Application of Kalman filter and k-NN classifier in wearable fall detection device

This paper presents the research and application of real-time fall detection system based on wearable data fusion. Firstly, it builds the activity model of elderly people based on attitude angles, designs and develops the sensor board integrated with three-axis accelerometer, gyroscope and Bluetooth to collect the activity data of the elderly in real time and send them to the smart mobile phone through Bluetooth. Secondly, it chooses attitude angles and acceleration signal vector magnitude as the features of fall detection, and accomplish the denoising treatment for attitude angles by Kalman filter. It uses the sliding window and k-NN algorithm to extract features and implements the system that can detect the fall of the elderly and give an alarm. Finally, the simulated experiment results provided by Weka show that the accuracy rate of the fall detection method is 95.8%, and the average sensitivity and average specificity of different activities are up to 95.8% and 99.2% respectively, which proves that the method has the feature of good real-time performance and high accuracy.