A Deep Learning-based Approach to Line Crossing Prediction for Lane Change Maneuver of Adjacent Target Vehicles

Abstract

Forecasting the motion of surrounding vehicles is a key issue for autonomous vehicles to assess potential risks and avoid collisions. Among them, the sharp lane change of vehicle in adjacent lane has a greater impact on the ego vehicle. In this paper, we propose a deep learning-based approach to predict the lane change maneuver of adjacent vehicles and quantitatively estimate the position and time to line crossing point (PTLC). In order to distinguish the real lane change from an unintentional drifting between lane boundaries and make accurate prediction of the line crossing point, the features of vehicle kinematics and the driver's driving style as well as the interaction with surrounding vehicle are extracted. Furthermore, a deep neural network is used to process and fuse these features to obtain the probability distribution of PTLC, in which a gated recurrent units (GRU) is adopted as an improvement to robustly learn the historical trajectory of the adjacent target vehicle. Experiments using the data collected from highways show that the proposed method can achieve a reliable prediction of the driver's intention and line crossing point.