Collision-warning system integrated with merging behaviour prediction model based on multi-sensor fusion

One of the most dangerous situations on roads is that drivers choose to merge into traffic without warning. This paper presents a real-time collision warning system in merging scenario and our approach mainly focuses on the forward vehicle in different lane. First, multi-sensor is used to detect the distance and speed information of forward vehicles. Based on the detection result, a neural network is designed to predict whether they are going to merge into ego lane or not. The prediction model correctly classifies 92% of merging behaviour in our test dataset. Then, a collision warning algorithm is proposed to cope with different merging manoeuvres. The algorithm is tested on a real road on our embedded platform and the results show that the system can effectively alert drivers to brake when collision threats are posed.