Computational Speed and Qualitative Assessment of Real-Time Image Stitching Algorithm

In remote sensing and environmental mapping, Unmanned Aerial Vehicle (UAV) has been used extensively to capture images. For many years, digital maps are generated by using a method called image stitching. It is a method of combining multiple images to produce a segmented panorama. Since this method has been commonly used, many users produce an accurate map by using commercial software. However, a downside of this commercial software is a long computational time which is not appropriate for immediate mapping activities at chaos areas in particularly during rescue missions. This paper proposes a method to speed up the process of map generation by using a revised real-time image stitching algorithm including the qualitative assessments. In this research, the images are extracted from a video taken by a drone that flies in a dedicated flight path. These videos are then immediately transmitted to a ground station for further image processing and computation. The overlapping images are stitched and later undergoes features extraction process to identify the common features between the images. These common features are used to compute homography matrix which beneficial for image wrapping. The finding of this study suggests that ORB and AKAZE are the most suitable descriptors to be used in real-time image stitching because of their fast computational speed at adequate level quality. For instance, at 5 number of skip frame, ORB is at least 2-fold faster than AKAZE and goes up to 10-fold faster than BRISK.