基于Transformer改进的YOLOv5+DeepSORT的车辆跟踪算法

Vehicle Tracking Algorithm Based on Transformer’s Improved YOLOv5+DeepSORT

针对传统目标检测跟踪算法检测精度低、全局感知能力差、对遮挡和小目标物体的识别能力差等问题,提出了一种基于轻量化Transformer改进的YOLOv5和DeepSORT算法的车辆跟踪方法。首先,利用EfficientFormerV2模型改进YOLOv5算法模型,增强车辆的目标检测能力;然后,利用移位窗口(Swin)模型的优点改进DeepSORT多目标跟踪算法中的重识别(Re-Identification)模块,提高车辆的跟踪能力和精度;最后,通过数据集KITTI和VeRi开展对比试验和消融实验。结果表明,在复杂工况下,该方法的性能在车辆遮挡和小目标识别方面显著提高,平均准确度达到96.7%,目标跟踪准确度提高了9.547%,编号(ID)切换总次数减少了26.4%。

In order to solve the shortcomings of traditional object detection and tracking algorithms,such as low detection accuracy,poor global perception ability,poor recognition ability of occlusion and small target objects,this paper proposed a vehicle tracking method based on YOLOv5 and DeepSORT algorithm improved by lightweight Transformer.Firstly,the EfficientFormerV2 model was used to improve the YOLOv5 algorithm model to enhance the target detection ability of the vehicle,and then the advantages of the Swin model were used to improve the Re-Identification module in the DeepSORT multi-target tracking algorithm to enhance the tracking ability and accuracy of the vehicle.Finally,the dataset KITTI and VeRi were used to carry out comparative experiments and ablation experiments.The results show that under complex conditions,the performance of the proposed method is significantly improved in vehicle occlusion and small target recognition,with an average accuracy of 96.7%,an increase of 9.547%in target tracking,and a reduction of 26.4%in the total number of ID switching.