Real-time stereo-vision localization system for safe landing of unmanned aerial vehicles

Abstract

Orchestrating a safe landing is one of the greatest challenges for Unmanned Aerial Vehicles (UAVs). This paper aims at the autonomous localization and landing bottleneck by developing a real-time ground-based stereo visual system. This novel architecture consists of two separate perception components which are mounted with a pan-and-tilt unit (PTU) and optical sensors. Furthermore, a tracking-inspired stereo detection algorithm is proposed to improve localization accuracy. The algorithm synthesizes a Bounding Box Shrinking (BBS) approach into the Generic Object Tracking Using Regression Networks (GOTURN) method. Both datasets driven offline simulation, and online flight experiments are conducted to validate effectiveness as well as better performance of the novel system and the overall accuracy during the landing process. Also, this autonomous landing system caters for different UAV systems in operation, such as fixed-wing and rotary wing, particularly in GNSS-denied or-impaired environments.