FPGA-based approach for runway boundary detection in high-resolution colour images

Systems for aerial vehicles have to face tight constraints on weight, space, and energy consumption due to limited payload and energy resources of aircrafts. This leads to the use of optimised, application-specific components. In exploration and surveillance scenarios, electro-optical (EO) sensors in combination with embedded systems are very suitable to contribute to various perception tasks. EO sensors are lightweight, affordable and provide a high-quality representation of vehicle's environment. Embedded systems are energy-efficient, space-saving and provide powerful computing capabilities. But processing of high-resolution images is challenging, especially in the context of embedded computing and real-time data exploitation. Considering these conditions, the article introduces a novel FPGA-based approach for runway boundary recognition. The source image is scanned line-by-line to identify colour variations. Locations with strong colour discontinuity are grouped to lines which are used for runway pattern extraction in image. The classifier-less approach is independent from runway colour, brightness and contrast and doesn't require additional markers. The final detection is evaluated by a confidence value indicating its trustiness. The determinability of the worst case execution time and the robustness over a wide dynamic range demonstrate the certifiability of the implementation. It will be tested on an unmanned aerial vehicle for automated landing.