram difference method are not suitable for segmentation of moving clips,while common motion detection methods such as optical flow method have high computational complexity and are sensitive to light.In this paper,inspired by slices in motion detection,motion segment segmentation was realized by extracting pixels of the full video boundary,fitting slice background,extracting frame value features and boundary features combined with time probability model and boundary flow model.Since a small number of pixels were taken in each frame image,the accuracy of video segmentation was improved.In this paper,on the basis of video segmentation,a key frame extraction method of foreground target proportion based on time trajectory model is proposed.On the basis of video segmentation,the foreground target is automatically updated through the background,the target area proportion is calculated and the key frame extraction is achieved by combining the time trajectory model,which improves the accuracy and efficiency of key frame extraction.An object detection algorithm of 3D feature model based on external rectangle was proposed.The detection and segmentation of foreground objects were achieved through morphology and bounding box operation,combined with the elimination mechanism and correction method,and speed of detection was improved.In this paper,security video retrieval is divided into video segmentation,key frame extraction,target detection and similarity measurement.A large number of experiments show that on the basis of the proposed video segmentation method,the key frame extraction algorithm and target detection algorithm have the advantages of low data redundancy and low computational complexity while ensuring the accuracy rate.In terms of similarity measurement,this paper combines the characters of surveillance video with the existing perceptual hash algorithm to show a relatively ideal retrieval effect,which basically meets People’s Daily needs.