Research On Ground-based Visual Landing Technology For An Unmanned Aerial Vehicle With Infrared Detectors

Autonomous landing of UAV has already been one of the inevitable trend of the unmanned aerial vehicle system. The key point of the autonomous recovery is the high precision positioning of autonomous navigation. Along with the development of computer vision and photogrammetry, visual based landing guidance has become a new hotspot for researches on UAVs recovery. This paper is based on the background of ground-based visual guidance, focusing on the key technology in autonomous UAV landing. The infrared camera is chosen as the detector for quickly detecting and positioning of target thanks to the noticeable infrared features of UAV.The main work and contributions are summarized as follows:(1) A new designed calibration pattern for infrared cameras.This paper proposes a method for large FOV thermal binocular vision system calibration based on a new designed calibration pattern. Small circular constant heat reservoirs are used as calibration markers for infrared cameras while the visible cameras can obtain the center of the concentric black-white circles as calibration markers with high precision.(2) Infrared small targets detection.Based on the analysis of characteristics of infrared image, the paper studies the pre-processing algorithms of infrared small target image in thespace and frequency domain. Experiment results indicate that the algorithms can detect the small targets in real-time with background suppression and image segmentation.(3) The calculation of coordinates.Researches on the field of visual guidance are mostly based on the airborne vision design. In this paper, ground-based visual guidance is set up with cameras mounted on pan-tilt units(PTU) to dynamically measure the movement parameters of the UAV. The model of ground-based visual measurements is built. The calculation of coordinates is based on the angles of the PTU and the image coordinates with triangulation method. The 3D positioning principles adopted in the model are simulated and verified. In addition, the influences of azimuth angle error of the calculation are simulated and gradient vector fields are drawn for visual representation.