Speed prediction model at urban intersections considering traffic participants

In order to improve the performance of speed prediction in the state of free driving at urban intersections, a new method for speed prediction that considers the interaction characteristics of the host vehicle with other traffic participants is proposed. First, a vehicle target classification method is proposed to distinguish the driving direction of other vehicles relative to the host vehicle, and the target detection algorithm YOLOv5 is used to identify potential traffic conflicts and vulnerable traffic participants. Then, the identified traffic participant and historical speed are combined to establish a speed prediction model based on long short-term memory network. The effectiveness of traffic participant information in improving speed prediction performance is verified in three different driving scenarios, i.e. left turn, right turn and straight. The results show that compared with the baseline model that only takes historical speed as input, the speed prediction model considering traffic participants shows better performance. It solves the problem of the gradual decline in the accuracy of the prediction model in a prediction domain, and shows stronger adaptability to the complex traffic environment of urban intersections.
QC 20230829