Improvement of Object Segmentation Accuracy in Aerial Images

Abstract

With recent advances in UAV technology, research-based on UAV images is underway. UAVs can easily access places that are difficult for people to access and take a wide range of target areas. However, UAV images taken at high altitudes using a drone have object images with a tiny size in the entire background image, resulting in a more significant area error in the area of the detected objects. This paper proposes an accurate area measurement algorithm within an object based on image processing. Also, we evaluated the proposed algorithm by implementing it. The experimental results show that the average duplicate error rate decreased by 14% compared to mask instance segmentation. Finally, the proposed algorithm can more accurately extract small potholes in the images taken at high altitudes.