基于微分信息融合的Mean Shift改进跟踪算法

为提高传统均值漂移算法对低对比度图像的跟踪性能，提出一种融合图像微分特征信息的改进算法。根据图像8邻域微分值建立微分图像，利用微分特征建立目标模板和候选区域的微分直方图模型，并确定候选区中心位置的更新向量。将其与利用颜色特征信息确定的候选区中心位置的更新向量相融合，得到改进算法的更新向量。图像的微分信息包含了图像的细节信息以及像素的相对空间位置信息，提高了模型建立时信息的利用率，能够提高目标模型的建模精度。仿真实验结果表明，与传统均值漂移算法相比，改进算法在复杂的背景情况下具有更强的抗干扰性能，能够有效提高目标跟踪的稳定性。

Improved Mean Shift tracking algorithm based on differential information

In order to improve the tracking performance of the low-contrast images, an improved Mean Shift tracking algorithm fusing differential information of images is proposed. The differential image of the origin image is got according to its eight neighborhood differential values. The differential histogram models of the target template and the candidate region are built based on the differential characteristics. The iterative vectors of the central position of the candidate region can be determined by the differential histogram models and the color histogram models. The two iterative vectors can be fused to get a new iterative vector of the improved algorithm. The differential image contains detail information and space position relations, which increases the utilization of information, and enhances the model precision. Simulation results show that the improved algorithm has a better anti-interference performance than conventional methods in complex background, and the stability of the target tracking can be improved.