Vehicle Forward Collision Warning Algorithm Based on Multi-Information Fusion and Improved Warning Strategy.

Aiming at the problem that traditional forward collision warning (FCW) system cannot deal with some emergencies (such as the sudden appearance of target) accurately and timely in the real world, we present MIFWS-FCW- a novel system that applies multi-information fusion and improved warning strategy to improve the accuracy of emergencies warning. In this work, we construct two dynamic safety warning areas changing with the speed of the vehicle according to the detected lane lines and the safety distance model. We predict the trajectory of target based on the information output by YOLOv3 (You Only Look Once v3) at the same time. Then, three-level warnings are designed based on the improved warning strategy, and we combine the distance perception to decide whether the system sends out strong reminders. We transplant our system to Atlas 200 Developer Kit and conduct quantitative evaluation on 20 videos with different environment. The experimental results show that the MIFWS-FCW system can achieve the competitive warning accuracy of 95.5% with the warning time of 1.16s, to some extent, it can reduce the incidence of accidents and achieve the purpose of auxiliary driving.