Research On Passenger Flow Counting Method In Rail Transit Scene Based On Deep Learning

In the field of rail transit,passengers counting can monitor the passengers number in different areas in real time,provide security for guiding security measures such as traffic diversion,and also providie an effective basis for ticket clearance.With the rapid development of deep learning,the use of neural networks has become a common method of computer vision.How to design a more lightweight model without greatly reducing the detection effect is a very realistic problem in pedestrian detection based on deep learning;meanwhile,Low-monitoring monitoring angles in rail transit scenes lead to more severe pedestrian occlusion,and more applicable pedestrian tracking algorithms need to be studied.This paper studies the lightweight,high-precision pedestrian detection algorithms required for passenger flow counting and multi-target tracking algorithms for severe occlusion scenarios in rail transit scenarios.The work and results obtained are as follows:(1)Aiming at the problem that the existing target detection algorithms consume a lot of computing resources,design a lightweight convolutional neural network LPDNet based on convolution,separable convolution and Inception structure.Experimental results from actual scenes show that the proposed method can quickly and accurately detect pedestrian targets.(2)Aiming at the problem of serious occlusion of the pedestrians in the monitoring scene in the subway station,design a multi-pedestrian target tracking algorithm.This algorithm builds association features based on the block HSV color histogram feature and target position information.This feature has a lower dimension and can effectively improve the tracking accuracy without adding too much computational complexity.Experimental results from actual scenes show that the method is highly accurate and robust to pedestrian occlusion.(3)Aiming at the problem of wrong pedestrian direction judgment caused by the instability of pedestrian detection and tracking frame,design a pedestrian counting method based on successive multi-frame comparison.Experimental results from actual scenes show that the method has high counting accuracy.At the end of the thesis,the full text is summarized,and prospect the problems worthy of further research in the future.