Research On Surveillance Video Synopsis Based On Spatio-Temporal Slice

With the rapid development of information technology,the application of video surveillance system has become increasingly popular and gradually realized networking and high-definition,and the amount of surveillance video data has shown a blowout growth.Facing the vast sea of surveillance video big data,efficiently integrating key information and realizing real-time surveillance video synopsis is a hot and cutting-edge research issue of intelligent video processing and a challenging problem to be solved urgently present.Therefore,this paper,supported by the project of the National Natural Science Foundation of China,researches the theory and algorithm of surveillance video synopsis based on the spatio-temporal slice and strives to achieve real-time surveillance video synopsis.The innovative achievements and the main work of this paper are as follows:(1)A method for the segmentation of surveillance video motion segments based on spatio-temporal flow is proposed.Different from the previous full-volume video data analysis methods,the proposed method only sparsely sampled the boundary pixels of the video surveillance to form the video spatio-temporal profiles.On this basis,the moving targets in the spatio-temporal profiles are extracted through background modeling.Then,the target that crosses the surveillance boundary is modeled as spatio-temporal flow,and the spatio-temporal flow model is revised according to the target feature matching.Finally,according to the spatio-temporal flow model,the surveillance video motion segment is segmented.Experimental results show that the F1 value of the proposed method for segmentation of motion segments reaches 88.3%,and it has the highest efficiency compared with the current mainstream algorithms.(2)A method for extracting surveillance video motion segments based on spatio-temporal tunnels is proposed.Aiming at the problem that the spatio-temporal boundary analysis method is hard to detect the target in the surveillance area,a method for extracting motion segment of surveillance video based on the spatio-temporal tunnels is proposed.Firstly,the method extracts the video circular sampling line progressively to form a nested spatio-temporal tunnel to divide the surveillance area into multiple sub-surveillance areas.Secondly,we expand the nested spatio-temporal tunnels to form the spatio-temporal tunnel extension map.Then,we calculate the flow of the spatio-temporal tunnel extension map to construct a spatio-temporal tunnel flow model.Finally,the motion segments of the sub-surveillance area are fused to obtain the motion segments of the whole surveillance video.The experimental results show that the proposed method can accurately detect the moving targets in the surveillance area,and the F₁ value of motion segment extraction is increased by 0.6%compared with the spatio-temporal flow method.(3)A surveillance video synopsis method based on spatio-temporal feature fusion is proposed.Aiming at the target sparseness and random distribution in motion segments,a surveillance video synopsis method based on spatio-temporal feature fusion is proposed.This method first obtains the target trajectory in the horizontal slice,adaptively extracts the horizontal slice and vertical slice of the video and performs feature fusion.Secondly,the target trajectories are clustered,and the trajectories within the class are sorted in order of appearance.Finally,the target is translated along the trajectory to form a synopsis video.The experimental results show that the video synopsis rate and space utilization of the proposed method are relatively high,and it has obvious advantages in computing time.(4)A video synopsis method for cross-camera surveillance video based on spatio-temporal constraints is proposed.Moving targets that cross the boundary of the surveillance area are the most sensitive.The proposed method first detects sensitive targets based on spatio-temporal slices.Secondly,it uses the boundary spatial information of the surveillance area to match the targets.Then it constructs the background of the overall surveillance area.Finally,the sensitive targets are marked to form a synopsis video.Experimental results show that the proposed method can efficiently synopsis the video,which helps users to quickly perceive the complete movement process of the target,and has a better user experience.