Small-Object Detection for UAV-Based Images

Abstract

Unmanned aerial systems (UAS) are increasingly being deployed in civilian and commercial areas. The application of machine learning in UAS image analysis greatly promotes the progress of target detection and tracking algorithms. However, current object detection and tracking system algorithm can hardly be applied to detect aerial targets. Because the view of UAS changes and rotates quickly during the flight. In this paper, we propose a fast and accurate real-time small object detection system based on a two-stage architecture. The proposed addresses the small object detection challenges by combining the traditional target detection with deep learning. More precisely, it uses conventional background subtraction and deep learning algorithm to get the initial detection box, and then use target tracking to get the final result. We evaluated our approach on the small object data sets. Experimental results show that the proposed method has improved the aerial object detection performance compared with other conventional approaches.