UWDoor: Vehicle Dooring Prevention System Using Vehicle-Mounted UWB Nodes

Dooring, a type of accident in which the door of a vehicle collides with an oncoming bicycle as an occupant exits the vehicle, results in a high number of injuries and fatalities. Existing occupant exit warning systems struggle to accurately monitor low-speed, narrow target such as bicyclist accurately. UWB nodes are deployed as a digital key infrastructure in an increasing number of commercial vehicles, which brings new opportunities for dooring prevention. In this paper, we propose UWDoor, a system that forecasts cyclist trajectory and prevents dooring event using vehicle-mounted UWB nodes. We implement a passive sensing platform based on vehicle-mounted UWB nodes for sensing the dynamic target outside the vehicle. To eliminate the frame offset caused by the dynamic influence of target on signal propagation environment, UWDoor takes the static background signal as a reference to calibrate the offset frames. As distance changes, the signal coverage varies rapidly and the reflected signal strength of the target with constant width fluctuates drastically, resulting in the neural network failing to adequately extract target movement features. We design a compensation factor independent of the unknown target width to suppress the fluctuation. We propose a target occupancy grid forecast method based on deep learning to combine target trajectory prediction and dooring event prevention effectively. The experimental results show that UWDoor achieves 89.5% accuracy in the dooring event early warning.