为了提高复杂背景下目标跟踪的稳定性,在模板匹配失败时引入目标分割过程,根据新目标区域与模板的相似度分析得出导致匹配失败的具体原因,提出了一种新的自适应模板更新算法。根据模板图像各像素到模板中心的距离构造加权函数,对传统归一化相关算法进行改进,使新算法具有一定的抗遮挡能力。实验结果表明,该算法与固定模板跟踪算法和逐帧模板更新算法相比,在目标尺寸变化和被物体局部遮挡时,跟踪的连续性和稳定性更好。

A new adaptive template update strategy was proposed to increase the stability of target tracking in complex background. When the template matching fails similarity analysis between new target area and the template based on segmentation and correlation was applied to find the matching failure causes. A weighted function based on the distance of each pixel to the template center was put forward. And the normalized cross-correlation algorithm was improved to solve the occlusion issue. Experimental results indicate that the template can be modified adaptively. When the target is occluded or the target size is changed, the new algorithm is steadier compared with fixed template algorithm and each-frame template update algorithm.