基于时间感知和自适应空间正则化的相关滤波跟踪算法

Correlation Filter Tracking Algorithm Based on Temporal Awareness and Adaptive Spatial Regularization

针对相关滤波器的空间正则化权重与目标内容无关和跟踪过程中模型退化等问题,提出一种基于时间感知和自适应空间正则化的相关滤波跟踪算法。首先,提取灰度特征、CN(color name)特征和方向梯度直方图(HOG)特征来提升算法模型对目标的表达能力;其次,通过图像显著性检测算法获得带有目标内容信息的空间正则化初始权重;然后,在目标函数中加入自适应空间正则化项来缓解边界效应对相关滤波器的影响;最后,加入时间感知项使相关滤波器学习到相邻帧之间的信息,降低算法模型在处理不准确样本时发生过拟合的风险。在OTB-2013和OTB-2015公开数据集上对所提算法进行性能评估实验,结果表明,所提算法在多种复杂场景下都有良好的稳健性,在跟踪成功率和距离精度上优于其他对比算法,且速度达到24.2 frame/s,能满足实时性要求。

To solve the problem related to spatial regularization weight, which is independent of the target content, and model degradation during tracking in correlation filter, a correlation filter tracking algorithm based on temporal awareness and adaptive spatial regularization is proposed herein. First, we extract the gray, color name, and histogram of oriented gradient(HOG) features to improve the aptitude of the model. Second, the initial spatial regularization weight with the target information is obtained using an image saliency detection algorithm. Subsequently, we add an adaptive spatial regularization term to the objective function to alleviate the influence of the boundary effect on correlation filter. Finally, a temporal awareness term is introduced to make the correlation filter learn the information between adjacent frames and reduce the overfitting risk of the model when handling inaccurate samples. In this paper, we evaluate the performance of the proposed algorithm on the public OTB-2013 and OTB-2015 datasets. The results show that the algorithm in this paper has good robustness in a variety of complex scenes, better than other comparison algorithms in tracking success rate and distance accuracy, and the speed reaches 24.2 frame·s-1, which can meet the real-time requirements.