Research On Pedestrian Flow Statistics Algorithm Based On Deep Learning

Pedestrian flow statistics cover a variety of computer vision technologies,such as pedestrian detection and pedestrian tracking,and are the research areas that have attracted much attention in artificial intelligence.This technology can obtain basic pedestrian quantity information and multiple pedestrian status information in the video area,which provides a powerful guarantee for public safety and business management.Due to the low accuracy and poor real-time performance of current pedestrian flow statistics algorithms,the results of pedestrian flow statistics are unsatisfactory.Therefore,this paper uses deep learning technology to optimize the process of pedestrian detection and improve the performance of pedestrian flow statistics algorithms.Based on the framework of "detection-tracking-correlation",this paper designs pedestrian flow statistics algorithm based on deep learning.The algorithm uses YOLO v3 convolutional neural network and MOSSE tracking algorithm to achieve accurate and fast pedestrian flow statistics.The main research work of this paper is as follows:In the pedestrian detection section,three commonly used object detection algorithms are introduced,and YOLO v3 convolutional neural network is selected to complete pedestrian detection.Aiming at the overlapping box problem of YOLO v3 convolutional neural network,this paper uses the Soft-NMS algorithm to optimize the output of the network and improve the accu racy of pedestrian detection.The experimental results show that the improved YOLO v3 convolutional neural network has m AP(mean average precision)of 91.4% and detection time19 ms.Its overall performance is better than other detection algorithms,and it is more conducive to improving the accuracy and speed of pedestrian flow statistics.The pedestrian tracking section,it introduces three pedestrian tracking algorithms commonly used in pedestrian flow statistics.In order to improve the speed of the pedestrian flow statistics algorithm,this paper chooses the MOSSE algorithm with faster tracking speed to compl ete the pedestrian tracking.Finally,the pedestrian position information provided by the pedestrian detection section is used to initialize the MOSSE tracker in the Open CV library to implement pedestrian tracking.In the pedestrian flow statistics section,by establishing data associations between pedestrian detection and pedestrian tracking,the occurrence of duplicate and missing statistical phenomena can be avoided.The performance of the pedestrian flow statistics algorithm designed in this paper is te sted in different scenarios.The experimental results show that the average accuracy and average frame rate of the pedestrian flow statistics algorithm based on deep learning are96.20% and 27.6fps,respectively.the algorithm can better overcome the problems of low accuracy and poor real-time performance of the current pedestrian flow statistics algorithm.