A Vision-based Pedestrian Comity Pre-Warning System

Despite the rapid development of assisted driving systems in recent years, vehicle-pedestrian accidents still account for a large proportion of serious traffic accidents. And most existing pedestrian protection methods in the ADAS system only focus on pedestrian detection and immediate collision warning, they fail to fully combine the lane detection to pedestrians for early warning. So the most existing pedestrian protection methods do not know when the warning should be produced, even maybe produce an unnecessary warning to distract the driver's attention. So we combine traditional detection algorithm with deep learning to develop a pedestrian comity pre-warning system based on vision algorithm and embedded system. The system includes a series of innovations such as dynamic ROI and similarity-based pruning. Finally, the experiment shows that the system has good performance and low consumption of computing resources.