Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (10): 186-192.DOI: 10.3778/j.issn.1002-8331.1809-0198

Previous Articles     Next Articles

Object Tracking Algorithm Fused with YOLO Detection and Meanshift

WANG Zhongmin1,2, DUAN Na1, FAN Lin1   

  1. 1.School of Computer Science and Technology, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
    2.Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
  • Online:2019-05-15 Published:2019-05-13

融合YOLO检测与均值漂移的目标跟踪算法

王忠民1,2,段  娜1,范  琳1   

  1. 1.西安邮电大学 计算机学院,西安 710121
    2.西安邮电大学 陕西省网络数据分析与智能处理重点实验室,西安 710121

Abstract: Aiming at the rapid movement and drift problem of object tracking algorithm, a new object tracking algorithm is proposed combining YOLO(You Only Look Once) with Meanshift. Image enhancement mechanism is applied to remove illumination variations interference while maintaining image information. Binary classifier is used to distinguish object from background for fast object detection, in order to reduce computational complexity of YOLO algorithm. According to object position information, the image sequence is processed by Meanshift, and the object is detected and updated to avoid the object drift phenomenon caused by object moving rapidly. Experimental results demonstrate that the proposed algorithm is more accurate and improves tracking precision by 10.2% on average and improves operation efficiency by 12.56% on average in comparison with DLT(Deep Learning Tracker). The proposed algorithm suits the rapid movement, which has strong robustness and tracking efficiency.

Key words: rapid movement, YOLO algorithm, Meanshift, image enhancement, binary classifier

摘要: 针对视频目标跟踪算法中物体快速移动以及均值漂移算法误差累积造成的目标漂移问题,提出了一种融合YOLO(You Only Look Once)与均值漂移的目标跟踪算法。采用图像增强机制对视频帧进行预处理,在保持图像信息的同时去除光照强度的干扰;为了降低YOLO算法的计算复杂度,使用二分类器区分目标和背景进行物体的快速检测。根据目标物体的位置信息,使用均值漂移处理后续图像序列,并对目标物体进行检测更新,避免物体快速移动造成目标漂移问题,从而进行有效的检测跟踪。实验结果表明,该算法与DLT(Deep Learning Tracker)算法相比,运算效率提高了12.56%,跟踪精度提高了10.2%,能够较好地适应物体快速移动,具有较强的鲁棒性和实时性。

关键词: 快速移动, YOLO算法, 均值漂移, 图像增强, 二分类器