Research on collision avoidance control of autonomous vehicles based on pedestrian trajectory prediction

Release time:2026-07-01  Hits:

• DOI number:10.1016/j.displa.2026.103474
• Journal:Displays
• Key Words:Autonomous vehicles; Pedestrian detection and tracking; Pedestrian trajectory prediction; Collision avoidance control
• Abstract:Autonomous driving technology not only promises to significantly improve road safety, but also effectively alleviate urban traffic congestion. However, current autonomous driving systems still face numerous challenges in pedestrian safety, especially in dealing with the dynamic variation of pedestrian behavior. To address these issues, this study proposes an active collision avoidance control strategy based on pedestrian trajectory prediction. The strategy first introduces a lightweight YOLOv5-Light model for object detection, combined with the DeepSORT algorithm to accurately obtain pedestrian historical trajectories. Subsequently, an improved LSTM model is used to predict pedestrian movement trends, which considers the influence of different traffic participants on pedestrian trajectories through a dynamic interaction mechanism. The prediction model demonstrates excellent performance on the InD dataset, with average displacement error and final displacement error reduced by 15.94% and 18.39%, respectively. Based on the prediction results, a MPC framework is presented for collision avoidance control by utilizing the dividing collision risk areas. Finally, comprehensive simulation tests have fully validated the reliability and effectiveness of this approach.
• First Author:武广鑫
• Correspondence Author:李绍松
• Discipline:Engineering
• Document Type:J
• Volume:94
• Page Number:103474
• ISSN No.:0141-9382
• Translation or Not:no
• Date of Publication:2026-04-07
• Included Journals:SCI