吉林大学学报(信息科学版) ›› 2014, Vol. 32 ›› Issue (5): 458-464.

• 论文 • 上一篇    下一篇

基于梯度投影的视频跟踪算法

李志军, 陈雪, 李志刚, 顾海军   

  1. 吉林大学 通信工程学院, 长春 130012
  • 收稿日期:2013-09-24 出版日期:2014-09-26 发布日期:2014-12-26
  • 作者简介:李志军(1971—), 男, 长春人, 吉林大学工程师, 硕士, 主要从事无线通信网络及嵌入式技术研究, (Tel)86-13504456889(E-mail)zhijun@jlu.edu.cn; 通迅作者: 陈雪(1988—), 女, 吉林九台人, 吉林大学硕士研究生, 主要从事图像处理研究, (Tel)86-13944857168(E-mail)cxxiaoheihei@163.com。
  • 基金资助:

    吉林省科技发展计划基金资助项目(20110356); 吉林省自然科学基金资助项目(201215011)

Video Tracking Algorithm Based on Gradient Projection

LI Zhijun, CHEN Xue, LI Zhigang, GU Haijun   

  1. College of Communication Engineering, Jilin University, Changchun 130012, China
  • Received:2013-09-24 Online:2014-09-26 Published:2014-12-26

摘要:

为实现复杂场景内目标的准确捕获与跟踪, 结合目标特征, 提出基于梯度投影的视频跟踪算法。根据目标先验知识, 对视频流开窗并进行梯度投影, 获取目标区域的位置信息; 通过特征提取和形态学分析提取目标特征参数, 实现目标判定与捕获; 利用目标质心坐标更新跟踪窗位置信息, 实现对目标的跟踪。实验结果表明, 该算法降低了运算量, 实时性强, 实现了对目标的准确、 稳定的跟踪, 对实验场景中的光照变化和疑似目标的干扰具有很强的鲁棒性。

关键词: 视频跟踪, 梯度投影, 特征提取, 目标检测

Abstract:

In order to capture and track the target accurately in complex scene, combined with the target feature, video tracking algorithm which based on gradient projection is put forward. The algorithm based on prior knowledge of object, we open window for the video and conduct gradient projection in the window to obtain the position information of object areas. By feature extraction and morphological analysis of target extracting feature parameters, to determine goals and capture. Using the object centroid coordinates to update the information of tracking window and achieve the target tracking.The results fully prove that the method reduces the amount of computation, gets strong real-time and tracks target accurately and stably, for the scene of experiment illumination change and suspected object interference has a strong robustness.

Key words: video tracking, gradient projection, feature extraction, target detection

中图分类号: 

  • TP311