Ebook: Intelligent Environments 2025: Combined Workshop Proceedings

Intelligent Environments (IEs) represent a transformative approach to enhancing physical spaces by integrating cutting-edge technologies. These include information and communication systems, sensing and actuation, artificial intelligence, robotics, and human-computer interaction. These enhancements aim to enrich user experiences and facilitate effective environmental management while simultaneously showcasing the diverse capabilities of the underlying technologies.

The multidisciplinary research community dedicated to IEs delves into foundational concepts and core ideas across various domains. This exploration addresses significant challenges such as improving healthcare and well-being, ensuring system reliability and privacy, and integrating IEs with advanced computational models like deep learning and large language models.

The 21st International Conference on Intelligent Environments (IE 2025), held in Darmstadt, Germany, from 23 to 26 June 2025 featured several workshops that underscored the dynamic nature of this field:

∙ 14th International Workshop on Intelligent Environments Supporting Healthcare and Well-being (WISHWell’2025)

∙ 14th International Workshop on the Reliability of Intelligent Environments (WoRIE’25)

∙ 5th International Workshop on Artificial Intelligence and Machine Learning for Emerging Topics (ALLEGET’25)

These workshops attracted a substantial number of submissions, and following a rigorous peer-review process, only a select set of high-quality papers were accepted for presentation and publication in the conference proceedings.

The sustained success and relevance of the WISHWell and WoRIE workshops, now in their 14th editions, testify to their importance in the field.

The ALLEGET workshop provided a dynamic forum for exploring recent advances in deep learning and large language models within Intelligent Environments, reflecting the community’s ongoing innovation and adaptability.

In summary, the success of IE 2025, including the workshops, engaging tutorials, and demo sessions, was made possible by the organizers’ dedication and the authors’ outstanding contributions. Their commitment, time, and effort have ensured the continued excellence of the Intelligent Environments community, for which we are sincerely grateful.

Raúl AQUINO SANTOS

University of Colima, Mexico

Silvia FAQUIRI

RheinMain University of Applied Sciences, Germany

Information and communication technology (ICT) plays a crucial role in developing tools to enhance efficiency and reduce uncertainty in health-related procedures. The widespread adoption of sensors has facilitated the generation of large amounts of data that can be exploited with advanced data processing techniques. This research studies depth cameras (3D cameras) to assist health professionals in assessing frailty. It proposes using depth cameras to generate a dataset that can be used to train machine learning algorithms to automatically evaluate balance tests (side-by-side stance, semi-tandem stance, and tandem stance) for frailty assessment. Non-frail individuals participated in performing the three balance tests. Virtual reality lenses were used to induce imbalanced behaviours in the participants, generating data representing both balanced and imbalanced behaviours during the tests. The article presents the methodology that underpins the generation of this dataset, including the tools employed to validate the adequate performance of the cameras and support the labelling process. The dataset will be made publicly available upon completion of the labelling.

Generative artificial intelligence (AI), a rapidly evolving field of AI, holds transformative potential in mental health by addressing critical challenges such as accessibility, affordability, and personalization of care. Generative AI facilitates innovative solutions in illness screening, diagnosis, treatment, and psychoeducation. Despite these promises, integrating generative AI into mental health care poses significant risks, including misinformation, algorithmic bias, data privacy concerns, and a lack of regulatory oversight. This paper examines the opportunities and risks associated with generative AI in mental health and emphasizes the need for robust safeguards. Recommendations include implementing ethical design principles, developing clear regulatory frameworks, ensuring mental health professionals’ involvement, and prioritizing data privacy and security. By balancing innovation with caution, generative AI can advance mental health care responsibly and effectively.

