Preventive Safety for Unmanned Door Operation Using Integrated Passenger Body Parts Tracking and Rush Prediction

We propose an industrially realistic method to introduce a preventive safety mechanism for door operations of unmanned buses. The method adopts object detection to detect passenger's body parts and applies multi-object tracking (MOT) to track the movements of individual passengers inside and outside of the bus. In addition, we propose a lightweight method based on optical flow to predict objects rushing towards the door of the bus while the door is being closed, which contributes to preventing collision. These methods complement each other to enable preventive safety with respect to door operation.