位置-尺度异空间协调的多特征选择相关滤波目标跟踪算法

针对传统相关滤波跟踪算法在目标发生尺度变化和遮挡时容易导致跟踪失败的问题，本文提出位置-尺度异空间协调的多特征选择相关滤波目标跟踪算法。首先，提取目标区域的快速方向梯度直方图特征、颜色空间特征和灰度特征，特征间的不同组合方式构成特征池以加强滤波器的判别性能，将组合得到的特征分别进行相关滤波跟踪；其次，依据每种特征响应的鲁棒性得分，选择分数最高的响应图最大值预测目标位置；然后，转换坐标至对数极坐标中，使用相位相关滤波器进行目标尺度估计；最后，设计一种高置信度模型策略更新模板。在标准数据集TB-50和OTB-2015上的实验结果表明，本文提出的算法在目标发生尺度变化、遮挡、旋转、出视野和背景杂乱等情况下，仍具有较好的跟踪有效性。 

Multi-feature Selection Correlated Filtering Target Tracking Algorithms Based on Location-Scale Different Space Coordination

Aiming at the problem that the traditional correlation filtering tracking algorithm is easy to fail when the target is occluded and scale variation occurs, this paper proposed a multi-feature selection correlation filtering target tracking algorithm based on position-scale spatial coordination. Firstly, the features of fast histogram of oriented gradient, color names and average grey value of the region are extracted then different features are fused to form feature pools for correlation filtering tracking to enhance the discriminant capability of tracker,. Secondly, according to the robustness score of each feature response, the maximum value of response map with the highest score is selected to predict the target location. Then, the target scale is estimated using phase correlation filter in log-polar coordinates. Finally, updating filter template with the strategy of high confidence model updating. The experimental results on standard datasets of TB-50 and OTB-2015 prove that the proposed algorithm performs effectively under the circumstances of scale variation, occlusion, out-of-plane rotation, out-of-view and background clutters.