Human Activity Recognition (HAR) refers to a set of techniques and systems aimed at identifying and classifying human activities based on data collected from the surrounding environment. This is achieved through the collection and processing of data from heterogeneous sensing sources such as wearable devices, ambient sensors, and wireless communication channels (e.g., Wi-Fi Channel State Information). In the context of Smart Environments, such as homes, public buildings, and transportation systems, HAR plays a fundamental role in enabling adaptive, user-aware services that support energy efficiency, comfort, safety, and healthcare.
The activities of the Net4U laboratory in this research area focus on several key aspects:
Multisensor Fusion for Human Monitoring
HAR systems can rely on heterogeneous data sources to increase recognition accuracy and contextual understanding. Wearable sensors (e.g., accelerometers, gyroscopes, heart rate monitors) provide high-resolution information about physical activity and physiological state. At the same time, ambient sensors (e.g., PIR, temperature, humidity) and signal-based features (e.g., CSI from Wi-Fi signals) capture the environmental and behavioural context in a non-intrusive way. The integration of these sources can enable the system to distinguish activities such as sitting, walking, cooking, or sleeping, even in complex or crowded environments.
Activity Profiling and Predictive Modelling
In Smart Buildings, HAR supports adaptive comfort management by aligning HVAC, lighting, and automation rules with users’ real activities rather than pre-defined schedules. Sensor data is used to construct dynamic user profiles that represent individual habits, routines, and preferences. These profiles, combined with contextual data (e.g., time, room occupancy, illumination), allow machine learning models to predict forthcoming activities and adapt the environment accordingly. In residential settings, such predictive capabilities support anticipatory home automation (e.g., pre-heating a room before waking).
In the broader context of Smart Cities, HAR systems can enable the real-time monitoring of crowds in areas of interest, such as train stations, airports, or public squares. These systems can estimate the number of passengers in public transport, as well as movement patterns, density dynamics, and even detect critical events like falls, sudden congestion, or people running. These solutions can leverage signal-based and device-free methods, enabling crowd monitoring without compromising users’ privacy.
Combined with wearable-based health monitoring, such systems pave the way for personalized and inclusive services in both private and public spaces.
Privacy-aware Recognition
Given the sensitive nature of behavioural data, privacy is a key concern in HAR systems. It is necessary to explore strategies to preserve privacy across all sensing modalities. In CSI-based recognition, data obfuscation techniques are evaluated to mitigate unauthorized inference. In wearable-based HAR, data minimization and local processing approaches are explored to reduce exposure. Moreover, lightweight encryption mechanisms are applied to sensor data in order to prevent unauthorized access while ensuring compatibility with resource-constrained devices. These approaches aim at striking a balance between data protection and energy efficiency, allowing HAR systems to operate reliably and ethically in sensitive contexts such as healthcare or assisted living.