Pedestrian tracking and counting based on YOLOv4 and DeepSORT in Subway Stations

Real walking parameters of pedestrians are the basis of research on pedestrian dynamics. The detection and tracking technology provides the possibility of accurate walking parameters. To tackle the problem of lower detection accuracy of small-scale pedestrians, the YOLOv4 algorithm has been improved in two aspects. Then, the improved YOLOv4_I model acts as the detector of DeepSORT algorithm for pedestrian tracking and counting. After the improvement, the YOLOv4_I algorithm performs well in both the public and private datasets. The mAP is 90.61% in the private dataset, raising about 3.42%. Detection speeds reach 28.2 and 18.2 fps for processing images and video, respectively. Based on the YOLOv4_I-DeepSORT algorithm, trajectories and walking parameters of pedestrians can be obtained. Results show the average speed of pedestrians on stairs in subway stations follows the normal distribution, the mean value is 1.27 m/s, and the standard deviation is 0.33 m
²
/s
²
.