基于核相关滤波的自适应目标跟踪算法

Adaptive Object Tracking Algorithm Based on Kernelized Correlation Filter

利用核相关滤波器跟踪框架,提出了一种结合自适应目标颜色属性的目标位置检测和自适应的目标尺度检测的核相关滤波跟踪方法。首先构建循环样本矩阵,引入颜色属性对目标进行特征描述并获得目标位置,然后在计算出的位置处进行目标尺度检测,并在线更新相关滤波器。最终实验结果显示,算法的中心位置误差减少了20.47 px;在阈值为20 px时,平均距离精度提高了24.0%,在如目标尺度变化、光照变化、部分遮挡、运动模糊等复杂场景下有较强的鲁棒性。

Based on the kernelized correlation filter tracking framework, a tracking algorithm based on adaptive target color attribute detection and adaptive target scale detection was proposed. Firstly, constructing a cyclic sample matrix, introduce the color attribute to describe the target and get the target position, then, measuring the target scale at the calculated position and update the correlation filter online. Experimental results show that the error of the center of the algorithm was reduced by 20.47 px, the average distance accuracy was improved by 24% at the threshold of 20 px, and the algorithm has strong Robustness in complex scenes like change of target scale, illumination change, partial occlusion and motion blur and so on.