ViNavUT: Vison Based Navigation System for Unmanned Tractor in Airport
Yichen Zhou, Mingxing Wen, Jun Zhang, Danwei Wang · 2023
Airplane towing operations in the apron demand seamless cooperation of multiple operators, thus usually face several challenges including high manpower cost, poor safety and low efficiency. To address these problems, ViNavUT, a vision based navigation system for unmanned tow tractor has been proposed, where only one camera is leveraged to perceive environment and derive a low-price but robust autonomous navigation solution. ViNavUT consists of three main modules. Object detection and segmentation: a multi-object detection and tracking has been utilized to perceive the dynamic entities in the apron. And image segmentation is also used to achieve pixel level classification. Vision based object localization: benefiting from the segmentation results, a refined depth estimation method based on the pinhole camera model has been proposed to determine accurate localization of the objects for path planning. Adaptive navigation: motion predictions of surrounding objects are taken into consideration by path planning to realize a robust autonomous navigation adaptively. Extensive experiments have been conducted to verify its efficacy, and the experimental results prove that ViNavUT is a potentially robust and cost affordable navigation solution for autonomous apron towing operation.