The importance of intelligent transportation system in improving traffic efficiency and safety is increasingly prominent. In order to realize real-time and accurate identification and tracking of multi-targets in traffic scenes, this paper proposes an innovative algorithm fusion strategy, which combines the YOLOv8 target detection algorithm with the DeepSORT multi-target tracking algorithm. YOLOv8 can effectively identify traffic participants such as vehicles and pedestrians with its efficient and accurate target detection ability. DeepSORT, on the other hand, shows stable tracking performance in complex environment by combining deep learning features with traditional tracking algorithms. In order to further improve the robustness of tracking, this paper introduces a dynamic feature update strategy to update the feature representation in real time according to the appearance change of the target. At the same time, the tracking decision is optimized by using the confidence score output by YOLOv8, and the target with high confidence is given priority and the tracking is suspended when it is low. The experimental results show that the fusion algorithm performs well on traffic surveillance video data sets, which is significantly better than using YOLOv8 or DeepSORT alone, and shows higher accuracy, recall and F1 score compared with other advanced algorithms. This study provides strong technical support for intelligent transportation system and lays a foundation for realizing more efficient and safe traffic management.
Design of Intelligent Transportation Multi-target Recognition and Tracking Algorithm Combining YOLOv8
2024-08-14
351050 byte
Conference paper
Electronic Resource
English