基于骨架模型的人体动作识别方法

Human Action Recognition Based on Skeleton Model

深度传感器以其高性价比特性越来越受到研究人员的青睐,结合姿态估计算法来自动识别人体动作,是当前计算机视觉领域中的研究热点之一。本文构建骨架序列模型作为人体姿态描述器,提出一种基于Citation-KNN算法的动作分类框架,可以有效应对类内差异和类间相似问题。此外,对于调节噪声和时间不一致的容忍度来说,该方法亦提供了一个有效途径。仿真实验证明了该方法在两个动作识别数据集上能取得非常好的效果。

Depth sensor been increasingly become the focus of researchers,since it could be combined with the human-pose estimation algorithms to achieve human action recognition which is one of the research hotspots in computer vision field.In this paper we model skeleton sequences as the pose descriptors and propose a novel framework for human action recognition based on Citation-KNN.Our method could effectively avoid intra-class variability and inter-class similarity problem,and for regulating the noise and temporal misalignment tolerance,an effective way is also provided.Simulation experiments show that the proposed method on two human action recognition datasets could obtain pretty good results.