Asynchronous Localization of Ground Objects Using a 2-UAV System

摘要

Vision based target tracking has immense application in areas such as security, traffic control, and defense. We present a novel method to estimate the position of a ground target using joint visual observations from a pair of unmanned aerial vehicles (UAV). Our method use a fast recurrent convolution neural network (RCNN) for low-level object detection. The results of RCNN is then fed to an EKF model for position estimation. The integrated vision pipeline and target recognition subsystem is evaluated with soft-in-the-loop simulation and the result shows substantial increases in tracking coverage.