Research On Recognition Of The Human Forearm Motion Based On Acceleration Signal

As a kind of emerging natural way of human-computer interaction, the technology of human forearm motion recognition based on acceleration signal has widely attracted the attention of researchers, and shows great potential applications in lots of fields, such as the intelligent human-computer interaction, physical exercise, virtual reality,somatosensory games, intelligent control. Although the technology has achieved rapid development in recent years, there are still many technical difficulties to consider during the research of the forearm motion recognition based on the acceleration signal, that are how to design the acceleration signal data acquisition scheme of the forearm motion, how to design signal processing solutions, and how to achieve high-precision forearm motion recognition algorithm. In this paper, a series of research work is carried out around the above issues.Firstly, on the basis of consulting a large number of domestic and foreign literatures, the research background and significance of this topic are briefly summarized, and the analysis of research status at home and abroad is given. Then introduce the main methods of forearm motion recognition technology and determine the specific research content, and the general research scheme of forearm motion recognition is put forward.Secondly, the implementation of the forearm motion recognition algorithm is illustrated from the three aspects,which are data pre-processing, feature extraction and selection,classifier design, respectively. Combined with wavelet theory, the original acceleration signal preprocessing methods including the wavelet threshold de-noising method and normalization are introduced. For the three-dimensional acceleration signal after pretreatment, extracting time-domain feature values and energy information values in the three dimensions with wavelet packet decomposition to build the feature vector. At the last, for classifier design,choose the SVM classifier to achieve motion classification and the LIBSVM software as the implementation platform for classification.Finally, the specific experiment is designed to do validation and analysis on the given research scheme. To carry out the experiment, 10 test subjects are selected to involve in data collection, and seven kinds of forearm motion patterns are defined. To obtain the original data set for the research experiment, the data acquisition device was placed on the right forearm of every experiment object. Since the original data set can't be directly used for classification recognition, it is required to do preprocessing and extraction of characteristic parameters. The much better generalization capability of the SVM classifier is built based on the extracted feature vector. In order to get the optimal decision model of SVM classification, the optimal kernel function is determined by cross-validation approach and the optimal kernel parameters(C, g) is obtained by grid search algorithm. At the last, the cross-validation experiments are carried out on two different types of data sets named specified data set and non-specified data set. The experiment results are analyzed and discussed. And then the research plan of this topic is verified to be effective.