Research On BAs And PTs Motion Recognition Algorithm And Its Application

Human motion recognition is a method of understanding human behavior through interpreting attributes from motion,position,physiological signals,and environmental information.It is an important research topic in the field of pattern recognition and intelligent control.And the study of human motion recognition theory is of great significance and has wide application prospects.In this paper,theoretical and applied researches are carried out for the needs of rapid and accurate identification of human movements in the fields of medical monitoring and behavioral assistance.Firstly,a data acquisition and motion recognition experiment platform based on sensors and Bluetooth communication is built.The main control chip is an SOC chip nRF51822 that can realize ultra-low power Bluetooth 4.0.The accelerometer is a MEMS sensor LIS3 DSH,it is embedded with two independent programmable state machines.Secondly,motion feature extraction and classification methods are studied.For the collected basic activities(BAs)and postural transitions(PTs)data,interquartile range(IQR),kurtosis,skewness,wavelet packet entropy and other feature extraction methods are studied.The effects of several classification methods such as support vector machine(SVM),probabilistic neural network(PNN)and random forest(RF)were compared.And then,genetic algorithm is used to optimize the parameters of the random forest classification method.The generalized error of the OOB data estimation RF is used as a fitness function,and the optimal parameter combination is searched by performing genetic operations such as selection,crossover,and mutation.This reduces the error in the feature space,improves the recognition effect,and the overall recognition accuracy reaches95.36%.Finally,a feature extraction method based on EEMD combined with multivariate multiscale entropy is proposed for the PTs of difficult recognition.Through the analysis and experiment of four typical PTs,the advancement of the proposed method is verified.Based on the above,the motion recognition method based on sensor embedded state machine is also studied.On the experimental platform,the simulation and experimental study of the state machine programming were carried out by taking the true and false fallaction as an example.It was found that the motion recognition based on the state machine has high speed and good real-time performance,but the accuracy of the motion recognition depends on the state machine programming of the specific motion.