Artificial Neural Network for Human Activity Recognition by Use of Smart Insoles

Human Activity Recognition (HAR) is an area with high interest. It can be used, for example, for medical purposes, for rehabilitation, for monitoring in sports as well as for prevention and monitoring of the elderly. The most used devices for this purpose are wearable devices. These devices have small sizes and can integrate different sensors inside. The only problem with these devices is that when using multiple devices at the same time, it can create an annoyance for the user. In addition, processing and understanding the data provided by such devices are sometimes not immediate and there is no standard rule that can be followed. In this study, a solution to HAR is presented by integrating a pair of smart insoles as a non-hindering device for the user and as a secondary purpose, there is the creation of a pipeline that can be simply recreated and extended to multiple subjects as needed. Smart insoles integrate pressure and inertia sensors inside. Alongside this device, an Artificial Neural Network model is developed to autonomously extract salient information directly from the raw data. Three subjects were included in the study. Each of them completed a series of activities among a well-defined set (fast walking, normal walking, slow walking, sitting, standing, downstairs, upstairs, and sit to stand). The results obtained with this solution achieved an average accuracy of 99.47%.