Motivation
Traditional pre-flight inspections of commercial airliners are labor-intensive, time-consuming, and prone to human error, often requiring workers to manually access hard-to-reach areas to identify dents and other defects. This process can take up to four hours, resulting in costly downtime for airlines, with potential losses averaging $10,000 per hour in missed earnings. Implementing drone-enabled inspection systems can significantly reduce inspection time to under 30 minutes while increasing accuracy and enabling remote analysis of high-resolution images, ensuring quicker detection of issues such as cracks, popped rivets, and leaks.
A manual pre-flight inspection
Automatic Inspection of the aircraft with a drone
Problem Statement
The objective of our project is to develop a drone-based inspection system capable of accurately detecting dents as small as 1-5mm in depth and 1cm in width from a distance of 2-4 meters. To ensure reliable performance, our algorithm must be robust to varying aircraft types, illumination conditions, and surface textures.
A plane wing with dents on it
Part of a fuselage with multiple dents on it
Challenges
The main challenge in detecting dents is that photographic images alone can be unreliable under certain viewing conditions. Consider a simple example: a metal plate with several dents. The two images below show the same plate photographed from different angles. In the first image, the dents are clearly visible, but in the second, changes in illumination and surface reflections render them nearly invisible.
Photos of the same plate with dents with different viewing angles and illumination
To address this problem, we employ active illumination techniques. Specifically, we project a straight laser line onto the aircraft surface. Any dents will distort the laser line, and by detecting these deformations using classical computer vision methods, we can identify candidate dent locations.
By projecting laser line onto the aircraft, we can identify candidate dent locations