摘要
针对Mean shift(即MS)算法理论上的不足以及跟踪目标时的邻域跟踪局限性,提出将Mean shift算法与尺度无迹卡尔曼滤波器(Scaled unscented Kalman filter,SUKF)相结合的实时目标跟踪算法。该算法利用尺度无迹卡尔曼滤波器获取Mean shift算法的初始位置,然后,利用Mean shift算法获取跟踪位置。通过分析跟踪区域内横纵向直线的统计变化获取目标的尺度变化,依此自适应调节Mean shift跟踪算法中核函数带宽,并对高速公路上快速运动的车辆进行跟踪实验。研究结果表明:该算法与固定核窗宽Mean shift算法相比,对目标跟踪更准确;SUKF滤波使MS的迭代次数减少,跟踪的实时性提高;核窗宽自适应调节可使跟踪误差降低到50%以下。
Aiming at theoretic and neighborhood tracking limitation of Mean shift （MS）, a real-time target tracking algorithm combined with Mean shift and scaled unscented Kalman filter （SUKF） was proposed, The algorithm firstly used SUKF to estimate the starting position of the Mean shift in every frame. And then the Mean shift was used to locate the target position. The target scale changing was estimated by analyzing the statistical changes of the horizontal and vertical lines in the tracking region. According to the estimated scale, the kernel-band width changed adaptively for Mean shift object tracking. Experiments on the highway vehicle tracking were done. The results show that the proposed tracking algorithm can locate the target more accurately than the traditional Mean shift; the SUKF can reduce the iteration times of MS and improve the real time of tracking. Adaptive adjustment of kernel-band width further can reduce the tracking errors to less than 50% of the MS algorithms.
出处
《中南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2011年第5期1338-1343,共6页
Journal of Central South University:Science and Technology
基金
国家自然科学基金重大专项项目(90820302)
国家自然科学基金面上项目(60805027)
国家博士点基金资助项目(200805330005)
