Research Of Fall Detection Algorithm Based On Deep Neural Network And System Implementation

With the increasing trend of aging population in China,health protection for the elderly has become the focus of current research.The current wearable fall detection system is mainly composed of a single sensor or multiple sensors to collect different information of the human body,which is transmitted to the PC through the combination of Bluetooth,Wi-Fi,Zig Bee and other short distance communication and GPRS/GSM remote communication.Then fall detection algorithms such as threshold method,CNN,SVM and random forest are used on PC.The Narrow Band Internet of Things(NB-IOT),which has been developing rapidly since its birth,provides a new solution for remote data communication.Therefore,based on NB-IOT technology and cloud platform,this paper designs a wearable fall detection system through Gate Recurrent Unit(GRU)algorithm for fall detection.The main research contents of this paper are as follows:(1)Research on fall detection method based on GRU algorithm.Firstly,the process of human falling is analyzed,and the characteristics of human falling are extracted.And compared with the human body’s daily activities.The GRU algorithm is used to train the traditional fall detection model because the time constraint characteristics are not used much.Aiming at the situation that the hyperparameters of GRU algorithm are not easy to train and the accuracy is not high,Particle Swarm Optimizer(PSO)is used to optimize the hyperparameters of GRU algorithm,and a better model is obtained.In order to improve the PSO algorithm by adjusting flight inertia and introducing learning factors,a better detection model was obtained.(2)Research on fall detection algorithm based on SFM algorithm.In order to dig deeper frequency domain information of human daily behavior activities,improve the accuracy of fall detection algorithm.In this paper,THE SFM model based on PSO optimization is used,which can not only retain the time domain information of the mining data,but also capture the frequency domain information of the data and learn it.Experimental verification shows that the accuracy of this model is higher than that of the improved PSO-GRU model.(3)Design and implementation of prototype system for fall detection.According to the functional requirements of the fall detection system,the prototype of the fall detection system was designed,the hardware circuit of the lower computer was designed and the driver was written,and the functions of the upper computer login authentication,message queue data transmission,algorithm model call detection,command delivery,historical data display and data storage were completed.Finally,the prototype system of fall detection designed was tested,including hardware uploading data to the cloud platform,data transfer from the cloud platform to the upper computer,detection and display of human body posture by the upper computer,command delivery by the upper computer,command delivery by the lower computer and alarm,data storage by the upper computer,etc.Verify the effectiveness of the prototype fall detection system.