In the paper, we introduce a real-time hand gesture recognition method using a neural network. The underlying system is an automatic music display system which consists of three modules; feature extraction module, pattern classification module, and display control module. | Journal of Automation and Control Engineering Vol. 4, No. 2, April 2016 A Real-time Hand Gesture Recognition Technique and Its Application to Music Display System Jun-Yong Lee, Joong-Eun Jung, and Ho-Joon Kim Dept. of Computer Science and Electrical Engineering, Handing University, Pohang, South Korea Email: {leejunyong, jejung}@, hjkim@ Abstract—In the paper, we introduce a real-time hand gesture recognition method using a neural network. The underlying system is an automatic music display system which consists of three modules; feature extraction module, pattern classification module, and display control module. To reduce the computation time of the feature extraction process and the pattern classification process, a threedimensional data representation called motion history volume has been adopted. In addition, we propose a feature selection technique based on a modified fuzzy min-max neural network. We have defined a relevance factor which can measure the relevance of a feature to classify the specific pattern classes. The feature selection method can remove ineffective features and erroneous features in the learning data set by using the relevance factor data. network. However, in some problems such as image recognition, the learning data may include erroneous data or ineffective data because there may exist unexpected variations in the image data. They degrade the recognition rate and effectiveness of the learning process. In this research, we propose a methodology to select an effective feature set by extending the FMM neural network model. We have defined a relevance factor which measures the relevance of a feature with its pattern classes. The feature selection method removes ineffective and abnormal features from the learning data set by using the relevance factor data. The applicability of the music display system and its performance of recognition have been tested with the residual features. Index Terms—hand gesture recognition, .
