A Vision-Based Method for Vehicle Forward Collision Warning

Forward Collision Warning (FCW) system can automatically measure the distance between obstacles and provide early warning, which can effectively provide safety guarantee for vehicle travel and reduce the probability of traffic accidents. The vision-based approaches are always popular because vision sensors have the characteristics of low cost and rich image information. In this paper, a vision-based forward collision warning method is proposed. The method contains three main stages: (1) detect obstacle based on multi-feature fusion using convolution neural networks (CNN), (2) estimate the relative distance, relative velocity from vehicles and collision time, (3) design obstacles hazard level discrimination strategy for different road scenarios. The algorithm is tested on our own dataset and the experiment results showed that the method has good feasibility and robustness.