Research On Foreground Object Detection Technology In Intelligent Video Surveillance System

With the rapid development of emerging technologies such as artificial intelligence,big data,cloud computing,and 5G,intelligent video surveillance has become a research hotspot and driven the development of many industries,such as medical,industrial,security,etc.,which has a significant impact on people's production,life and social development.In intelligent video surveillance system,foreground object detection technology is the foundation and key of the system research.The detection results have an important impact on the subsequent target recognition,target classification,target tracking,and have a close relationship with the degree of intelligence of the system.In this paper,we deeply research the foreground object detection technology in intelligent video surveillance system,focus on the foreground object detection in the static scenario,the dynamic scenario and lens shaking scenario,establish temporal differencing,background subtraction,Gaussian Mixture Model,improved ViBe model,feature match and other models,design and implement a simple surveillance system according to the foreground object detection technology.Finally,we carry out the test and algorithm evaluation for the system.The main research work is as follows:(1)In different surveillance scenarios,establish a variety of algorithm models to achieve the foreground object detection and analyze the model according to the experimental results.For the static scenario,establish temporal differencing model,in order to solve the problem of inaccurate detection when the moving foreground object is too slowly or too fast,propose an improved adjacent frame difference method based on gradient coefficient;meanwhile,establish Single Gaussian Background Model and ViBe model based on background subtraction,analyze and compare these models.For the dynamic scenario,establish Gaussian Mixture Model,reduce the effect of dynamic interference in background of the result for foreground object detection;adopt TOM update mechanism in SACON algorithm,establish an improved ViBe model to solve the problem that the background model updates too slowly to produce ghosting.For lens shaking scenario,in order to stabilize the video by extraction of feature corners,affine transformation,and image filling on the video image,then adopt an improved ViBe model to achieve the foreground target detection.(2)For the purpose of foreground object detection,design an intelligent video surveillance system based on B/S architecture.On the server side,mainly achieve the streaming media services,video transmission,video playback,algorithm analysis and other functions;in the client side,mainly achieve display of video real-time and foreground object detection results.(3)Through the functional test of the system,it verifies whether each module can work properly,adopt different algorithm models to test the actual scene of the surveillance video,and observe whether the detection results can meet the requirements.In the evaluation of the algorithm,analyze the performance of the algorithm based on the evaluation index,analyze the running speed of the algorithm according to the average detection time of the model,and analyze the advantages and disadvantages of the algorithm model in the end.Through system testing and algorithm evaluation,the algorithm model established in this paper can realize the foreground object detection in the surveillance system,which can meet the requirements of the system and lay an important foundation for the follow-up research in the intelligent video surveillance system.