自适应跟踪窗的Mean Shift算法

Mean Shift Algorithm of Adaptive Target Tracking Window

摘要:

针对经典Mean Shift算法在目标跟踪过程中不能自适应目标的形状变化, 提出了一种跟踪窗自适应的Mean Shift目标跟踪算法。该算法利用权值图像的矩特征推导出了一种目标真实面积的估计方法, 同时利用估计出的目标真实面积校正可表达目标窗口内图像特征的协方差矩阵, 从而计算出目标区域的主轴长度和方向。实验结果表明, 该方法保留了Mean Shift算法简单、有效的特点, 同时该算法的跟踪窗具备当目标发生旋转和尺度变化时的自适应能力, 并且具有较强的鲁棒性。

Abstract:

Aiming at the unadaptive problem of classical Mean Shift algorithm on changing target shapes in tracking process, in this paper, a tracking window adaptive Mean Shift target tracking algorithm is proposed. This algorithm uses the moment characteristic to derive out an evaluate approach of target true area. By the true target area, the covariance matrix of image characteristics in target window is corrected, and the length and orientation of principal axis are figured out. Test results indicate that this algorithm retain the simple and validity of the classical Mean Shift algorithm, at the same time the tracking window of the proposed approach can adaptive the rotation and scale changing in target tracking, and also having robust to a certain degree.