【摘要】 为了增强彩色视频中目标外观描述能力和解决跟踪过程中目标尺度变化的问题,提出一种基于分块的多特征融合变尺度目标跟踪算法。设计了一个能处理不同挑战因素下对目标的精确跟踪算法,首先提取HSV分块的颜色直方图特征和PCA-HOG特征并采用多通道线性核函数对两种特征进行融合构建训练样本,然后求解线性岭回归函数获得位置核相关滤波器模型,并以线性核函数来计算候选区域在7个尺度空间上与跟踪目标的响应值,最后利用尺度自适应模板更新模型参数。实验结果表明,提出的算法在彩色视频中不仅能较好地自适应目标尺度的变化,在复杂场景下也具有较强的鲁棒性。更多还原

【Abstract】 To enhance description capability of objects appearance in color video and to solve the problem of object scale-variant during tracking process,object tracking algorithm based on blocking multiple feature integration and scale-variant is proposed. The object can be tracked accurately under different challenge factors by proposed algorithm. Firstly,HSV color histogram features and PCA-HOG features of object region block are extracted. After using linear kernel function fuses the two features,kernelized correlation filter models are obtained by linear ridge regression function. The degree of similarity between the tracking object and candidate region can be calculated by linear kernel function and the maximum response of the classifier on 7 scale spaces are obtained.Finally,the model parameters are updated with the scale adaptive template. The experimental results show that the algorithm can better adapt to object scale-variant in color video and remain high robustness in complex scenes.更多还原