宁夏大学学报（自然科学版）2017，Vol.38Issue(2)：147-152,6.
基于关节点运动轨迹的人体动作识别
Action Recognition Using Trajectories of Joints
摘要
Abstract
A novel action recognition method is proposed by using trajectories of joints to improve the accuracy and performance.Inspired by the experiment of biological motion in psychophysics the trajectories of joints is used for the representation of human action,which can express entirely the action in spatial-temporal dimension.On the basis of above work,Gaussian mixture model is applied for clustering the trajectories.Feature quantization is computed by Fisher vector.Taking into account the real-time requirement of action recognition task,kernel extreme learning machine is adopted to improve the performance.Experimental results on the UTD-MHAD and KARD dataset are provided to demonstrate the proposed method effectiveness.关键词
运动轨迹/高斯混合模型/Fisher向量/核极限学习机Key words
trajectories/Gaussian mixture model/Fisher vector/kernel extreme learning machine分类
信息技术与安全科学引用本文复制引用
王松,杜晓刚,王阳萍,杨景玉..基于关节点运动轨迹的人体动作识别[J].宁夏大学学报（自然科学版）,2017,38(2):147-152,6.基金项目
国家自然科学基金资助项目(61162016 （）
61562057) （）
甘肃省国际科技合作项目(144WCGA162) （144WCGA162）
甘肃省自然科学基金资助项目(145RJZA080) （145RJZA080）
兰州交通大学校青年基金资助项目(2013009,2013005) （2013009,2013005）
