基于LERBF算法的人体行为自相似识别  

Self Similarity Recognition of Human Behavior Based on LERBF Algorithm

作　　者：王志瑞[1] 刘正涛[1] 黄慧[1] 

机构地区：[1]三江学院计算机科学与工程学院,南京210012

出　　处：《控制工程》2016年第11期1784-1789,共6页Control Engineering of China

基　　金：江苏省科技支撑计划(BE2014388)

摘　　要：在当前人体动作行为识别过程中,一般侧重于对某一类动作行为的识别,算法的通用性和实用性均较差,对此基于局部证据RBF算法(Local Evidence RBF Algorithm,LERBF),在高层行为特征上实现对人体动作的自相似识别。首先,基于广义自相似时变特性,并结合光流场时空关注点,进行人体动作行为的特征局部提取,以此建立人体行为自相似局部描述特征矩阵;其次,基于SVM学习过程建立个体行为独立识别机制,并结合所提LERBF特征高层融合理论,建立起结构的分类优化机制,进而获得分类精度提高;最后,通过实验分析,LERBF具有90%左右的识别精度和最快的运算速度,验证所提识别算法具有更高的计算效率和精度。The current human action recognition process generally focuses on the identification of a class action, and the versatility and practicality are poor. This paper proposes a local evidence RBF algorithm （local evidence RBF algorithm based on LERBF）, implementing the self similarity recognition of human motion behavior on the high-level featuxe. Firstly, based on generalized self similarity characteristics, combined with the spatial and temporal concerns of optical flow field, local feature extraction of human behavior is performed, in order to establish the self similarity local description feature matrix of human behavior; Secondly, he independent individual behavior recognition mechanism is established based on SVM learning process, and combined with the LERBF feature high-level fusion theory, the structural classification optimization mechanism is established, then the classification accuracy is improved; Finally, by the experimental analysis, the LERBF has 90 % recognition accuracy and the fastest operation speed. The proposed algorithm has higher computational efficiency and accuracy.

关 键 词：局部特征 证据理论 神经网络 融合识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]