基于DTW约束的动作行为识别  

Action Recognition Based on Constraintive DTW

作　　者：李海涛[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《计算机仿真》2014年第11期227-230,共4页Computer Simulation

摘　　要：针对固定视角下的人体动作行为准确识别的研究,通过姿态关节点标定,结合不同关节点对动作内和动作间起到不等的权重区分度的问题,提出一种加权量化矩阵的姿态特征表示方法,将关节点相对坐标经过权重修正作为姿态特征,减少因关节点标定误差或图像帧的缺失导致的识别精度下降和误判率增加等问题,同时,对DTW算法加入动态时间规整加以约束来降低算法的复杂度,最后在家庭智能空间内获取动作姿态图像序列,通过实验验证关节点标定及行为识别算法的有效性。According to the research of human action recognition in fixed angle, through the point calibration of the postures, combined with weight discrimination of different joint to distinguish actions, a method of posture feature attitude features is proposed based on a weighted quantization matrix. It reduces some questions of recognition accuracy and misjustice due to the lackness of joint point or image frame, then reduces the complexity of the DTW algorithm with dynamic time warping restraint. Finally, some gesture image sequences are obtained from the smart home space, and some simple action recognitions are completed with improved DTW algorithm.

关 键 词：关节点标定 姿态特征 加权量化矩阵 动态时间规划 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]