Regional Trajectory Analysis through Multi-Person Tracking with mmWave Radar

Understanding characteristics of an environment can lead to performing more reliable multiple object tracking (MOT). This paper proposes an approach to extract entry and exit points for an environment using MOT trajectories. The observed environment is divided into a grid, in which regions are defined for. The trajectories obtained from the MOT data are organized into their respective regions. For each region an activity heatmap is formed and classified using a convolutional neural network (CNN) to determine if the region consists of either an entry or exit point. The entry and exit points classified are projected onto the MOT plane to illustrate the entry and exit points of the observed environment. The approach proposed in this paper provides a foundation in which future work can build on to enhance the MOT capability in real-time through a greater understanding of the tracking environment characteristics.