Research On Path Trajectory Detection Method Based On Multiple Video

At present,video surveillance systems are widely used in unattended areas such as units,communities,and road monitoring.Post-analysis based on surveillance video sequences plays an important role in finding out the causes of accidents.However,the detection of a specific target in a multi-channel video using an artificial method is not only time-consuming and labor-intensive,but also easily affected by human factors.Therefore,the study of path trajectory detection methods for specific targets has certain practical significance for improving the intelligence of video target detection.In order to accurately extract the same moving target from a multi-channel video sequence,this paper optimizes the existing target tracking algorithm and supplements it with the time and location of the target object.A specific target trajectory tracking algorithm based on multi-channel video is studied to achieve the detection of a specific target trajectory.This article uses the pedestrians in the multi-channel video as the research object and takes the campus intersections as the background.It focuses on the detection,identification and tracking of specific targets under the fixed conditions of the camera in order to provide the required data for the follow-up target trajectory detection process.In the low-level visual processing stage,adaptive background Gaussian mixture model is used to achieve the foreground detection of the motion.Target recognition based on Hausdorff distance is used to complete the target recognition,and then an optimized compression tracking algorithm is used to track the target.In the high-level visual processing stage,this paper divides the monitoring area of each intersection to improve the detection accuracy.At the same time,it comprehensively considers the factors such as the time and location of the target at each intersection and the speed and direction characteristics of the target at the moment.Based on this,a predetermined threshold value is set in advance.When two or more intersections detect the target at the same time,some interference can be eliminated according to a preset threshold.According to a specific target,the position of each frame is used to obtain its action trajectory in each video,and the specific target position information in the multiple channels of video collected at different moments is connected to form a complete multi-channel video target action trajectory.In order to verify the effectiveness of the proposed algorithm,the paper collects 100 real trajectory data collected at each intersection of a campus,and verifies the accuracy of the proposed algorithm under simple and complex scenarios,respectively.It is also compared with the performance of the trajectory prediction algorithm based on Kalman filter.Experiments show that the proposed algorithm is a kind of target action trajectory acquisition method with good practical effect and strong robustness.