Human-Cognitive Buildings Interaction through Smart Gloves

Cognitive Buildings (CBs) exploit Internet of Things (IoT) devices, distributed intelligence, and adaptive services to improve comfort, safety, and quality of interaction with occupants. Within this perspective, gesture-driven human–machine systems are attractive because they enable natural and fast interaction without relying on fixed interfaces. This paper presents a smart-glove-based dynamic gesture recognition system for CBs. Using MANUS Quantum Metagloves as wearable IoT devices, a dataset from 26 participants performing 11 interaction-oriented activities has been collected. Two classical Machine Learning models, namely Support Vector Machine (SVM) and k-Nearest Neighbors (kNN), with two Deep Learning (DL) models, namely a Feedforward Neural Network (FNN) and a Long Short-Term Memory network (LSTM), have also been compared. The paper contributes: i) a CB-specific interaction vocabulary with a mapping from recognized gestures to building responses; ii) a complete acquisition and preprocessing pipeline for temporal smart-glove data; and iii) a comparative evaluation of classical and deep models for smart-glove-based human-machine systems. Results show a clear advantage of temporal DL, with LSTM reaching 98.25% accuracy, well above FNN (83.92%), kNN (59.67%), and SVM (53.14%). These findings support the adoption of sequence-aware learning for reliable smart-glove interaction in CBs.