Research On Passenger Flow Detection And Passenger Navigation Technology In Metro Stations

In recent years,with the gradual expansion of the scale of rail transit,the passenger flow is rising rapidly,the phenomenon of passenger congestion in rail transit stations,seriously affecting people’s quality of life,so it is imperative to improve its service quality and management efficiency through intelligent means.The current rapid development of technologies such as passenger flow detection and passenger flow guidance provide strong support for achieving these goals.Most of the current passenger flow detection algorithms are deployed on cloud platforms,which are prone to excessive network bandwidth pressure,severe waste of computing resources and high transmission latency.How to detect the passenger flow with high accuracy in real time at a low cost,and how to analyze it effectively to realize the dynamic guidance of passenger flow in the station.In response to the above challenges and based on the existing research results,this paper explores and researches the problem of real-time passenger flow detection and guidance in rail transit stations,the main work and innovation points of this paper are as follows.Firstly,in order to solve the problem that it is difficult to detect passenger flow efficiently and flexibly in a rail transit station,we design and implement real-time and reliable passenger flow detection algorithm that can run in edge devices.The Faster RCNN algorithm was found to have superior detection effect by test analysis of existing target detection algorithms.First,the K-means-based anchor frame selection algorithm for optimal scale anchor frame selection is proposed in response to the low detection accuracy of the Faster R-CNN algorithm in rail transit stations.The experimental results show that the optimized algorithm can reach a maximum m AP value of 94.85%,which is 13.29% higher than the original algorithm.Further,considering the huge memory storage and computational overhead required to run the Faster R-CNN algorithm,it cannot be deployed directly on the In the edge device,a pruning-based algorithm for the compression of the passenger flow detection model is proposed to address this problem.The experimental results show that the detection speed of the compressed VGG16 as the base network for the passenger flow detection model exceeds the original detection speed using the Mobile Net is used as a traffic detection model for the underlying network,with m AP values not lower than Mobile Net The detection speed can be increased by up to 51.91% compared to the conventional one.Secondly,after we get the passenger flow detection data,we need to forecast the passenger flow through the historical passenger flow data,and then guide the station’s passenger flow based on the forecast results.Firstly,based on the passenger flow detection data and weather data,passenger flow prediction is carried out,a BP neural network is introduced,and a passenger flow prediction model is designed and implemented,which can obtain the degree of passenger flow congestion in a rail transit station at a certain moment in the future.Secondly,a dynamic information-based D* LiteF algorithm is proposed to guide the passenger flow in the rail transit station so that pedestrians can pass through the rail transit station quickly and achieve the purpose of station diversion.The experimental results show that compared with the D* Lite algorithm,the D* Lite-F algorithm can reduce passengers’ path costs by an average of 33.35%,with the highest reduction being 46.92%,which significantly reduces passengers’ path costs and improves the efficiency of passenger flow guidance.