Feature-based real-time aerial image stitching and quality assessment for post-disaster application

Abstract

In the past, digital maps were created using a photogrammetry framework where the Unmanned Aerial Vehicle (UAV) would collect the aerial images; then, images would be post-processed through commercial software using the Structure From Motion (SFM) method. Creating digital maps has been helpful for remote sensing, especially for studying and observing the terrain. However, one disadvantage of this method of creating digital maps is that it consumes more computational time. Although commercial solutions are widely used, they are not suitable in disaster-affected areas because of the long computational time. Disasters such as earthquakes, floods, and landslides would happen without prior notice, and areas affected by such a disaster would suffer heavy damage. In such a situation, the authorities need an instant digital map to observe the affected areas and decide. Hence, this study focuses on accelerating the creation of a digital map using the real-time image stitching method. Image stitching itself can be divided into feature-based and region-based methods. This study uses a feature-based image stitching method to accelerate the map creation process. This research formulated an image stitching algorithm to stitch aerial images in real-time. A processing speed of 37 frames per second was achieved. The image stitching algorithm was optimized to stitch large areas captured using the multi-grid flight path; a processing speed of 2 frames per second was achieved. Finally, an image selection algorithm was introduced to improve the stitch image quality by 14% and the computational time by 2-fold for a multi-grid flight path. In conclusion, the developed image stitching algorithm can reduce the computational time needed to produce a digital map at the disaster site. Although the developed image stitching algorithm can stitch faster with improved quality, more testing needs to be conducted using aerial images from disaster sites.