摘要
针对复杂背景下目标发生旋转、遮挡、尺度变化和摄像机运动时不能实时跟踪到目标的问题,将最佳核窗宽方法和信息量度量方法相结合,用于粒子滤波框架中,各个粒子通过均值偏移来搜索峰值,同时加入了相应的遮挡策略。实验结果表明,该算法在目标发生旋转、遮挡后仍能很好地跟踪到目标,同时跟踪窗能随目标尺度的大小变化作相应调整,大大提高了算法的实时性和稳健性。
Robust real-time object tracking is the critical task while there are rotation, occlusions, target scale variations and camera motion. It presents the information measure into particle filtering, which has typically been used in combination with best bandwidth. The proposed tracker first produces a few of samples and then shifts the samples toward a close local maximum using Mean-Shift, while multiple hypotheses are processed in cases of occlusion. In the presented tracking examples, the improved algo- rithm can successfully track objects more reliably after rotation, occlusions and can select the proper size of the tracking window in the scenarios that not only the object scale increases but the scale decreases as well.
出处
《电视技术》
北大核心
2009年第1期24-26,53,共4页
Video Engineering
基金
四川省教育厅重点资助项目(2006A097)
国防基础科研项目(C1020060355)
关键词
遮挡
尺度变化
信息度量
粒子滤波
均值偏移算法
occusion
scale variations
information measure
particle filter
Mean-Shift algorithm
