Abstract: Hand Position Detection is crucial in real-time interactive virtual reality applications. This study presented a new method to detect a wide range of human hand movement efficiently. Camera array system was used instead of single or binocular vision system to monitor the whole hand existence region. In order to extract stable hand information from video sequence, our method transformed pictures color space from RGB (Red, Green and Blue) to Lab (Lightness, color position between red and green, color position between yellow and blue) and the image histogram was used as the feature parameter of video sequence. Then the differences of single video frequency feature parameter and global feature parameter were putted into neural network. Through neural network training and machine learning the target hand position in three-dimensional space could be calculated. Experimental results showed that this method could detect and locate hand efficiently.