Forward collision warning system using multi-modal trajectory prediction of the intelligent vehicle

As one of important advanced driving assistance systems (ADAS) for intelligent vehicles, forward collision warning system is an effective solution to avoid the traffic accidents. Accurately trajectory prediction can obtain the future spatial distribution of the vehicles in contextual traffic, which will help the driver make the decisions and avoid potential collisions. This paper proposes an effective trajectory prediction method based on long short term memory model integrated with attention mechanism and regularization strategy (AR-LSTM). The long short-term memory and convolutional neural network (LSTM-CNN) is utilized to recognize the driver’s lane change intention. Combined with the multi-modal information of the vehicles such as vehicle status variables, traffic information, and driver’s lane change intention, the AR-LSTM model is designed to predict the vehicle’s future trajectory. Besides, the AR-LSTM and LSTM- CNN model are trained and tested by the real traffic data set NGSIM (next generation simulation). Finally, considering two maneuvers to avoid the collisions by braking or changing lane, the proposed forward collision warning system is verified by the hardware-in-loop (HIL) platform compared to the existed fixed TTC (time to collision) method. And the statistical results demonstrate that the proposed system can provide correct warnings of braking and decrease invalid warnings of braking.