基于自适应初始搜索点预测的目标跟踪算法

Object tracking based on adaptive prediction of initial searching points

为了得到恰当的初始搜索点以使得目标跟踪算法避开背景干扰并缩短搜索距离,提出了一种自适应初始搜索点预测的算法。该算法通过对坐标变换参数的变化率进行Kalman滤波来更好地预测初始搜索点;更重要的是,该算法有效地在线估计Kalman滤波器中的模型噪声功率,而非先验地对它们的取值做出假设,因而能够在没有任何人工干预的情况下动态地根据不同的目标运动状况和搜索精度进行实时调整。大量实景视频流上的实验结果均证实了该算法显著提高了跟踪稳定性,并且大幅降低了计算量。

To obtain a good initial searching point that enables object tracking algorithms to circumvent interference from background and to reduce searching distance, an algorithm of adaptive prediction of the initial searching point is proposed. The algorithm tracks the changing rate of each coordinate transformation parameter through Kalman filter to facilitate better prediction of the initial searching point. More importantly, the powers of the noise models of the Kalman filter are effectively estimated online rather than making artificial assumptions on their values, The algorithm can hence adapt to various target motions and searching precisions in a real-time manner without any manual intervention. Experimental results on a large number of real-world video sequences confirm that substantial enhancement of tracking stability and considerable drop of computational burden are achieved by the proposed algorithm.