Design And Implementation Of The Pedestrian Video Synopsis System Based On Machine Vision

Most of the current substation base stations adopt unattended mode.In the substation monitoring center,the monitoring of the target is completely based on manual staring.It is inevitable that there will be omissions,and it takes a lot of time to browse the video search problem after an accident.The angle of the substation surveillance video is fixed,and we are more concerned about the activities of pedestrians.In the substation monitoring video scene,the moving targets are very few,the monitoring video angle is fixed,the repeated data is more,and the valuable information data is relatively less.Considering the scene characteristics of the substation,video summary technology can be used to compress the original video length in time and space dimensions,remove non-pedestrian information data,reduce video length and storage size,and retain the basic content and important information in the original video.Improve the efficiency of monitoring managers.The system input is a piece of original surveillance video.Firstly,the original video is preprocessed,the background difference method is used for motion detection,the mixed Gaussian model is established,the moving target is detected from the original image,and the deep learning algorithm is applied to the moving target,using the convolutional nerve.The network model distinguishes pedestrians from moving targets,saves the pedestrian's motion information,and then trajectories the trajectory of each pedestrian.The trajectory image of the pedestrian is saved according to the tracking result,and the pedestrian trajectory segment is generated.Finally,the simulated anneal principle is used to optimize each trajectory.The combination of ways to reduce occlusion and collision between pedestrians,generate the final summary video and save.The system realizes the detection of moving targets,detects pedestrians in moving targets and records the trajectories of pedestrians.Finally,the trajectories are recombined in time and space to generate a summary video of pedestrians.The pedestrian learning method based on deep learning can accurately detect pedestrians in the video,and the accuracy rate meets the requirements of the project.The generated summary video retains the pedestrian motion of interest,and the length of the video is greatly reduced.The method in this paper is suitable for scenes with similar moving targets such as substation,which are not applicable to scenes with more moving targets,longer target periods or more complicated scenes.