基于空间边缘方向直方图的Mean Shift跟踪算法

Mean Shift Tracking Algorithm based on Spatial Edge Orientation Histograms

传统的基于色彩直方图或空间色彩直方图的Mean Shift跟踪算法,在诸如跟踪目标出现尺度变化的复杂条件下,无法得到准确的跟踪结果。这是因为色彩直方图或空间色彩直方图无法显著区分颜色相近的目标和背景。鉴于此,提出了一种基于空间边缘方向直方图的Mean Shift跟踪算法,使用空间分布和纹理信息作为匹配信息。实验结果表明,该算法能够有效的处理遮挡、光照变化和尺度缩放等复杂情况,对目标进行准确有效的跟踪,改善了传统方法在尺度缩放等方面的局限性。

Traditional mean shift tracking algorithm based on color histogram or spatial color histogram sometimes fails to get accurate results under intricate condltions, such as scale modifications occur to the objects. That is because the histograms based on color cannot distinguish objects and background have the same color. This paper presents a new mean shift algorithm based on spatial edge orientation histograms, using space distribution and texture information as matching information. Experiments proved the new algorithm can deal with intricate conditions, such as occlusion, different luminance and scale modifications, and track objects accurately, effectively and real time. The new algorithm also overcomes the limitations of the traditional one.