FIBiLS: Fall Detection of Healthy Elderly Using IMU Sensor and BiLSTM Model

Fall recovery refers to an individual’s ability to recover from external perturbation, and it is an obvious phenomenon in cluttered environments. Fall detection systems have evolved primarily for elderly individuals, who are more prone to falls that may lead to permanent disability or even death. In this article, a novel deep learning-based approach for early fall detection and monitoring individuals is proposed, while they perform daily tasks. The current Industry 4.0 revolution has witnessed the growing popularity of Internet of Things (IoT)-based solutions, with wearable sensing technology driving the use of wearable sensors for detecting such activities. This research utilizes dual wearable inertial measurement unit (IMU) sensors, worn on the body, to monitor elderly individuals and detect potential falls during daily activities. The SisFall dataset, a publicly available dataset of falls, is used to train the deep learning-based model. The dataset includes 14 distinct fall categories, which account for various directions and magnitudes of falls. Data are collected for experimental purposes to evaluate the fall detection capabilities of elderly individuals. The SisFall dataset includes two age groups: adults (18 years and older) and the elderly (60 years and older). The research utilizes a 1-D convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM) model to predict fall categories. The proposed FIBiLS (fall detecting using IMU sensor data and BiLSTM model) deep learning model demonstrates superior performance, achieving an accuracy of 99.68% with fast inference time. To facilitate edge computing, the method is implemented on a node MCU microcontroller board for fall detection. This approach outperforms previous research in both accuracy and complexity, providing better results with a more compact and less complex solution. This solution provides confidence to elderly individuals, enabling them to walk safely and independently.