计算机工程

• 图形图像处理 • 上一篇    下一篇

基于BRISK特征点改进的跟踪学习检测方法

祝贤坦,石繁槐   

  1. (同济大学 电子与信息工程学院,上海 201804)
  • 收稿日期:2016-06-02 出版日期:2017-02-15 发布日期:2017-02-15
  • 作者简介:祝贤坦(1989—),男,硕士研究生,主研方向为计算机视觉;石繁槐,副教授。
  • 基金项目:
    国家自然科学基金(61175014)。

Improved Tracking Learning Detection Method Based on BRISK Keypoints

ZHU Xiantan,SHI Fanhuai   

  1. (College of Electronics and Information Engineering,Tongji University,Shanghai 201804,China)
  • Received:2016-06-02 Online:2017-02-15 Published:2017-02-15

摘要: 为提高长时目标跟踪的鲁棒性和准确性,提出一种改进的跟踪学习检测(TLD)方法。利用少量具有尺度不变特性的BRISK特征点和均匀分布点组成跟踪点集合代替TLD中的均匀分布跟踪点。这样不仅可以减少跟踪部分的计算量,而且可以提高跟踪的鲁棒性。当跟踪器利用前后项误差检测到遮挡时,通过使用目标的空间上下文信息扩大跟踪范围再次跟踪,进而解决遮挡的问题。实验结果表明,改进的TLD方法在多个测试序列上都有较好的跟踪性能,与传统的TLD相比,鲁棒性更好,准确率更高。

关键词: 跟踪学习检测方法, BRISK特征点, 跟踪点集合, 目标跟踪, 空间上下文

Abstract: In order to improve the robustness and accuracy of the long-term object tracking,this paper proposes an improved Tracking Learning Detection(TLD) method.Tracking points set consists of a few BRISK keypoints with scale-invariant feature and uniformed-distributed points.Tracking points set is used to replace the uniformed-distributed points in TLD to reduce the computation of tracking part and improve the robustness of tracking.The tracking range is extended and the track is taken again by using the spatial context information of the target when forward-backward error meets the condition of occlusion,and then solves the problem of occlusion.Experimental results show that the improved TLD method has better tracking performance on several test sequences,and outperforms the traditional TLD in terms of robustness and accuracy.

Key words: Tracking Learning Detection(TLD) method, BRISK keypoints, tracking point set, object tracking, spatial context

中图分类号: