Static Gestures Recognition Based On KELM

Hand gesture has become an important way for people to communicate,so gesture recognition is becoming more and more important.Especially for the deaf and dumb groups,the progress of gesture recognition is related to whether they can rely on this technology to get closer to normal people.Gesture recognition can be divided into static gesture recognition and dynamic gesture recognition.Static gesture recognition is the basis of dynamic gesture recognition and dynamic gesture recognition is the extension of static gesture recognition.In practical situations,the study to dynamic gesture recognition is very difficult,so it is very important to study the static gesture recognition as the basis of dynamic gesture recognition system.Gesture recognition can also be divided into vision-based gesture recognition and sensor-based gesture recognition.According to practice,the sensor-based gesture recognition has a hand gesture identification limitation and its accuracy is not good enough and can be easily affected by the environment.Therefore vision-based gesture recognition is more significant than sensor-based gesture recognition.This paper deals with static hand gesture recognition based on vision.Aiming at the problem of static gesture recognition based on vision,a gesture recognition system based on improved ELM classifier is built.The main contributions of this paper are as follows:Gesture segmentation is the first step in a gesture recognition system and it's very important.Because the effect of segmentation affects the feature extraction of the next step,affects the accuracy of gesture recognition.In this paper,the common methods of hand gesture segmentation are discussed and the paper mainly focus on the gesture segmentation based on the skin color model,In order to minimize the effects of illumination and threshold selection on segmentation,the trained non-linear YCrCb color model is used for image segmentation to improve the accuracy and robustness of skin color model segmentation.On the basis of this background,in order to separate the human face from the hand gesture in the skin-like region and get the hand gesture,and eight-connected region labeling method is used.By filtering the complexity of the connected region several times,we can select the most complex gesture contour image.In this paper,the Hu invariant moments(feature selection)and PCA(dimensionality reduction)are adopted respectively.After the validation,the dimensionality reduction method PCA is selected and used in the feature extraction of the gesture recognition system.Among methods of the gesture classification,this paper deeply studies neural network algorithm,such as BP and ELM algorithm.Also,use two SLFNs classifiers including RBF neural network and ELM classifier to deal with gesture recognition.Among the gesture classification methods,the paper innovatively proposed the kernel limit learning algorithm(KELM)for the classification of hand gesture and achieved a better gesture recognition.Experimental data of small sets shows that the recognition accuracy rate is more than 80% or more,and experimental data of large sets shows the result is more than 90% or more.In the aspect of project implementation,an adaptive nonlinear threshold segmentation model based on linear YCrCb skin color model and a gesture recognition model based on KELM classifier are proposed in this paper.The simulation is carried out on multiple sets of data,and the version-based static gesture recognition is realized.This method can achieve a good recognition rate with number 0 to 8.