A real-time system for monitoring and classification of human falls on stairs using 2.4 GHz XBee3 micro modules with a tri-axial accelerometer and KNN algorithms

Abstract

In this paper, a monitoring and classification system for human activities on stairs is presented. The contribution of this work is that, first, we develop the real-time wireless sensor monitoring system for measuring human motion data using 2.4 GHz IEEE 802.15.4 XBee3 micro modules as the low-power wireless modules, where the GY-521 accelerometer sensor is attached. Here, human activities on stairs, including stair ascent, stair descent, turning around, and falling, are mainly focused on preventing any dangerous accidents. Second, using the measured data, the signal vector magnitude (SVM) calculation, signal filtering using an exponentially weighted moving average (EWMA), feature extraction using the mean, maximum, interquartile range (IQR), standard deviation (STDEV), variance, and peak-to-peak (PTP) amplitude, and classification using the K-nearest neighbors (KNN) algorithm are applied. Experiments have been conducted in a home scenario. Results indicate that the proposed system can efficiently monitor human activities on stairs in real-time with reliable communications, as indicated by a strong level of the received signal strength indicator (RSSI), and a packet delivery ratio (PDR) of 100 % for both line-of-sight (LoS) and non-line-of-sight (NLoS) communications. Additionally, the proposed system using only one variance feature and the KNN classifier provides classification accuracy of 89 % for stair ascent, 70 % for stair descent, 95 % for turning around, and 100 % for falling (a critical or focused event); 88 % on average results. Thus, our system, which includes devices and classification algorithms, has the potential to monitor and categorize human falls on stairs via wireless communication, and it can be applied in practical situations.