LSTM을 이용한 주변 차량 경로 예측 모델 개발

초록

In this paper, we design a surrounding vehicle trajectory prediction model based on deep learning. The most important goal of the autonomous driving system is ensuring driver safety. Identifying future driving intentions of surrounding vehicles may be applied to risk assessment and contribute to securing safety of autonomous vehicles. It is possible to predict avehicle trajectory using a vehicle model, but this method does not reflect interaction between surrounding vehicles. It differs from the actual movement in a driving environment where many surrounding vehicles exist, such as urban areas and highways. To improve this, a long short-term memory(LSTM) based surrounding vehicle trajectory prediction model is designed in consideration of mutual effects between vehicles. This proposed method was designed and validated with High D datasetwhich is vehicle trajectories recorded on German Highways. The sequence of the speed of the Ego vehicle, the heading angle, the relative coordinates of the surrounding vehicles, and the relative speed were used as inputs for vehicle trajectory prediction.