基于YOLO和KCF的目标跟踪算法研究

Research on Target Tracking Algorithm Based on YOLO and KCF

摘要: 为了解决在目标跟踪过程中因录制设备发生偏移带来的跟踪偏移和目标丢失问题，提出了基于YOLO和KCF的目标跟踪算法。本文使用基于回归的端到端设计思想的YOLO算法实现目标检测，在目标检测前，对视频图像帧进行灰度化和均值滤波法实现图像增强，减少冗余数据，去除噪点。随后通过YOLO算法进行KCF算法跟踪框的初始化，在跟踪过程中设置偏移误差率(Offset error rate, OER)实时调整跟踪窗口位置，实现目标实时准确跟踪。实验结果表明，本文所提出的方法在面对录制设备发生偏移时相比于KCF算法、Camshift算法有较高的跟踪准确率和鲁棒性。

Abstract: In order to solve the problem of tracking offset and target loss caused by the offset of the recording device during target tracking, a target tracking algorithm based on YOLO and KCF is proposed. This paper uses the YOLO algorithm based on the end-to-end design idea of regression to achieve target detection. Before target detection, the video image frame is grayed and average filtered to achieve image enhancement, reduce redundant data and remove noise. Afterwards, the tracking frame of the KCF algorithm is initialized by the YOLO algorithm, and the offset error rate (OER) is set in the tracking process to adjust the tracking window position in real time to achieve real-time tracking of the target. Experimental results show that the method proposed in this paper has higher tracking accuracy and robustness than KCF algorithm and Camshift algorithm when facing the recording device.