Research On Activity Recognition Based On MEMS Inertial Sensor

In the wake of science and technology,motion state detection and trajectory detection of objects have been widely applied in civilian fields and high-precision military in recent 10 years.MEMS inertial components have the advantage of small package size,light weight,low power consumption,low cost and high stability,which can overcome complex external conditions when using GPS and computer vision and image technology,in addition MEMS is independence and autonomy,not subject to any other equipment.In the meantime,with the increase of human income,people pay more and more attention to the applications of accelerometer sensors in human-machine interaction,health care and sports information.This paper mainly uses accelerometer to collect the acceleration data of daily motion of human body and studies the related motion after data-processing.Compared with other methods of human motion detection,such as computer vision,EMG sensor,the characteristics of acceleration sensor include: the external environment has little influence on it;the test procedure is simple;the way of raw data acquisition is free.In this paper,MEMS inertial sensor components are used in the following aspects:(1)An overview of research methods of recognition of human motion the state.Comparison of advantages and disadvantages among EMG sensor,acceleration sensor,computer vision and image,we think acceleration sensor is a good choice.And then advantages,related technologies and study on the implementation process when using the original data signal collected by the sensor of accelerator are elaborated.The basic principles of different classifiers are analyzed,and fuzzy neural network and extreme learning machine are selected as classifier to classify the motion.(2)Gait data is collected through three-axis acceleration sensor which is fixed on the ankle.In order to count steps,preprocessing including data filtering,integral envelope detection,acceleration threshold,are used.After that,the number of steps is transmitted by BLE 4.0 wireless transmission to the mobile phone and real-time display.The three cases of weight,thin,normal and fat of adult male who is about 170 cm were tested,and the average error of the step numbers were less than 1%.(3)Data of human motion including walk,upstairs,downstairs,running,standing is collected through three-axis acceleration sensor which is fixed on the waist.Before classification using extreme learning machine and fuzzy neural network respectively,preprocessing is essential procedure,including position correction,denoising,segmentation,feature extraction,zero velocity update.The influence of different training sets and sample length on the recognition rate when using the two algorithms is analyzed.The experiment shows that the recognition rate of the extreme learning machine and the fuzzy neural network is 90.5%,94% respectively.