多特征联合时空正则化相关滤波目标跟踪鲁棒算法

Correlation Filtering Target Tracking Algorithm with Multi⁃feature Joint Spatio⁃temporal Regularization

针对相关滤波跟踪框架中快速运动带来的边界效应和遮挡情况下模型错误学习的问题,提出多特征联合时空正则化的相关滤波目标跟踪算法。算法在第一帧提取目标区域的快速方向梯度直方图特征、颜色空间特征和深度卷积特征,并使用主成分分析法降低特征的维度;然后在相关滤波跟踪框架中加入空域和时域正则化项,来缓解跟踪过程中边界效应和模型退化等问题;最后结合尺度池方法,对跟踪目标进行自适应的尺度估计。实验结果表明,该算法在目标发生尺度变化、遮挡、快速运动等情况下,仍具有较好的跟踪有效性。

Aiming at the problem of boundary effects caused by fast motion and the model error learning under occlusion in the correlation filtering tracking framework,a multifeature joint spatiotemporal regularization correlation filtering target tracking algorithm was proposed.Using the algorithm,the features of fast histogram of oriented gradient,color names and deep convolution of the target area were extracted in the first frame,then the principal component analysis was used to reduce the dimensions of the features;at the same time,the spatiotemporal regularization was added to the correlation filtering tracking framework in order to alleviate the problems of boundary effects and model degradation in the tracking process.Finally adaptive scale estimation was performed with the help of scale pool method.The experimental results prove that the proposed algorithm performs effectively under the circumstances of scale variation,occlusion and fast motion.