基于分块的多特征融合变尺度目标跟踪算法  

Object tracking algorithm based on blocking multiple feature integration and scale-variant

作　　者：李凯[1] 刘颖[1] 李娜 戚秀真[2] 

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710061 [2]长安大学信息工程学院,陕西西安710064

出　　处：《电视技术》2017年第1期6-13,共8页Video Engineering

基　　金：公安部科技强警基础专项(2015GABJC5);陕西省国际科技合作计划资助项目(2015KW-014);陕西省教育厅专项科研计划资助项目(15JK166I);西安邮电大学研究生创新基金资助项目(CXL2015-22);西安市碑林区应用技术研发项目(GX1502)

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。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 corre- lation filter models are obtained by lineal＂ ridge regression function. The degree of similarity between the tracking object and candi- date 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.

关 键 词：目标跟踪 多特征融合 变尺度 线性核函数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]