Research On Human Activity Recognition Based On Triaxial Accelerometer

Human activity recognition is an important research topic in the field of artificial intelligence and pattern recognition.There are great theoretical significance and broad prospects for development in intelligent human-computer interaction, intelligent monitoring, health monitoring and human movement energy consumption assessment and other fields. With the acceleration sensor technology continues to progress and development, human activity recognition based on acceleration sensor receive more extensive attention and become a hotspot in research.In recent years, human activity recognition based on acceleration sensor has been gradually moving towards practical application, and has made great progress. But it is still in a period of comparative basis. Due to the fact that there are a lot of influence in the real environment and a variety of human activities, there are still many problems need to be solved in human activity recognition. Including how to reasonable design for the practical application of feature extraction and more effective feature selection methods, how to make the action recognition classifier classification accuracy high, low complexity, and has strong generalization ability, how to make recognition method of the classification and recognition performance is good. Based on the above problems, we mainly present three aspects of the research work:(1)We introduced and analyzed six human activity acceleration features, including the average value, standard deviation, kurtosis, skewness, interquartile range and the correlation coefficient between three axes on the feature extraction and selection. We train and identify the acceleration data by combining these six features.(2)We propose a method of classifier Weight Features Weight Bounds Weight Support Vector Machine(WFWBWSVM) based on the support vector machine(SVM), and the classification boundary of the classifier is weighted and combined with the weighted sum of the classification samples. Then we obtain the WFWBWSVM classifier which is more appropriate for human activity recognition.(3)We propose directed acyclic graph support vector machine(DAG SVM) optimization based on feature difference for classification and recognition by introducing SVM DAG. The optimization is according to the size of the classifier training samples set between feature differences. We distribute and arrange the classifiers and we get the optimal DAG SVM by the difference of feature.Finally, we verify the feasibility and effectiveness of the three aspects of the research work through experiments.