基于改进的Heaviside核函数新的目标模型追踪算法

为更好地解决前景和背景相似程度较大或目标运动较为复杂的问题,提出了基于改进的Heaviside核函数新的目标模型追踪算法.在初始帧中,使用改进的Heaviside核函数来表示目标区域,然后分别计算目标区域前景和背景元素的颜色纹理直方图特征分布,并通过前景和背景特征分布差异建立新的目标模型,它可更好地代表目标.对于候选模型,结合传统Epanechnikov核对目标模型建模,通过Bhattacharyya系数进行迭代搜索,最终收敛的位置即为下一帧的目标中心.实验结果表明:提出的算法和传统的Mean-shift算法和基于颜色纹理直方图的Mean-shift算法相比较精确度高、速度快、鲁棒性强.

New Target Model Based on Heaviside Kernel Function for Tracking

In order to better solve the tracking problem with compromised target identification arising from the similarity between foreground and background or higher motion mobility, a new target model is put forward based on the modified Heaviside function. The target region is first represented by the proposed function, then the background and foreground color-texture histograms is constructed respectively in the region in the initial frame, in so doing the new target model is built through comparing the difference between the foreground and background model. In the candidate target area, the traditional Epanechnikov kernel is adopted in modeling target, and the target center is determined in the next frame using the Bhattacharyya coefficient. The experimental results show that the new algorithm performs better in accuracy and robustness than the traditional Mean-shift tracker and the existing color-texture-combined histograms based Mean shift.