Measurement of Office Activities Using Non-Contact Ultrasonic Sensors and Accelerometer: Accuracy Comparison and Optimal Use Cases

Human Activity Recognition (HAR) plays a crucial role in optimizing indoor environmental conditions, improving occupant comfort, and enhancing energy efficiency in smart buildings. Various sensing technologies, including ultrasonic (US) sensors and accelerometers, are commonly used for HAR, each offering distinct advantages. US sensors provide non-intrusive monitoring while preserving privacy, whereas accelerometers enable personalized activity tracking but require user compliance. This study presents a comparative analysis of HAR performance using both technologies, focusing on their accuracy, reliability, and optimal application scenarios. The experimental setup included a US sensor and a accelerometer to capture movement patterns of office activities. A machine learning (ML) framework, combining Support Vector Machines (SVM) and Convolutional Neural Networks (CNN), was implemented to classify activities. Results show that the hybrid SVM-CNN model achieved 99.3 % accuracy with US sensors and 97.0 % accuracy with accelerometers, demonstrating the effectiveness of ML techniques in HAR. Findings indicate that US sensors excel in differentiating fine-grained movements, making them ideal for Personalized Comfort Models (PCMs) and energy-efficient HVAC systems, while accelerometers are better suited for individual activity tracking. The study provides guidelines for optimal sensor deployment in indoor environments, contributing to the development of smart, adaptive buildings.