Research On Pedestrian Flow Counting Algorithm Based On CNN

With the strengthening of security awareness in all walks of life,the demand for the installation of surveillance cameras is increasing,which leads to a sharp increase in surveillance video data.Video data cannot be processed in a timely and efficient manner when relying on labor to count the number of people.Pedestrian flow counting based on computer vision is one of the key technologies of intelligent video surveillance system,which can effectively reduce costs and improve the efficiency of video processing.However,there is still exists some problems such as pedestrian missed detection,false detection and repeated tracking due to the complexity of the external environment and the uncertainty of pedestrian movement,which will directly affect final statistical results of pedestrian flow.In view of the above problems,the main work of this thesis include:1.Aiming at the problem that the detection accuracy is low due to the pedestrian occlusion in a complex environment,considering that there is a one-to-one correspondence between the head and the person,and the head is not easily occluded during pedestrian movement,a pedestrian head detection algorithm based on clustering and Faster RCNN is proposed.All labeled detection boxes are clustered by using the newly designed distance measurement method and k-means++ algorithm,to determine the anchor size and aspect ratio.The penalty function of the NMS algorithm is optimized to remove invalid prediction boxes,which can alleviate the problem of low recall due to pedestrian occlusion.The simulation experiments show that the algorithm can effectively improve the detection accuracy,and the highest AP on the two datasets reached 90.2% and 87.7% respectively.2.Aiming at the problem of identity switch after pedestrian disappears and reappears during multi-target tracking,a pedestrian tracking algorithm based on the IOU and color similarity is proposed.Based on the IOU matching algorithm,color similarity is introduced and a temporary queue is created for disappearing pedestrians,and the appearance similarity and location information between objects are integrated to complete pedestrian rematching.The experimental results show that the proposed algorithm reduce the problem of identity switch due to transient disappearance of pedestrians,and effectively improves the ability of multi-target tracking.3.Aiming at the problem that the pedestrian flow counting algorithm has a low accuracy rate in bidirectional statistics,a bidirectional crossing-line pedestrian flow counting algorithm based on tracking trajectory is proposed.Preset a line of interesting counting based on the application scenario.The direction and position of each trajectory are analyzed on the basis of pedestrian tracking results.Then the vector cross product is introduced to evaluate whether the pedestrian passes the preset line of interest,so as to count the number of people in different movement directions.The experiments show that the algorithm can improve the accuracy of bidirectional pedestrian counting to some extent.