摘要
提出了一种基于Hu矩的以支持向量机(Support Vector Machine)为基分类器的Tri-Training分类器。首先采用背景差分法得到运动人体轮廓,然后通过Hu矩的特征提取方式,对运动人体轮廓进行特征提取,并对提取的数据集经过数据清洗和归一化处理后,使用以SVM为基分类器的Tri-Training分类器进行训练和分类来实现人的身份识别。实验结果表明,这种新的学习方法,能较好地弥补以往很多学习方法在小样本、非线性、过学习、高维数等问题上的不足,在人体行为分析智能监控系统中具有很强的推广能力。
Aim. The introduction of the full paper discusses some relevant matters and then proposes the method mentioned in the title, which we believe is new and better and which is fully explained in sections 1 through 3. Their core consists of： ＂We introduced a kind of tri-training classifier using Hu moments and SVM （Support Vector Machine） classifier. First we obtained sports silhouette by background subtraction method; then through the Hu moment feature extraction method, we obtained sports body contour feature extraction; then we obtained the extraction of data sets through data cleansing and normalization; finally, through using tri-training classifier, we obtained the person identity. Section 4 analyzed the simulation results of the human behavior of a particular example ; the results of analysis, presented in Table 5, show preliminarily that our new method is indeed better.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2011年第6期871-876,共6页
Journal of Northwestern Polytechnical University
关键词
行为分析
协同训练
不变矩
TRI-TRAINING
analysis, classification （of information）, errors, feature extraction, information technology, invari-ane, learning systems, nonlinear systems, simulation, sampling
activity analysis, tri-training
