Multi-Target Tracking and Trajectory Planning Framework for Autonomous Vehicles Based on Monocular Vision

In the pursuit of enhancing autonomous vehicle capabilities, efficient multi-object tracking and trajectory planning are critical components. This paper proposes an integrated framework for multi-Target tracking and motion planning, leveraging monocular vision. The framework employs the You Only Look Once v8 (YOLOv8) algorithm as the object detector, which accurately identifies multiple obstacles in real-Time. For tracking these detected obstacles, the Deep Simple Online and Realtime Tracking (DeepSORT) algorithm is utilized, which ensures robust multi-Target tracking by associating detected objects across consecutive frames. Based on the trajectories of tracked obstacles, the Probabilistic Roadmap (PRM) algorithm is utilized to generate a reliable and efficient path for the self-driving vehicles. To evaluate the feasibility and effectiveness of the proposed framework, we designed and executed a series of experiments.