Leveraging Hyperspectral Data in Cognitive Environments

Hyperspectral Imaging (HSI) captures data across multiple wavelengths of light, including visible and non-visible spectra (e.g., infrared or ultraviolet). It is usually used to analyze the composition, structure, and condition of objects of interest in a given environment through the spectral signature that, for a given material, represents its unique reflectance pattern of electromagnetic radiation across different wavelengths. Cognitive Environments (CEs) are intelligent systems designed to sense, interpret, and respond to context information within a physical space, enhancing user interactions through personalized and adaptive services. A key component in these environments is the ability to acquire and process rich and multidimensional data about dwellers and the environment itself, which is crucial for understanding and optimizing important aspects like comfort, safety, and users’ overall experience. HSI can be effectively exploited for retrieving rich contextual information from the environment in a non-invasive way. In particular, the ability to detect human skin and identify textile fabrics provides valuable insight into the clothing and physical state of a person, which in turn can be used to infer their activities. For example, distinguishing between exposed skin and covered areas can help estimate the type of clothing worn, offering clues about whether a person is exercising, resting, or engaged in specific tasks. In this paper, we propose an HSI-based framework for extracting representative spectral signatures of materials, with a focus on the recognition of human skin and textile fabric. A key strength of the approach lies in its use of standard and reproducible algorithms that operate directly on the observed scene, without requiring spectral references from controlled environments. The framework is validated using some publicly available hyperspectral datasets acquired with different cameras under various environmental conditions. Experimental results demonstrate that the extracted spectral signatures are robust and suitable for reliable matching and target detection despite the heterogeneity of the considered datasets. We also discuss key insights gained from the application of the framework on real data.