Understanding the characteristics of pedestrian flows in large-scale railway stations is demanding. Not only for designing new stations or renovating existing ones, but also for applying new technologies such as autonomous robots for goods transportation, it is essential to conduct appropriate evaluations based on a fine measurement of pedestrian flow. However, there are only a limited number of cases of large-scale measurements in public spaces, and quantitative verification of tracking accuracy targeting high-density crowds has not been conducted. Furthermore, there are no publicly available 3D point cloud datasets for station environments. In this research, we developed a wide-area, high-density pedestrian flow measurement and tracking system using twenty 3D-LiDAR sensors in a real-world environment to accurately capture and quantify pedestrian flow in crowded large-scale station concourse. By employing an offline person detection model using deep learning and tracking compensation processing using both past and future point cloud data, we confirmed that high-accuracy person detection and tracking with HOTA exceeding 90% is possible in an area of approximately 900 m². Using the developed system, we successfully measured walking trajectory data with high accuracy and visualized flow trends using basic indicators and heatmaps derived from the measurement data, thereby evaluating the pedestrian flow characteristics in the station concourse environment.
Hishida et al. (Thu,) studied this question.