Research On Gesture Recognition Based On Wearable Data Gloves

Gesture recognition is an important field of artificial intelligence.It has Wide application prospects in virtual reality,deaf-mute communication,robot control,etc.Two major research areas in the field of gesture recognition are the problem of isolated gesture recognition and continuous gesture segmentation.Considering carrier of gesture recognition used,gesture recognition can be classified into based on vision,based on touch screen,based on sensor,based on wearable device.Gesture recognition based on visual is greatly affected by some factors such as illumination and environment,which limits the application of gesture recognition in different scenes.Ordinary visual images only include information of two-dimensional planes.Gesture recognition methods based on sensor can obtain three-dimensional spatial information including depth and acceleration information of hand motion by using different sensors,which provides new possibilities for the research of gesture recognition.This paper studies and summarizes the solutions to current problems in the field of gesture recognition,and introduces several classic gesture recognition algorithms based on the basic theory.Through the understanding of different algorithms applied to the field of gesture recognition,it is found that there are still some problems to be improved.In order to solve the corresponding problems,this paper proposes a method to solve the isolated gesture recognition and continuous gesture segmentation based on two algorithms.The main contributions of this paper are as follows:Firstly,in the process of pre-processing hand posture data,performance of clustering algorithm used in the discretization process is greatly affected by its parameters.This paper proposes a method called selecting state fuzzy C-means clustering algorithm.It can automatically determine the appropriate number of clusters and high-quality initial clustering centers in the gesture recognition.A more effective clustering results can effectively improve the performance of gesture recognition.The high-quality initial clustering center can speed up the convergence of clustering algorithms and avoid clustering results into local optimal solutions.Secondly,in the process of completing gesture.Due to the individual behavior difference and outliers,it may lead to a sparse transmit matrix of hidden Markov model in the process of gesture modeling.The problem of data sparseness is generated.Data sparseness can be understood as the number of states of a certain gesture is 0,and the state transition probability is 0,which causes the gesture recognition process to be sensitive to the influence of noise andother outliers.In order to solve the problem of data sparseness,this paper adopts the idea of data smoothing in natural language processing to improve the hidden Markov model.By adding an adaptive weak disturbance value to transmit matrix of Hidden Markov model,it is avoided to become a sparse matrix.Reduce the sensitivity of the model to outliers.Experiments show that improved model has a significant improvement in gesture recognition accuracy.Recall rate and accuracy are up to 93.88%? 96.92%.Finally,through the research of dynamic time warping algorithm,it is difficult to select a appropriate template that in the process of gesture recognition.Ordinary template is non-representation and usually too long.This paper proposes global template dynamic time warping algorithm to select a appropriate global template.Combined with knowledge of statistics,representative data of different defined gestures are selected to compose a global template.Combined with length characteristics of different defined gestures,a gesture segmentation method is proposed.Experiments show that improved algorithm has obvious improvement in accuracy and time efficiency.Recall rate and accuracy are up to 99.40%?95.60%.