基于支持向量机的多种人体姿态识别.pdf

第23 卷 第3期 重 庆 工 学 院 学 报( 自然科学) 2009 年3 月 Vol.23 No. 3 Journal of Chongqing Institute of Technology(Natural Science) Mar. 2009 谢 非, 徐贵力 ( , 210016) : , ., , , 2 , RBF . ,SVM , RBF , RBF ,SVM . , 96% , , 0. 0172S, . : ; ; :TP391. 4 :A : 1671- 0924( 2009)03- 0138- 06 Recognition of Multiple Human Body Postures Based on SVM XI Fei, XU Guili ( Collage of Automation ngineering, Nanjing University of Aeronautics Astronautics,Nanjing 210016,China) Abstract: In order to realize intelligent surveillance in some places such as sentry and so on, this paper present an approach to multiple human body posture recognition based on Support Vector Machine. Having built the sample library of all researching object, shape complexity, rectangular degree, and ratio of width to height etc. are extracted using SVM as classifier between two kinds and multiple kinds of objects. SVM has global optimality and a good generalization ability. Using the RBF neural network as classifier, SVM classifier is contrasted with RBF neural network classifier. The experimental results show that SVM doesn t need network iteration and the SVM classifier has a faster classification speed and a good recognition ratio than RBF neural network. The method presented has a good recognition ratio, which reaches 96%, and a fast recognition speed, which reaches 0.017 2 s averagely. The method provides a theoretical and technical basis for two or multiple classification. Key words: human body posture recognition; Support Vector Machine; intelligent surveillance :2008- 12- 26 : (