针对现实场景中跟踪目标的光照变化、尺度变化、遮挡等问题,提出了一种基于通道可靠性的多尺度背景感知相关滤波跟踪算法。通过提取方向梯度直方图特征、灰度特征和颜色属性特征作为目标表观模型,提高了目标跟踪方法在复杂场景中的稳健性;独立训练每个通道的背景感知相关滤波器,采用通道可靠性系数衡量每个通道响应图的置信度;根据所有通道的响应图和可靠性系数,合成多通道背景感知相关滤波跟踪器的最终响应图,对目标进行精确定位;运用尺度池方法估计目标的最优位置和尺度。实验结果表明:与现有跟踪算法相比,所提算法可以有效地处理光照变化、尺度变化、遮挡等复杂因素的干扰,取得较高的跟踪精度和成功率,其整体性能优于其他算法。

To solve the problems of illumination variation, occlusion, and scale variation during object tracking, we propose a multi-scale context-aware correlation filter tracking algorithm based on channel reliability. First, we extract the histogram of the oriented gradient (HOG), gray features, and color name (CN) features as the appearance model of the object, which can enhance the robustness of the tracking algorithm in a complex scene. Second, the single-channel context-aware correlation tracker is independently trained by applying the related channel feature samples. The channel reliability factor is applied to evaluate the confidence of each channel. Then, the final response map of the multi-channel context-aware correlation tracker comprises the response maps and the channel reliability values of all the channels, and it is used to accurately locate the object. Finally, the scale pool method is applied to estimate the optimal position and scale of the object. When compared with the results obtained using the state-of-art trackers, the experimental results show that the proposed algorithm can effectively tackle the illumination variation, occlusion, scale variation, and other complicated factors, and achieve relatively high tracking accuracy and success rate. The overall performance of the proposed algorithm is superior to those of other algorithms.