Research On Human Activity Recognition Based On Ultrasonic Array

With the global population growth slowing down,the proportion of the elderly is on the rise.Many elderly people have health problems when they live alone at home because of the lack of necessary supervision.The use of technology of HAR(Human Activity Recognition)to monitor the activity of elderly people living alone helps to grasp the living conditions of the elderly and prevent accidents.Based on single-person basic activity monitoring,this paper builds a HAR system based on the acoustic method with better privacy,and studies the key technologies.The system designed in this paper performs some typical basic activities(such as standing,sitting and lying postures).The hardware system transmits ultrasonic waves and receives echo information.Then,the received ultrasonic information is segmented and preprocessed to extract each active unit.In this paper,time-domain statistical analysis and frequency-domain analysis are used to obtain the characteristics of human activity from segmented data,and a method for constructing human activity data sets is designed.In addition,according to the different data characteristics,an empirical study was conducted to test the discriminability of these features.In this paper,first we built two traditional machine learning classification algorithms KNN and SVM for experiments,which have obtained relatively good recognition effect.Then,the defects of the machine learning algorithm were improved,and the CNN model was built and optimized to complete the identification of indoor single-person activity.The improved CNN algorithm has a recognition rate of 92.8% for single-frequency time-domain dataset and 97.5% for single-frequency frequency-domain dataset.The conclusions show that the optimized CNN algorithm has advantages in storage,calculation and accuracy.In order to evaluate the effectiveness of the ultrasonic array,a comparative study of the ultrasonic array and the single sensor and the effect of the number of ultrasonic array sensors on the accuracy were studied.It was found that the use of ultrasonic arrays as a means of acquisition not only makes data processing easier,but also significantly reduces algorithm complexity.