局部分布信息增强的视觉单词描述与动作识别an-电子与信息学报.pdfVIP

第38 卷第3 期 电 子 与 信 息 学 报 Vol.38No.3 2016 年3 月 Journal of Electronics Information Technology Mar. 2016 局部分布信息增强的视觉单词描述与动作识别 张 良* 鲁梦梦 姜 华 (中国民航大学智能信号与图像处理天津市重点实验室 天津 300300) 摘 要：传统的单词包(Bag-Of-Words, BOW)算法由于缺少特征之间的分布信息容易造成动作混淆，并且单词包 大小的选择对识别结果具有较大影响。为了体现兴趣点的分布信息，该文在时空邻域内计算兴趣点之间的位置关系 作为其局部时空分布一致性特征，并提出了融合兴趣点表观特征的增强单词包算法，采用多类分类支持向量机 (Support Vector Machine, SVM)实现分类识别。分别针对单人和多人动作识别，在KTH 数据集和UT-interaction 数据集上进行实验。与传统单词包算法相比，增强单词包算法不仅提高了识别效率，而且削弱了单词包大小变化对 识别率的影响，实验结果验证了算法的有效性。 关键词：人体行为识别；局部分布特征；增强单词包模型；支持向量机 中图分类号：TP391 文献标识码：A 文章编号：1009-5896(2016)03-0549-08 DOI: 10.11999/JEIT150410 An Improved Scheme of Visual Words Description and Action Recognition Using Local Enhanced Distribution Information ZHANG Liang LU Mengmeng JIANG Hua (Key Laboratory of Advanced Signal and Image Processing, Civil Aviation University of China, Tianjin 300300, China) Abstract: The traditional Bag-Of-Words (BOW) model easy causes confusion of different action classes due to the lack of distribution information among features. And the size of BOW has a large effect on recognition rate. In order to reflect the distribution information of interesting points, the position relationship of interesting points in local spatio-temporal region is calculated as the consistency of distribution features. And the appearance features are fused to build the enhanced BOW model. SVM is adopted for multi-classes recognition. The experiment is carried out on KTH dataset for single person action recognition and UT-interaction dataset for multi-person abnormal action recognition. Compared with traditional BOW model, the enhanced BOW algorithm not only has a grea