Research On Long Time Tracking Of Pedestrians In Intelligent Surveillance

In recent years,with the rapid advancement of science and technology and the continuous improvement of the people's living standards,people's demand for public safety precautions has also continued to increase.After the topics such as "Smart City" and "Safe City" were put forward in China,video surveillance equipment was gradually spread over every corner of public places.The video surveillance system became the core part of the urban social public safety monitoring platform.Target detection,tracking,and recognition based on computer vision are the key technologies of intelligent surveillance systems,and they are also hot topics in academic and engineering research at home and abroad.Pedestrian detection and tracking technology is one of the key issues in the field of intelligent monitoring.It detects and tracks pedestrians in the field of vision,and obtains behavioral trajectory information.It is the precondition for automatic analysis and understanding of the behavior of monitoring targets.In the actual application scenario,the current tracking technology still has many challenges.Changes in light intensity,rapid movement of the target,severe deformation,and occurrence of occlusion may easily lead to the loss of the target.To track the trajectory of the tracking target more completely,higher requirements are placed on the robustness and real-time performance of the tracking system.In this paper,aiming at the long-term tracking problem of specific pedestrian targets,a mechanism based on re-detection and re-recognition after loss of target tracking is proposed to implement a long-term tracking method for pedestrians with high robustness and real-time performance.The main work of this article includes the following aspects:(1)Research the pedestrian tracking method based on computer vision.In order to ensure the tracking accuracy and real-time at the same time,an end-to-end off-line deep learning model is used for the online tracking scheme and the algorithm is improved.Through experimental tests,the improved tracking algorithm in this paper shows better tracking ability in both accuracy and real-time.(2)The method of pedestrian detection and re-recognition based on computer vision is studied.A lightweight end-to-end convolutional neural network is designed based on the framework of fast-RCNN to search the target in the full visual field and achieve rapidity.Specific pedestrian target detection and re-identification.In this paper,the detection and re-identification of pedestrian targets are not separated from two separate tasks,but they are collectively handled in an end-to-end network.Experimental results show that the network has high accuracy and robustness.(3)Research on the method of long-term tracking of pedestrian targets.By designing the short-time target tracking algorithm combined with the algorithm of pedestrian re-detection and re-identification after tracking loss of target,a long-term tracking of the target with high robustness is proposed.Methods.The method can solve the problem of tracking loss caused by complex scenes such as deformation,rapid movement,severe occlusion,and out of view in the tracking process of the pedestrian target.After the tracking target is lost,the pedestrian re-detector is started to re-detect the target and initialize The tracker continuously tracks pedestrians' goals and records their behavior.The experimental results show that the long-term tracking method in this paper can record the behavior trajectory of the pedestrian target in the surveillance field to a certain extent.