Design and implementation of fall detection system based on PCA-ANN

In view of low accuracy and poor privacy protection in current wearable fall detection systems, we had designed and implemented a fall detection system with high precision and low delay. First, the Arduino Nano 33 BLE development board as the main control component of the detection device is used. A fall detection algorithm (PCA-ANN) based on principal component analysis combined with artificial neural network is designed and implemented. It uses a lightweight open-source machine learning framework TensorFlow Lite, which is specifically fit for IoT scenarios. Secondly, we trained and transformed the model based on the fall data set published online publicly using TensorFlow framework and deployed the model to the embedded fall detection terminal. Finally, the fall detection device was used to carry out the actual fall detection experiment on the volunteers The experimental results show that the fall detection accuracy of the system is 99.04%, the sensitive is 97.57%, and the specificity is 99.58%. The system uses edge computing technology to complete the task of running deep learning on embedded devices with limited computing power and storage units, and creatively applies this technology to wearable fall detection devices, which provides valuable data for subsequent research. Compared with the existing fall detection systems that require data transmission to cloud, this system has the characteristics of low latency and high accuracy. Besides, this system also eliminates hidden dangers in user privacy, which is suitable for elderly people to wear.