Dual UAV-based cross view target position measurement using machine learning and Pix-level matching

Abstract

The measurement of target position in dual or multiple unmanned aerial vehicle (UAV) scenarios presents significant challenges, including data fusion for accurate location and depth determination, image matching difficulties and high computational requirements. This paper proposes a methodology for precise target position measurement in dual UAV-based cross view by employing the Zhang algorithm for camera calibration, the YOLOv8 network model for target detection and the SuperGlue algorithm for feature point extraction and registration. The methodology offers unique contributions to the literature on target position measurement using UAVs, including a novel approach for matching identical targets in different images and a method for measuring the position of shared targets in dual drone images within real space using a target pixel positioning matrix. The experimental results demonstrate the effectiveness of the proposed methodology in offering a promising and systematic solution for accurate and reliable target position measurement using dual UAVs.