Evans,Guiomar Gaspar de AndradeBätge,NikolausFrancisco,Rodrigo Miguel2026-01-072026-01-072025http://hdl.handle.net/10400.5/116511Tese de Mestrado, Engenharia Física, 2025, Universidade de Lisboa, Faculdade de CiênciasActivity recognition plays a pivotal role in the evolution of wearable technologies and assistive devices, enabling them to adapt to user needs and improve mobility. Traditional methods often rely on multiple sensors or complex setups, limiting their practicality for everyday use. This dissertation introduces an approach to activity recognition utilizing Inertial Measurement Units (IMUs), specifically designed for seamless integration with an adaptive leg orthosis. The research employs a Raspberry Pi based system to classify activities such as walking, stair climbing, stair descending, ramp ascending, ramp descending, and standing in real-time using advanced artificial intelligence techniques. By addressing common challenges in existing solutions, such as real-time processing limitations, power inefficiency, and integration difficulties, this study provides a streamlined and cost-effective alternative. The orthosis in development, created in collaboration with Elysium Industries pretends to leverage fluidic muscles as a control mechanism, replacing traditional motors to enhance responsiveness, adaptability, and energy efficiency. Testing was conducted under varied conditions to validate the system’s accuracy and reliability, achieving classification performance comparable to multi sensor setups. Ultimately, this research emphasizes the feasibility of accurate activity recognition and control implementation using a single sensor integrated with AI. The proof of concept demonstrates the potential for developing more efficient and adaptable orthoses, which could be extended to other wearable assistive technologies, thereby enhancing their practicality and usability in real world applications.application/pdfengActivity recognitionInertial Measurement UnitLeg orthosisFluidic musclesmaster thesis204176255