Human Behavior Recognition Algorithms Based on Neural Network

• Main Authors: Yan-Jun Chen, 陳彥君
• Other Authors: Jeng-Shyang Pan
• Format: Others
• Language: zh-TW
• Published: 101
• Online Access: http://ndltd.ncl.edu.tw/handle/26376583697781126759

碩士 === 國立高雄應用科技大學 === 電子工程系 === 100 === This thesis suggested that there are three main parts in the algorithm structure inclnding Feature Extraction , Behavior Recognition, and Moving Object Detection, we used the KTH human behavior database for experiment. The database included six motions—boxing, hand–clapping, hand-waving, running, jogging, walking—and 25 people in four scenarios and Weizmann database. Firstly, used the method of Moving Object Detection is need to detect continuous images, and then the detected images exclusive or compressed into binary sequence compressed images to show the traced traits in their outlines. Finally, the Artificial Neural Network, using the the Back-Propagation Network along with the training of Scaled Conjugate Gradient Algorithm to speed up the convergence rate. Conparecl with the tracutional methods, the recognition rate is improved.