DYNAMIC HAND GESTURE RECOGNITION USING A 动态手势识别.pptVIP

DYNAMIC HAND GESTURE RECOGNITION USING A CNN MODEL WITH 3D RECEPTIVE FIELDS 指導教授：張財榮 學生：陳建宏 學號：M97G0209 出處：IEEE Int. Conference Neural Networks Signal Processing , 2008 作者：Ho-Joon Kim, Joseph S. Lee, and Jin-Hui Park * * OUTLINE * * Introduction The dynamic hand gesture recognition model Feature representation for hand gesture recognition An extended CNN model FMM-BASED feature analysis technique Experimental results Conclusion Introduction * * Human behavior recognition using computers It may not be easy to recognize the exact meaning of the dynamic hand gesture Utilize a CNN model with a receptive field extended to a three-dimensional structure Use a feature selection technique on the feature space using a weighted fuzzy min-max The dynamic hand gesture recognition model * * Feature representation for hand gesture recognition * * Skin color detection method Feature representation for hand gesture recognition * * Motion history volume An typical CNN model * * CNN (Convolutional Neural Network) An extended CNN model * * An extended CNN model * * Feature maps(5*5*5) CNN An extended CNN model * * Two similar feature maps generated from two different input data FMM-BASED feature analysis technique * * 傳統集合 模糊集合 FMM-BASED feature analysis technique * * FMM(Fuzzy min-max) FMM-BASED feature analysis technique * * Experimental results * * Using 125 features Experimental results * * Experimental results * * Using 63 features Conclusion * * The proposed system consists of two modules: CNN (Convolutional Neural Network) FMM(Fuzzy min-max) In selecting an effective feature to reduce dimensionality ,the technique improves the computation speed and robustness to noise.