针对目标发生形变、遮挡以及尺度变化导致跟踪失败的情况, 本文提出了一种改进的多特征融合的目标跟踪算法。首先, 通过计算方向梯度直方图(Histogram of Oriented Gridients, HOG)和颜色命名(Color Names, CN)特征响应相邻两帧峰值旁瓣比(Peak-to-Sidelobe Ratio, PSR)的差值得到这两种特征的融合权重, 用得到的权重对HOG和CN特征响应进行自适应融合, 将第一次融合后的响应与颜色直方图特征获得的响应以固定权重进行二次融合,并根据融合结果确定目标中心位置。其次, 结合最终目标响应值的PSR与其均值的差值变化, 对位置相关滤波器和尺度相关滤波器的学习速率进行动态调整。最后, 在OTB50标准数据集上进行实验验证, 并和其他跟踪算法进行对比。实验结果表明: 本文算法在多项性能指标上均优于其他算法, 其中精度为81.9%, 成功率为61.1%, 能有效适应形变、遮挡以及尺度变化场景下的目标跟踪。

Aiming at the situation that the target deformation, occlusion and scale variation lead to tracking failure, an improved multi-feature fusion target tracking algorithm is proposed. Firstly, the fusion weights of the two features are obtained by calculating the difference between the two adjacent PSRs of the HOG(Histogram of Oriented Gridients) and CN(Color Names), and the HOG and the CN are adaptively fused by using the obtained weights. The response obtained by the first merged response and the color histogram feature is secondarily fused with a fixed weight, and the target center position is determined according to the fusion result. Then, the learning rate of the position correlation filter and the scale correlation filter is dynamically adjusted by the difference between the PSR(Peak-to-Sidelobe Ratio) of the response and its mean value. Finally, the experimental verification is performed on the OTB50 standard dataset, and compared with other algorithms. The results show that the proposed algorithm outperforms other algorithms in many performance indicators. The accuracy is 81.9%, and the success rate is 61.1%. It can effectively adapt to target tracking under deformation, occlusion and scale variation scenarios.