Veneruso, Paolo (2025) Multi-sensor-based Enhanced Vision techniques to support approach and landing operations in Urban Air Mobility scenarios. [Tesi di dottorato]

	Documento PDF Paolo_Veneruso_37_COMPLETO.pdf Visibile a [TBR] Amministratori dell'archivio Download (7MB) | Richiedi una copia
	Documento PDF Paolo_Veneruso_37_PARZIALE.pdf Download (493kB)

Abstract

Approach and landing operations are among the most critical phases in aviation, accounting for a significant proportion of accidents due to challenges as low visibility conditions. These challenges are amplified in the emerging Urban Air Mobility (UAM) context, where Vertical Take-Off and Landing (VTOL) aircraft operate in densely populated urban environments, often with limited space and closer to potential obstacles. Existing navigation systems, such as the Instrument Landing System (ILS) and GNSS augmentation systems, are ill-suited to these scenarios due to their reliance on ground-based infrastructure, susceptibility to signal degradation in urban areas, or limitations in supporting variable approach trajectories. These factors necessitate the development of innovative onboard navigation solutions that minimize infrastructure dependency while ensuring robust, weather-resilient, and safe operations. To address these challenges, this dissertation proposes a novel multi-sensor navigation architecture specifically designed for UAM approach and landing operations. The architecture integrates a GNSS receiver, an Inertial Measurement Unit (IMU), two visual cameras and an Frequency Modulated Continuous Wave (FMCW) radar. Each sensor complements the others, enabling a navigation solution capable of providing accurate state estimation throughout the approach and landing phases. The radar and cameras design prioritizes compatibility with the operational and regulatory constraints defined by international aviation authorities, such as European Union Aviation Safety Agency (EASA), including the integration of enhanced visual and radar-aiding navigation cues tailored to specific vertiport infrastructure. The vision system leverages advanced computer vision techniques, including Convolutional Neural Network (CNN) for landing pattern detection and image processing techniques for tracking of the area of interest and keypoints extraction. The visual measurements are validated through integrity monitoring before being incorporated into a loosely coupled Extended Kalman Filter (EKF), ensuring precise pose estimation during critical flight phases. Meanwhile, the radar system complements the vision-based systems providing resilience to adverse visibility conditions and higher operative ranges. Radar data processing involves targets detection and different matching algorithms exploiting the navigation state prediction of the EKF and the known position of the targets on the vertiport, enabling early and reliable detection of the landing area. The radar data is tightly integrated into the EKF, allowing the filter to extract navigation information even from limited number of targets correctly matched. The performance of the proposed architecture is assessed through high-fidelity simulations and experimental validation. Simulations emulate realistic urban scenarios, including adverse visibility conditions, to evaluate the contributions of individual sensors and the overall multi-sensor system. The visual measurements prove essential to provide centimeter-level accuracy in the final phase of the approach trajectory, while the radar enables the navigation system to meet the assumed navigation performance requirements even at high ranges from the landing area and in adverse visibility. Experimental validation, conducted via scaled flight tests using Small Unmanned Aerial Vehicle (sUAV), demonstrates the effectiveness of the visual processing pipeline, including the ability to correctly distinguish between multiple landing patterns.

Downloads

Actions (login required)

Modifica documento