Research On The Application Of Bus Passenger Flow Statistics Based On Multi-Target Recognition And Tracking

In recent years,the process of urbanization has been accelerating.In order to address the problems of traffic congestion,uneven resource allocation,and road safety caused by the growth of urban population and private car ownership,various provinces and cities have also accelerated the pace of infrastructure construction.As an indispensable component of smart city construction,building a smart public transportation information infrastructure is particularly important.In response to the current problems in bus passenger flow statistics,such as unsatisfactory statistical accuracy and low operational efficiency,this paper proposes a bus passenger flow statistics model based on optimized YOLOv5 s lightweight single stage object detection algorithm combined with Deep SORT multi-target tracking algorithm,which improves the accuracy of passenger flow statistics while ensuring operating speed,and further achieves a balance between model operating speed and accuracy.The model is divided into three modules: passenger target detection,passenger target tracking,and passenger counting.In the passenger target detection module,in terms of data preprocessing,the passenger target detection module uses lable Img to label passengers and generate annotation files suitable for YOLOv5 s training in response to the lack of annotation information in the dataset;To address the issue of insufficient data volume,pedestrian categories were extracted from the COCO2017 dataset and the Pascal VOC2007 dataset to expand the dataset.In terms of model optimization:(1)Use Decoupled Head to improve YOLOv5 s coupled head and improve model convergence speed;(2)Improve the Global Attention Mechanism(GAM)by incorporating an architecture that combines group convolution and Channel Shuffle,and introduce the improved GAM structure into the Backbone and Neck networks of YOLOv5 s to enhance network feature extraction capabilities;(3)Alpha-Io U is introduced as the model regression loss function to improve the model positioning accuracy.Setting up ablation experiments and comparative experiments,after experimental verification,the YOLOv5 s algorithm proposed in this article has improved by 1.1% and 8.5% compared to the original YOLOv5 s algorithm on m AP0.5and m AP.5:95,respectively,and effectively balancing detection accuracy and speed.In the passenger target tracking module,the improved YOLOv5 s target detection model and Deep SORT multi-target tracking model are built to stably and accurately track passengers;A pedestrian tracking experiment was set up,which showed that the total number of passenger ID switches was reduced by 34%,and the tracking performance was improved.In the passenger counting module,design a line counting algorithm to perform bidirectional counting of passengers,achieving dynamic statistics of the total number of people in the carriage.Integrate the three modules,set up passenger boarding and alighting flow statistics experiments,obtain the model’s passenger flow statistics accuracy of 96.8% and 98.1% in boarding and alighting scenarios,and design a bus passenger flow statistics system.The average statistical accuracy of the system reaches 97.5%,and the operating speed can reach30 Hz.The system has high accuracy,fast real-time performance,and strong robustness.