Research On Gesture Recognition Based On Support Vector Machine

Gesture recognition based on acceleration has been more valued than the methodbased on image because of its less dependance on environment equipment and otherexternal conditions. It has a lot of application in the human-computer interaction,According to the characteristics of hand signals, the method based on support vectormachine (SVM) has been adopted. SVM is a method based on the traditional statisticallearning theory. The SVM has a lot of advantages on solving the small sample, nonlinearand high dimensional pattern recognition problem. The classification alglrithm based onSVM can get a better classification accuracy.Firstly, the signal sensed by AMI602motion sensor is processed so that the data canbe applied to the model of SVM algorithm. In this paper, the process of the signalincluding the extraction of acceleration signal、data sorting、normalization processing anddata smoothing filter.Secondly, this paper do some research on the recognition algorithm based on SVMmodel, through the study and summary of classical binary classification algorithms, themethod has been promoted to multiple classification algorithm and then applies themultiple classification algorithm to the gesture recognition system. In this paper, the studyof multiple classification method mainly includes the one-against-all approach method,one-against-one approach method, directed acyclic graph method and the binary treemethod. According to the characteristics of the gestures acceleration signal collected bythe system and through the comparison of advantages and disadvantages of each algorithm,the multiple gesture classification, one-against-one approach method is adopted in thispaper.Finally, this paper do some optimization selection on some important parameters ofthe multiple classification model algorithm, the penalty factor C and the parameter g of thekernel function of radial basis function is optimized, through the validation and contrast ofthe grid search method、ga method and pso method, this paper choose the grid searchmethod.In this paper, the MATLAB software is used as the display platform,through the simulation the feasibility and the recognition rate of the algorithm is demonstrated, thegesture recognition of ten digits’ recognition accuracy is95.8%, the gesture recognition offive alphabets’ recognition accuracy is95.6%.