基于Mean-Shift算子的多尺度视频跟踪算法研究  

Research on Multi-scale Video Tracking based on Mean-Shift Algorithm

作　　者：盛磊[1] 刘旨春[1] 于晓波[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《光机电信息》2010年第12期134-139,共6页OME Information

摘　　要：为了提高靶场光测设备视频跟踪算法的稳定性、准确性以及抗干扰性,开发了基于Mean-Shift算子的多尺度视频跟踪算法,对该算法所采用的小波变换、小波包及Mean-Shift目标跟踪算法进行了研究。阐述了整个跟踪算法的原理及多分辨率图像选择依据;介绍了小波变换和小波包原理,说明了利用小波包将视频图像分解为多分辨率图像的方法;介绍了Mean-Shift算子的原理以及对目标特征进行归一化表示;最后说明了利用Mean-Shift算子对归一化目标的搜索区域进行预测的算法。实验结果表明,本算法跟踪过程平稳、准确,且抗干扰能力强,收敛速度快。试验数据表明,经过小波变换的Mean-Shift算法的收敛速度提高约66%,搜索准确性提高约34%。基本满足了靶场测量中对视频跟踪算法更高的要求。In order to improve the stability, veracity and anti-jamming of video tracking algorithm in range instrumentation, an algorithm of multi-scale video tracking based on Mean-Shift was established and its applied algorithms such as wavelet transformation and wavelet-packet, moving target tracking based on Mean-Shift were investigated. The theory of the whole algorithm and the rule of selecting the multi-scale images were explained. The theory of wavelet transformation and wavelet-packet were presented. The method of image decomposing based on wavelet-transformation was analyzed. And then, the theory of Mean-Shift and unitary target expression were presented. Finally, the algorithm of forecasting searching area of unitary target was analyzed. Experimental results indicated that by using the algorithm, the procedure of tracking had more stability, veracity and anti-jamming, the convergence speed of Mean-Shift improved 66%, and veracity improved 34%. It could satisfy the higher requirement to video tracking in range instrumentation.

关 键 词：视频跟踪 小波变换 小波包 MEAN-SHIFT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]