核窗宽自适应的均值偏移跟踪算法

Mean-Shift Tracking Algorithm Based on Adaptive Kernel Bandwidth

针对固定窗宽的均值偏移算法对逐渐变大的运动目标跟踪不准确的问题,提出了一种窗宽自适应的均值偏移跟踪算法。先对当前帧进行均值偏移跟踪,再通过后向跟踪使跟踪窗口中心与目标形心匹配,利用巴氏系数最大化对窗宽进行±10%的修正,使跟踪窗口的尺度自适应变化。实验结果表明:该算法提高了跟踪精度,增强了跟踪稳定性,保证了跟踪的实时性。

Mean-shift algorithm with fixed bandwidth often fails in tracking the object that moves with obviously change in scale, especially changing bigger. To solve the problem, a new adaptive bandwidth mean-shift tracking algorithm is proposed. The algorithm first matches the center of the tracking window with the target center by the afterward-tracking method, then uses the principle of maximizing bhattacharyya coefficient to fix the bandwidth by _ 10%, thus makes the bandwidth change adaptively. The experimental results prove that the algorithm improves the tracking accuracy, enhances the tracking stability and ensures the real-time tracking.