Venous function assessments of the lower limb vascular bed are currently exclusively performed in medical offices and hospitals, requiring special equipment and trained personnel. Risk factors such as obesity, prolonged periods of sedentary activity, and older age, combined with rising life expectancy, are likely to contribute to a rising prevalence of deep venous insufficiencies. With potentially life-threatening secondary diseases, high treatment costs, and currently no curative therapy, these may become a growing burden on the health care system and public. Hence, an affordable monitoring system that is unobtrusive in everyday life, easy to use, and able to inform users and medical professionals about acute and prolonged negative deep venous health trends would be highly beneficial to counteract the aforementioned challenges. We therefore propose a novel lightweight algorithm that adapts the Light Reflection Rheography (LRR) using digital photoplethysmogram (dPPG) data collected wirelessly from a wearable sensor. Based on our col- lected test data of 18 subjects with 2 measurements taken per leg, our algorithm achieves an LRR segmentation performance of 86.15 % whilst being able to run on a smartphone. Integrated in our custom mobile app, our approach delivers a deep venous health progression control and critical early-stage information about pathological changes in the deep venous blood flow. Early medical results are consistent, comprehensible, and comparable with a medically performed LRR of one study subject with post-thrombotic deep vein damage, with a larger, more diverse follow-up study containing a medical ground truth planned for future work.

The increasing adoption of digital healthcare technologies is reshaping patient care, enabling intelligent environments that support both healthcare professionals and patients. However, challenges such as data heterogeneity, interoperability, security, and regulatory compliance hinder seamless integration into existing healthcare systems. This paper proposes a modular and scalable architecture for remote patient monitoring, designed to standardize data exchange, ensure secure communication, and enhance adaptability. The system leverages HL7 FHIR to facilitate interoperability across diverse healthcare platforms while incorporating privacy-preserving mechanisms, including encryption, authentication, and compliance with GDPR and HIPAA regulations. To validate its effectiveness, the proposed architecture is applied to cardiac surgery monitoring, integrating diverse devices and enabling personalized reporting to support patient care decisions.

Edge intelligence integrates edge computing and artificial intelligence for real-time data processing and decision-making at the data source, reducing latency and optimizing bandwidth for distributed IoT and smart device applications, empowering smarter, more efficient systems in real-world environments. However, this decentralized approach introduces significant privacy challenges, as sensitive information is processed across multiple, often resource-constrained, nodes. In this talk, we explore state-of-the-art privacy-preserving techniques within edge intelligence, with a focus on practical evaluations and real-world insights. We delve into methods such as differential privacy, secure multiparty computation, and homomorphic encryption, examining their effectiveness in safeguarding data while maintaining system performance. Through a series of empirical studies, we analyse the trade-offs between computational efficiency and privacy guarantees, highlighting the challenges and opportunities inherent in deploying these techniques in dynamic edge environments. Our findings underscore the need for a balanced approach that not only fortifies privacy but also supports the rapid processing demands of edge applications. Finally, we discuss emerging trends and propose a roadmap for future research to improve the performance of privacy preserving techniques with the operational constraints of edge intelligence. This talk will provide valuable perspectives for researchers and practitioners seeking to navigate the complexities of privacy in edge computing, ultimately fostering the development of more secure and efficient intelligent systems.

I have originally developed an indoor localization method called Visual-Geometric Matching (VGM), in which a line-segmented query image (visual) from an edge device is matched pixel-by-pixel with a line-segmented template image generated from an indoor building model (geometric) to determine the query image’s location. This method can be implemented as a server-client system, requiring only a monocular camera on the edge device, making it lightweight in terms of both weight and computational load. Through experiments, I have confirmed that a robot system integrating VGM with wheel odometry using an Extended Kalman Filter (EKF) successfully navigated an L-shaped corridor, ,repeatedly traveling back and forth for over an hour. The average localization error remained below 12 cm. The key advantage of this method is that it requires only a standard monocular camera for localization and does not rely on any additional physical infrastructure. This system holds great potential for various IoT applications.

With the increasing demand for real-time video processing in intelligent environments, optimising energy consumption while maintaining video quality remains a challenge. This paper presents a rule-based adaptive energy optimization framework for video compression, integrating dynamic decision-making techniques to regulate computational complexity based on system constraints. The proposed method employs an energy-aware loss function that dynamically adjusts key parameters based on inference conditions, real-time resource availability, and perceptual video quality. The model autonomously balances compression quality and energy efficiency by leveraging a rule-based approach, ensuring optimal trade-offs in resource-constrained devices. Experimental results demonstrate significant improvements in energy-aware video transmission, achieving adaptive complexity modulation with minimal loss in perceptual quality.

