A novel competing convolutional kernels method to CSI-based fall detection for disabled people

Abstract

Recent advancements in wearable Internet of Things(IoT) are enabling real-time health monitoring of the physically disabled and elderly people. However, there are several challenges associated with wearable IoT devices, such as inconvenience to the patients, their obtrusive nature, and the need for technological literacy. Addressing these issues, we propose a novel competing convolutional kernels-based method for CSI-based fall detection. Our proposed model is a time series classification that uses dictionary-based methods and convolutional kernel transformations. Random convolutional kernels are organised into groups, and each one of them is treated as a dictionary of patterns to count kernel outputs that match the input series. To ensure high accuracy and good efficiency, hard and soft counting methods are used along with first-order differences. We tested the proposed model on a publicly available dataset and compared the performance with the existing methods. The proposed method outperforms existing state-of-the-art methods for CSI-based fall detection. Our model achieves an accuracy of 98% and an F1-score of 89%, which is 9% higher than the state-of-the-art.