针对核相关滤波(KCF)在跟踪中由于目标出视野以及遮挡导致跟踪失败的问题, 提出一种基于核相关滤波的长时目标跟踪算法。该算法融合梯度直方图特征和颜色提名特征来增强特征的表达能力; 考虑到核相关滤波不能解决尺度变化的问题, 通过定义尺度池、采集不同尺度的样本计算响应值, 然后利用最大响应值得到最佳位置和尺度。最后, 针对在长时间目标跟踪过程中, 有时不可避免地会出现跟踪失败的情况, 通过训练支持向量机对目标进行重新检测以达到长时跟踪的目的。在OTB数据集上对提出的算法和其他主流算法进行对比, 实验结果验证了提出算法的有效性和优越性。

Aiming at the problem of tracking failure caused by out of view and occlusion in object tracking with kernelized correlation filters (KCF), a long-term tracking approach based on KCF is proposed. Firstly, the features of both gradient histogram and color namination are fused to enhance the expression ability of the features. Then, considering that KCF can not deal with scale variation, and by defining a scale pool and collecting samples in different sizes, the response values are calculated. Thus the optimal position and scale of the object can be obtained according to the maximum response value. Finally, a support vector machine(SVM) classifier is trained to re-detect the target so as to achieve a long-term tracking. The proposed algorithm was compared with other tracking algorithms on the online object tracking benchmark (OTB), and the experimental results validate its effectiveness and superiority.