| dc.description.abstract | This report addresses autonomous inspection of wind blades using a UAV by proposing to use computer vision combined with structured light to accurately segment and detect the wind blade edges. It will be proven that this method is invariant to both rotation and scale, while it is computationally efficient for use on real-time applications. The blade edges will be detected according to the segmented area of the blade, and identified as Hough transformed lines to easily develop a manouvering scheme which intends to follow the blade edges from root to tip, while maximizing the view of the blade without loosing view of the respective edges. It will also be shown that the method detects the blade tip, regardless of rotation and scale.
Furthermore, the structured light will be utilized to efficiently compute feature point matches from a stereo vision system. These matches will be used to conduct 3D reconstruction of the respective feature points using 2D triangulation. This will enable transformation of coordinates between the image frame and camera frame, including a real-time estimate of the distance between the blade and drone. | |
