Vision-based action recognition is an essential component of human motion analysis and has been receiving broad attention in the field of computer vision. Human action in the video sequence can be seen as a 3D space-time volume stacked by silhouette image of each frame according to temporal series. In this paper, a novel representation of local binary patterns, named volume semantic local binary patterns (VSLBP) is proposed and applied to extract low-dimensional features on the space-time volume. Action label can be assigned by computing the nearest chi-square distance of features between each probe silhouette sequence and gallery silhouette sequence sets. Experiment results on the publicly available "Weizmann" dataset demonstrate that the proposed approach outperforms the state-of-the-art methods and can also tolerate various challenging conditions, I. E. , partially occluded, changes in the viewpoints, motion style, etc.%基于视觉的行为识别是人体运动分析的重要组成,也是该领域一个富有挑战性的研究方向,因此获得广泛的关注.该文将视频中的人体行为看成由每帧轮廓图像沿时间轴堆叠而成的三维空-时体积,提出一种新颖的局部二值模式,即体积语义局部二值模式(VSLBP),用于提取空-时体积中的有效低维特征,通过计算测试序列与已标记的行为训练集特征间的最近卡方距离得到其所属行为类别.在行为库“Weizmann”上实验结果表明,该文提出方法的识别准确率略高于现有最新方法,且能容忍各种复杂的条件,如遮挡、拍摄角度、行为者外观穿着及行为方式等.
展开▼
