No abstract

People often use non-linguistic communication methods such as physical gestures, which compose over 70% of our overall interaction with others. These physical gestures have regularity and repeatability as their common traits. This study is mainly discussing gesture recognition system for regular and repeated gestures. Torso PCA frame method is applied to angle transformation for such gesture recognition. In addition, a feature set is defined through extracting the patterns using envelope detection method and is applied to multi-layer perceptron for gesture recognition. For our experiment, skeletal structure was collected using Kinect, and 8 gestures were selected that people regularly use in real life. The recognition system was confirmed based on variety of people as our sample and the average accuracy was 89%.