Multi-Approach Learning with Embedded Sensors Application in Gesture Recognition

Abstract

The increased attention to human daily activities in academic circles has proven highly valuable, serving various specific needs and producing desired outcomes across different fields. Evaluating human activity data opens up numerous possibilities for researchers, facilitating personalized support options such as timely stress interventions, real-time feedback mechanisms, and applications for assisting individuals with disabilities or monitoring mental health. This paper presents a comprehensive approach integrating multiple sensors to recognize human body movements, applicable to real-life scenarios such as classrooms, driving, and kitchen-related activities. Our focus is to enhance the precision of motion classification and improve motion classification rates by merging acceleration and rotation signals and analyzing an enhanced array of features using various high-caliber machine-learning models. This methodology achieves exceptional performance and flexibility, with accuracy rates ranging between 96% and 98%, substantiating activity recognition within diverse contexts. It aims to reduce system recognition errors, improve the classification process, and promote the advanced utilization of artificial intelligence algorithms in signal processing and in controlling and enhancing bionic hands.