基于关键点梯度特征描述的人体行为识别算法

摘要

A scheme of new human action recognition is presented and its algorithm is performed. The interest points in video sequence are extracted through Gaussian filtering in space of video sequence and Gabor filtering in time axis. The GLOH (gradient location orientation histogram) is adopted to describe the features of the threshold 20×20 region at each interest pointe. The described sequence may feature the video sequence. Comparing with other human action recognition algorithms, only by establishing a single support vector classifier can the algorithm realize the high recognition rate, that is, it need not label the given feature region and use the classification algorithm which wastes time. The algorithm is simple and efficient.%提出一种新的人体行为识别方案并进行了算法实现.通过对视频序列在空间上高斯滤波,在时间轴向上Gabor滤波,提取出视频序列的关键点,对每个关键点邻域20×20的区域使用梯度位置朝向直方图进行描述,描述的序列可以表征视频序列的特征.与其他人体行为识别算法比较,不需要标记特定的特征区域和比较耗时的聚类算法,构建单个支持向量分类器即可达到好的识别率,算法简单有效.