计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于DTW与混合判别特征检测器的手势识别

黄振翔a,彭 波a,吴 娟a,王儒朋b   

  1. (中国农业大学 a. 信息与电气工程学院;b. 工学院,北京 100083)
  • 收稿日期:2013-04-23 出版日期:2014-05-15 发布日期:2014-05-14
  • 作者简介:黄振翔(1987-),男,硕士研究生,主研方向:机器视觉,数字图像处理;彭 波,教授、博士、博士生导师;吴 娟、王儒朋,硕士研究生。

Gesture Recognition Based on DTW and Combined Discriminative Feature Detector

HUANG Zhen-xiang a, PENG Bo a, WU Juan a, WANG Ru-peng b   

  1. (a. College of Information and Electrical Engineering; b. College of Engineering, China Agricultural University, Beijing 100083, China)
  • Received:2013-04-23 Online:2014-05-15 Published:2014-05-14

摘要: 在动态手势识别领域,动态时间规整(DTW)算法在消除不同时空表示模式之间的时间差异方面具有优势,但作为一种模板匹配算法,受限于样本库的容量大小并且缺乏统计模型框架训练,其识别效果和稳定性较差,尤其在大数据量、复杂手势和组合手势的情况下。针对上述不足,提出一种基于DTW和混合判别特征检测器(CFDF)的手势识别算法。利用DTW只对手势信号在时域进行规整,通过CFDF将手势特征的概率分布转换成二值的分段线性函数,根据允许的偏差范围分别做归0或归1处理后,再进行二次分类。实验结果表明,该算法通过舍弃无辨识度特征有效地降低了维度和噪声,手势平均识别率可达91.2%,比单独采用DTW的识别算法提高了6.0%。

关键词: 手势识别, 动态时间规整, 隐马尔可夫模型, 归一化, 统计模型, 混合判别特征检测器

Abstract: In the dynamic gesture recognition field, the Dynamic Time Warping(DTW) algorithm, which has advantage in eliminating time differences between different space-time expression modes, is a template matching algorithm in essence, so its performance is limited by the capacity of the sample database and lacking statistical model framework to train. Its recognition result is not satisfactory and stability is poor, especially in the cases of large amount of data, complex gestures and combined gestures. In response to these deficiencies, this paper proposes a gesture recognition algorithm based on DTW and Combined Discriminative Feature Detector(CDFD). It warps gesture signals in the time domain only, uses combined discriminative feature detectors to transform probability distribution of gesture features to binary piecewise linear function and makes zero or one according to the permissible deviation ranges, finally classifies gestures. Experimental results show that this algorithm can discard non-discriminative features to reduce dimensionality and noise, and the gesture average recognition rate reaches 91.2%. Compared with individual DTW algorithm, gesture recognition rate increases by 6.0%.

Key words: gesture recognition, Dynamic Time Warping(DTW), Hidden Markov Model(HMM), normalization, statistical model, Combined Discriminative Feature Detector(CDFD)

中图分类号: