| The problem of aging is aggravated,and falls have become the leading cause of accidental death for the elderly.The cause of death is not only the fall itself,but the more direct cause is that it cannot be rescued in time after the fall.In recent years,products for the detection of falls of the elderly have been continuously introduced,but they all have problems of large size,high power consumption,and low detection accuracy.Therefore,the development of a wearable,low-power,high-precision and highly interactive fall detection device has important social significance and practical value for the safety and health of the elderly.This paper has developed a fall detection system that can be worn on the waist.The system consists of a fall detection device and mobile phone monitoring software.The motion data of the waist is collected by the nine-axis acceleration sensor,and the fall is judged by the algorithm.If the fall is detected,the alarm information and the location of the fall can be sent automatically.After receiving the alarm information,the guardian can call the falling device to determine the safety status of the elderly and locate the falling position of the elderly.Firstly,through the research of low-power hardware design and reasonable PCB layout,a small-scale,low-power fall detection hardware system is designed.Secondly,based on the research of MPU9250 interrupt wake-up and software optimization,a system low-power operation strategy combining sleep wake-up mechanism and power amplifier circuit control is proposed.Then,in order to improve the signal quality of GPS and GSM antennas,antenna simulation software and a network analyzer are used to determine the antenna network parameters to achieve high-precision positioning and high-quality transmission link.Finally,through the analysis and research on the characteristic information of daily activities and falls,two traditional fall threshold methods and Adaboost fall detection algorithms based on single-layer decision tree are proposed.Considering the limitations of the hardware system and the complexity of the algorithm,the traditional threshold method is finally transplanted to the embedded system to realize the real-time monitoring of falls.After measurement,the hardware size is 5.8cm*3.4cm,the weight is 48g,the average power consumption is 7.3mAh,and it can be used for 4 days at full power,which fully meets the requirements of wearability.For the real data in the database,the threshold detection algorithm and the machine learning algorithm proposed in this paper have a fall detection accuracy rate of 100%,and the simulated fall data for 12 volunteers have an accuracy rate of 97.73%and 99.08%,respectively. |
