The Study Of The Human Motion Data Recognition Based On Boosting Algorithm

Research on the recognition of human action and behavior based on wearable sensors is carried out,and the sensors are embedded into clothing to enable the elderly to receive health monitoring through wearing smart clothing.It is of great significance and broad application prospect for the health monitoring,fall prevention,rehabilitation and other aspects of the elderly.In human action recognition based on wearable sensors,most of the previous researches have focused on single type sensors and single classifiers.Because a small number of single type of sensors cannot capture the details of movement changes(such as joint parts),it brings trouble to the identification of complex body movements.However,increasing the number of sensors will increase the complexity of the device and bring inconvenience to the wearer.At the same time,the single classifier has obvious shortcomings in improving the accuracy of human action recognition and classification.In the face of these problems that need to be solved urgently in the study of human motion recognition,this paper adopts the combination of three-dimensional acceleration sensor and flexible sensor to collect the daily action data of the elderly according to the movement characteristics of the elderly,which can directly obtain the action data of the main body and the details of the movement changes of joint parts.The ensemble algorithm based on boosting framework to build the elderly action recognition model,analysis and comparison the difference of the integrated classifier and single classifier performance,and the performance difference of the several kinds of boosting algorithm model.This paper also improves the previous feature extraction methods.Compared with the wavelet transform to extract time-frequency features,the method of extracting time-domain features and associated features can quickly build the recognition model with high accuracy.Different from other feature dimension reduction method in the study of human action recognition,this article uses the linear discriminant analysis method to dimension reduction,found by experiment and compared with the principal component analysis method,linear discriminant analysis method in obtaining the lower dimensions can retain most of the original data information and recognition rate is higher than that of the principal component analysis method.When building the XGBoost integrated model,because of the deficiency of the automatic grid optimization method in finding the super parameters,the bayesian optimization algorithm was used instead of the automatic grid optimization method to find the optimal super parameter combination,and the experimental results showed that the action recognition rate was significantly improved.Through experimental analysis,it is concluded that the feature extraction method in this paper has great advantages over the previous methods,and the improved super-parameter tuning method can be used to build a better action recognition model.The research on the key technologies of motion recognition is of great practical value in the clinical management and health monitoring of the elderly population and in the rehabilitation treatment of people with motor dysfunction.