Inference of Vehicle Lane Change Intention Using Multiple Model Estimator in Automated Highway Driving

Abstract

One of the most critical topics in vehicle active safety control is collision avoidance(CA) maneuver. To ensure the robustness of the CA, it is essential to recognize the behavior of surrounding vehicles accurately. In particular, a safer path can be generated, if the intention of changing lanes of surrounding vehicles can be predicted. Existing studies on lane change intention prediction are primarily based on machine learning, and it is difficult to respond to unexpected situations that have not been learned. In this study, a method for predicting lane change intention in real time based on the trajectory of surrounding vehicles is presented. It is assumed that the location of the lane is known through the map, and the global coordinate system is transformed into the Frenet coordinate system to maintain generality regardless of the curvature of the road. And the paths that the target vehicle can travel are modeled as cubic spline curves on the Frenet coordinate system. Through the multiple model estimator, which operates the path models in parallel, it finds the most probable path and predicts the lane change intention. The performance of the lane change intention prediction algorithm is verified through highD, a German highway vehicle trajectories dataset.