Publications

This paper presents an offline speech-driven semantic input module born to support an indoor navigation application in the context of Sapienza University of Rome. The system addresses the difficulty international students face when searching for rooms, as Italian names are often mispronounced and incorrectly transcribed by standard speech-to-text (STT) systems. To operate in buildings with limited connectivity, the solution relies entirely on on-device processing and does not require cloud services or dedicated speech recognition models. Spoken queries are normalized through domain-specific vocabulary correction and spoken number extraction before being matched against a local repository of 243 rooms across four campus buildings. The matching engine combines Levenshtein similarity with an Italian-specific phonetic representation to improve robustness against transcription errors and pronunciation variation. Evaluation on 150 spoken queries achieved 92% top-candidate accuracy, with an additional 6% correctly resolved through disambiguation. The complete resolution pipeline executes in 8.0 ms on consumer mobile hardware, demonstrating the feasibility of lightweight, real-time, offline voice search to support indoor navigation applications.

Abstract under CC BY-NC-ND 4.0, by the authors.

Accurately estimating indoor occupancy is fundamental to the development of modern smart buildings, which aim to optimize critical parameters such as Heating, Ventilation, and Air Conditioning (HVAC) control, safety, and resource management in real time to reduce energy waste. Traditional sensing approaches, including cameras, Passive Infrared (PIR) sensors, and CO₂ monitors, often encounter high deployment costs, maintenance overhead, and significant privacy concerns, particularly under General Data Protection Regulation (GDPR) regulations. This paper presents the design, implementation, and evaluation of a non-invasive occupancy estimation system that exclusively relies on Bluetooth Low Energy (BLE) scans performed via a mobile device, eliminating the need for prior structural information or dedicated sensing infrastructure. The proposed method analyzes statistical differences between various university environments, such as Laboratories, Classrooms, and Corridors, while integrating variables such as the number of fixed and mobile devices, the average device density per person, and the interference caused by signal bleed-through between adjacent rooms.

Based on a foundational study of device ownership behavior, we develop context-dependent calibration coefficients to address the multi-device phenomenon, in which a single occupant may carry multiple Bluetooth Low Energy emitters. Our system utilizes a three-layer architecture that includes Passive Bluetooth scanning, Signal filtering with night-baseline infrastructure detection, and Automatic room-type classification. This design allows for the dynamic selection of estimation parameters without the need for manual input. Field experiments conducted across various university spaces over a multi-week data collection period demonstrate that our context-aware model significantly reduces estimation error compared to traditional device-counting methods. This approach offers a scalable, cost-effective, and privacy-preserving engineering solution for real-time occupancy monitoring in smart campus environments.

Abstract under CC BY-NC-ND 4.0, by the authors.

This poster presents an interactive mobile interface that helps users describe their indoor surroundings, enabling automatic AI-based localization on evacuation maps. The interface prompts users to select the environmental elements they see around them from a set of icons commonly used in evacuation maps, such as elevators, fire extinguishers, alarm buttons, doors, and corridor shapes. As selections are made, the pool of possible locations narrows, and incompatible icons are removed from the set. The approach builds on recent work in which users described their surroundings through open-ended conversations, often including irrelevant or unexpected information, resulting in poor localization results. As a first step in assessing the efficacy of the proposed interface, a user study (N = 16) was conducted across multiple locations within a university building to investigate the impact of the interface design on usability and the overall localization process. This work aims to contribute to ongoing investigations of localization via human collaboration by proposing an interface that reduces the cognitive effort required to provide significant information to the localization backend.

Abstract under CC BY-NC-ND 4.0, by the authors.