Automatic visual identification of correct matches in unmanned aerial vehicle images for visual-based attitude estimation

This paper presents a framework of automatic clustering to determine correctly matched keypoints locations in aerial images for visual-based attitude estimation. In this work, correct and false matches are automatically identified using a clustering technique which utilizes the outlier information to determine the initial number of clusters and cross-correlation. The proposed framework has been tested on a set of 152 Unmanned Aerial Vehicle-acquired images, and the results have been compared with the visual inspection. The comparison has shown that the proposed framework is able to provide an acceptable matching accuracy, minimize the size of region-of-interest images, and simplify the key points computation.