|
| 基于改进的DeepSORT算法的多目标车辆跟踪研究 |
| Research on Multi‑target Vehicle Tracking Based on Improved DeepSORT Algorithm |
| |
| DOI：10.16018/j.cnki.cn32-1650/n.202403011 |
| 中文关键词: YOLO v8 卡尔曼滤波 多目标跟踪 DeepSORT 数据关联 |
| 英文关键词: YOLO v8 Kalman filter multi-target tracking deepSORT data association |
| 基金项目: |
|
| 摘要点击次数: 45 |
| 全文下载次数: 48 |
| 中文摘要: |
| 为了解决传统多目标车辆跟踪中实时交通运行监控视频的DeepSORT跟踪算法精度低，以及因视频中车辆遮挡面积大导致车辆目标跟踪准确率降低的问题，提出了一种采用端到端的YOLO v8检测方法代替传统的YOLO v8检测方法，同时DeepSORT数据关联采用整合目标的运动状态和外观属性，再通过卡尔曼滤波器将车辆的静态特征、运动轨迹和外观细节结合起来，并根据环境光照条件动态调整两者的权重，以优化低照度下的追踪性能。试验结果表明: 相较于传统算法，本文提出的算法能够很好地处理遮挡和不同的照明条件，其精度和速度也均优于传统的多目标车辆跟踪检测算法。 |
| 英文摘要: |
| In order to solve the problems of low accuracy of DeepSORT tracking algorithm of real-time traffic monitoring video in traditional multi-target vehicle tracking, and the decrease of vehicle tracking accuracy due to large vehicle occlusion area in the video, an end-to-end YOLO v8 detection method is proposed to replace the traditional YOLO v8 detection method. At the same time, DeepSORT data association integrates the motion state and appearance attributes of the target, and then combines the static characteristics, motion trajectory and appearance details of the vehicle through Kalman filter, and dynamically adjusts their weights according to the ambient lighting conditions to optimize the tracking performance in low illumination. The experimental results show that compared with the traditional algorithm, the algorithm proposed in this paper can deal with occlusion and different lighting conditions well, and its accuracy and speed are also better than the traditional multi-target vehicle tracking detection algorithm. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |