Research On Road Target Detection And Tracking Algorithm Based On Video Processing

In recent years,with the comprehensive construction of smart cities,people's demand for social security is increasing day by day,which undoubtedly increases the burden on security personnel.Therefore,intelligent video surveillance systems with automatic analysis and processing of image data are important development directions for monitoring technology in the future.Different from the traditional human-management monitoring system,computer vision algorithm plays a key role in intelligent monitoring technology.Because of the complex and changeable monitoring environment,the shape change and occlusion of the target and other factors,fast and accurate detection and tracking of the target is still a challenging research hotspot.This paper automatically analyzes the vehicles and pedestrians in the monitoring scene through computer vision algorithms,it can automatically detect the intrusion problem of the warning area and issue alarms in time,which greatly improves the monitoring efficiency and has high application value.The main research work of this paper is as follows:(1)Due to the traditional detection algorithm such as background difference method,the optical flow method has great limitations and the effect is not satisfactory.This paper proposes a vehicle and pedestrian detection algorithm based on SSD convolutional network.Firstly,the framework and basic principles of SSD network are introduced,and the improvement ideas of further improving network detection performance and strengthening the detection ability of small targets are summarized.The algorithm modifies the aspect ratio of the a priori frame for the common targets in the surveillance video,and the low-level convolution layer is used to increase the position information of the feature map to further improve the network performance.In addition,this paper proposes a data enhancement method based on image pyramid,which improves the detection accuracy by increasing the number of small targets in the data set.(2)In order to continuously track the detected multiple targets,this paper designs a vehicle and pedestrian tracking algorithm combined with Camshift and Kalman filtering.Firstly,the position predicted by the Kalman filter is taken as the initial position of the Camshift algorithm iteration,and then the accurate tracking position is obtained by the Camshift algorithm.In addition,the possible occlusion situation is improved and the stability of the algorithm is improved.Secondly,a distance-based detection and tracking data association algorithm is proposed to match the same target in the two frames before and after,and the tracking chain is formed.Finally,the intrusion detection of the specified warning area is realized by the target trajectory analysis.The video test results show that this paper uses The tracking algorithm has a good tracking effect.(3)In the VS2013 development environment,the Caffe deep learning framework and the OpenCV visual library are used to complete the algorithm programming,and the overall performance of the algorithm is tested experimentally.The test results show that the optimized SSD detection algorithm has a mAP of 77.5 on the PASCAL VOC dataset,and the average tracking FPS is 27.3 in the test scenario.The average recognition rate is 93.3%,and the average miss detection rate is 6.7%.The false detection rate is 4.9%,and the performance indicators have reached the expected level,which basically meets the real-time monitoring needs.