Vision-Aided Navigation for UAM Approach to Vertiports with Multiple Landing Pads

The vertiport concept has spread widely as the future aerodrome that will allow Vertical Take-Off and Landing vehicles to operate in complex and congested scenarios, such as those foreseen within the Urban Air Mobility framework. Possible vertiport configurations have been proposed by many government agencies such as the European Union Aviation Safety Agency and the Federal Aviation Administration providing general design and development guidelines. This paper introduces an autonomous visual-aided navigation architecture able to estimate the aircraft state during approaches to vertiports exploiting visual observables gathered from multiple landing patterns. The implemented architecture exploits a Convolutional Neural Network for landing patterns detection; it then performs their discrimination and, for each of them, keypoints detection and identification to feed a perspective-n-point solver. The resulting pose measurements are input to an Extended Kalman Filter, which also processes data from an Inertial Measurement Unit and a Global Navigation Satellite System receiver. The implemented architecture is tested on synthetic and real data, showing the validity of the pattern discrimination strategies and the performance of the visual-aided filter as a function of the number of detected patterns along different approach trajectories.