Hand Gesture Recognition Based On Feature Fusion And Partial Least Squares Dimensionality Reduction

With the development of and computer vision technology and theemergence of the new type of human-computer interaction, gesturerecognition based on vision has caused extensive attention of the scholars.How can recognize the kinds of the gesture samples in real-time andaccurately has become the current focus of the research. On the basis of theprevious work, through taking the improvement of the gesture recognitionrate under the condition of taking the real-time into consideration as ourgoal and utilizing all the aspects of the theories comprehensively, we studythe approach based on feature fusion and partial least squares dimensionalityreduction to solve gesture recognition.Firstly, this paper summarizes the general process of gesture recognitionand then analysis the technologies of gesture segmentation, feature extractionand gesture recognition which are the process of gesture recognition inparticular.Secondly, propose a novel approach based on feature fusion and partialleast squares dimensionality reduction to solve gesture recognition based on analysis and study gesture recognition thoroughly. With our method, thegesture sample’s features of histograms of oriented gradients and local binarypatterns are extracted based on gesture segmentation and then would becombined to be the combined features of the gesture sample. Under thepromise of the recognition rate can not change, we adopt the partial leastsquares method to reduce the dimension of the gesture samples’ combinedfeatures for improving the performance of real-time, and then could get thelow combined features of the gesture samples after dimensionality reduction.Thirdly, the method of gesture training and recognition based on thesupport vector machine classifier is given. Train the low combined features ofthe train samples after reducing the dimension to get a gesture recognizer byusing this method and recognize the test samples using the recognizer,calculate the recognition rate of gesture recognition through adding up all therecognition results of the test samples.Finally, conduct the experiences to test our method based on the JochenTriesch and self-made gesture database, and compare the performance ofrecognition rate and real-time between our method and the existing gesturerecognition method.