摘要
提出了一种人工挑选关键帧的人体动作识别方法.先从标准视频中提取出能代表姿势的关键动作,然后对待测视频中每一帧图像中人体动作和关键动作比较分类来确定动作时间的相关度和相似度.实验结果表明人工挑选关键帧具有数量少和动作代表性强的特性,基于动作轮廓特征的人体动作识别方法在识别速度上比传统的方法快,识别率也较高.
A human action recognition method with artificial selection is presented .First, the key actions representing the posture are extracted from standard video key .then, the action in each image is compared and classified to define the correlation and similarity of action time .The experimental results show that the artificial selection of key frame has the characteristic of small quantity and strong representation .It also has the adoantage of high recognition speed and ratio .
出处
《兰州工业学院学报》
2014年第3期8-12,共5页
Journal of Lanzhou Institute of Technology
关键词
视频处理
机器学习
模式识别
HU矩
video process
machine learning
pattern recognition
Hu moment
K-Nearest Neighbour
