Structured Extraction And Management Of Video Targets Under Multi-camera Panoramic Surveillance

With the rapid development of smart city construction across the country,monitoring equipment is spread all over the city,and there is a huge amount of surveillance video data.Therefore,an intelligent monitoring system is urgently needed to process and analyze massive video data.Video structuring is one of the key technologies for intelligent monitoring.It includes technologies such as target detection,multi-objective association,target structured description,and structured data storage and management.This paper has carried out related research and development work on some key technologies in the panoramic video surveillance system,mainly including the updating of multi-camera panoramic background map,video target structuring and structured data management.A background map update method in panoramic video surveillance is designed,which is divided into two steps: image registration and color correction.Image registration consists of two steps: interactive rough registration based on interaction and refinement based on mutual information.Then based on the existing color correction algorithm of our research group,a color correction method based on local invariant region is designed.Finally,the panoramic video monitoring background map update software was designed and implemented.In order to extract useful information from the video,a method of extracting structured data of video object based on deep learning target detection is designed.Yolo algorithm is used to detect objects in video.Based on the detection results,A multi-featured Kalman filter correlation method based on detection target is designed.The similarity between targets is calculated by blending the target's block color histogram,bounding box size and spatial position.The Kalman filter algorithm is used to predict the moving state of the target,and finally an association matrix is constructed to realize multi-target correlation.After obtaining the target sequence,designed a structured description scheme and a database for storing these structured data.Based on the above-mentioned target structured data extraction method,the video structured calculation module in panoramic video surveillance is designed and implemented.Firstly,the decoding and format conversion of H.264 format video stream are realized by using FFmpeg open source framework;Secondly,the producer and consumer models are used to realize the extraction of target structured data.;and the storage of target structured data is implemented based on HTTP protocol.In order to unify the structured data of the target,Using Django open source framework work in with Nginx and uWSGI,a RESTful design style back-end server is implemented.Each computing node can request resources from the server through URI.In order to facilitate users to manage data,a Web client is designed and implemented.