Intelligent Environments (IEs) are complex user-centric systems that integrate a mix of sensors, software, and algorithms designed to react, handle and retain awareness to provided contextual information, user preferences, reasoning processes, and even system malfunctions. Developing IEs requires a structured, iterative approach that ensures context-specific requirements and maintains system quality throughout the development lifecycle. Although robust frameworks and strategies now exist to assess the quality of the systems, during or after development, tools to support developers in these stages remain scarce. This study aims to identify, synthesise, and present the findings on tools available to improve the evaluation and quality management of IEs. A systematic literature review was conducted to explore the current landscape of research on the development and engineering of intelligent environments. The initial findings suggest new avenues for research, particularly the need for automation in evaluation processes to improve framework applicability, efficiency, and reliability. Addressing this gap should be a target for research endeavors.

Wireless communication plays an essential role in daily life and industry. For stable communication, pointing the antenna of the receiver in the optimal direction becomes crucial, especially in activity recognition through Wi-Fi sensing. In Wi-Fi sensing, receiver parameters such as antenna angles have traditionally been determined through calibration involving installed Wi-Fi transmitters and receivers. Knowing the received signal strength for each direction in advance would eliminate or minimize the need for calibration efforts. To address this problem, we propose Calibration-less Wi-Fi sensing with NeRF2 (CAWN), a method that estimates the optimal receiver antenna direction for Wi-Fi sensing by adapting NeRF2 to Wi-Fi environments. While NeRF2 has proven effective only for long-wavelength radio waves, we investigate its application to 5-GHz Wi-Fi short-wavelength radio waves and evaluates its effectiveness in activity recognition using Wi-Fi sensing.

Ambient Assisted Living (AAL) technologies support ageing populations, particularly those with early-stage dementia, but existing middleware is often too complex for resource-constrained home environments. This research introduces a lightweight, flexible middleware for AAL systems, optimised for limited-resource settings. Implemented at the SEArch and Smart Spaces Lab, Middlesex University, it integrates components like a real-time event reasoner, location detection system, and behaviour learning module, enhancing interoperability. The event-based architecture improves processing and data transformation within an existing system architecture. Tests using three simulated daily activities implemented in the Smart Spaces Lab at Middlesex University demonstrate the middleware’s efficiency and effective communication among system components.

The talk will wrap up the experiences from the R&D projects that were aimed at benefiting from applying machine learning to real business needs, with a particular focus on emerging topics. It will include the image processing and anomaly detection tasks performed using airborne and satellite imagery, as well as sensor and telemetry read-outs.

Patterns are everywhere and this is the problem of nowadays AI algorithms. This in particular occurs for AI algorithms which are targeted on Object recognition or autonomous driving. In such applications mostly artificial neuronal networks are being used, which produces the best results when it comes to pattern detection, but they are all unhinged without a proper semantic model which builds up a common context for the detected patterns to validate them without a ground truth reference. Only because a semantic model can validate weather a detected pattern fits into a context or not. This behavior is often called hallucination, because it relies on the same principles like a human on psychedelic drugs: a pattern recognition unconstrained from a context.

Japanese agriculture faces labor shortages, challenges in skill acquisition for new farmers, and a low food self-sufficiency rate. Agricultural digital transformation (DX) is essential for improving productivity and meeting consumer demands. To facilitate the early acquisition of harvest information and reduce labor requirements, this study proposes a method for measuring the vertical projected area of cabbage using semantic segmentation to extract image regions from camera images for crop growth prediction. The proposed method is validated using evaluation images.

Accurate bird sighting predictions enhance tourism planning, helping visitors optimize birdwatching in natural parks. This study focuses on the Natural Park of Las Lagunas de la Mata y Torrevieja (Alicante, Spain), where AI models were developed for bird sightings forecasts. The analysis showed that data granularity affects model performance: high-frequency datasets with noise are harder to predict due to the lack of clear seasonal patterns, whereas aggregated datasets with weekly or monthly intervals reveal more structured trends with reduced noise. In all scenarios, the hybrid CNNLSTM model consistently outperformed the other studied models. Additionally, a web augmenter was implemented to enhance the Wikipedia page for the park, offering interactive visualizations, including dynamic maps with daily bird sighting predictions, aiding tourists in planning their visits.