基于方向梯度的相关图算法在人体动作识别中的应用

Application of Correlogram of Oriented Gradient Algorithm in Human Action Recognition

针对人体动作的特征提取法,该文提出了基于方向梯度的相关图算法Correlogram of Oriented Gradient（COG）,它是通过检查视频动作中的时间空间兴趣点和以兴趣点为中心的空间立方体,计算并提取空间立方体的时间空间方向梯度所具有的空间结构分布特性和外观信息,建立不同动作的特征模型,并用支持向量机（SVM）分类器来检验特征模型的识别准确率;最后,通过对比基于方向梯度的柱状图算法Histogram of Oriented Gradient（HOG）,该文提出的COG算法的识别准确率比HOG算法高15%左右。

This paper propose a novel feature description algorithm based on oriented gradient named as Correlogram of Oriented Gradient （COG）,which is to detect the space-time interest points and describe the unique features around the detected points （cuboids）,and compute the spatial temporal gradients of each cuboid taking advantage of both the spatial structure and appearance information. Then by building the recognition model for each action and utilizing the SVM classifier,we obtain the experimental results of COG which is 15% better than the Histogram of Oriented Gradient （HOG）.