Research On Pedestrian Detection And Tracking Algorithm Based On Convolutional Neural Network

In recent years,with the rapid development of artificial intelligence and machine vision technology,pedestrian detection and tracking technology has received widespread attention because of its important academic research and commercial value.Target tracking algorithms based on detection have gradually become a research hotspot.The performance of the detector plays a vital role in the final tracking performance of the algorithm.The detection-based tracking framework is adopted in this paper,pedestrian detection and tracking algorithm based on convolutional neural network is deeply studied.The research content is as follows:In terms of pedestrian detection,to solve the high missed rate and slow detection rate in current pedestrian detection process,the following improvements have been made on the basis of the YOLOv3 algorithm.Firstly,the K-means++ algorithm which has better clustering effect is adopted to cluster out the pedestrain anchor that meets the characteristics of the pedestrian's posture according to the pedestrian data set in a specific and real-time scene;Secondly,from the perspective of improving the real-time performance of the network,the deep separable convolution is introduced.Subsequently,the channels are separated,and then point-by-point convolution is used to expand the channels.The amount of calculation is reduced while keeping the size of the input and output feature maps before and after convolution.Finally,the SENet module is introduced to overcome the high false detection rate caused by the increase of invalid features caused by multi-scale prediction,and the SENet module is embedded in the YOLOv3 network prediction layer to increase the selection and capture of features by the entire network ability.In aspect of pedestrian tracking,the mainstream detection-based tracking strategy is adopted,the improved YOLOv3 algorithm is selected in the detection part,and improved DeepSORT algorithm is adopted in the tracking part.The DeepSORT algorithm uses Mahalanobis distances with different feature dimensions and cosine distances containing deep feature vectors to describe the motion and appearance information of the target.On this basis,a spatial distance model is introduced in this paper,which is combined with Mahalanobis distance and cosine distance to form a new appearance model as the basis for target association.The experimental results show that the proposed algorithm can effectively reduce the false detection rate and the missed detection rate in the algorithm detection process,which also consider the detection performance of the algorithm.In terms of system design,based on actual application requirements,relying on Pycharm and Py QT5 platforms,a pedestrian monitoring system is designed and implemented.The experiments show that the monitoring system has good robustness and detection effect,which can meet actual application requirements.