Abstract:

In order to enhance safety management within urban rail transit stations, a passenger flow detection and early warning model based on the YOLO11-DeepSort algorithm was proposed. First, a pedestrian head dataset was established and put into YOLO11 to train model parameters for extracting passenger flow density. Then, DeepSort was used for dynamic tracking of passengers' walking paths to extract parameters of passenger flow and walking speed. According to the parameters of passenger flow status at station bottlenecks such as transfer passageways, stairs and platforms, FCM algorithm was used to divide the passenger flow status into 6 subcategories, and 4 early warning levels were divided corresponding to the passenger flow status. Finally, taking Guanyinqiao Station of Chongqing Rail Transit as an example, the first 4 days of real-scene station videos were used to identify passenger flow and extract parameters, and the FCM model was trained based on those parameters to get clustering centers and thresholds. By using the model after training, the 5^th day's real-scene video was classified, and the classification results of full-day, peak and off-peak periods were compared. The results show that the platforms in the station have the longest congestion time and the greatest risk, with the peak-hour congestion duration accounting for 70% and the Level 1 early warning duration accounting for 15%. Stairs in the station have the second longest congestion time and risk, with the peak-hour congestion duration accounting for 46% and the Level 1 early warning duration accounting for 10%. Passageways in the station have shorter congestion time and lower risk, with the peak-hour congestion duration accounting for 41% and the Level 1 early warning duration accounting for 5%.

Key words: YOLO11, DeepSort, urban rail transit, passenger flow status, fuzzy c-means (FCM) algorithm, early warning