Research On Visual Target Localization System Based On UAV Platform

Localizing and measuring the spatial coordinates of the landing target is to test the accuracy of the weapon hitting the target.According to the requirements of localizing and measuring the spatial coordinates of targets which falling into the sea or ground,this paper designs a visual target localization system based on UAV platform,and studies key technologies and methods around the system.This paper analyzes the design requirements of the target localization system,and designs the overall scheme of the visual target localization system based on the UAV platform.It is divided into two localization methods: UAV’s camera combined with spherical targets and UAV’s camera combined with plane.The UAV’s camera combined with spherical targets localization mainly uses the principle of homography to solve the spatial coordinates of the target.The main principle of UAV’s camera combined with plane localization is similar to monocular structured light measurement.And an error correction method based on Pn P pose optimization is designed to improve the accuracy of the UAV’s camera combined with the plane localization method by adding spherical targets.In this paper,a method for extracting target pixel coordinates based on digital image processing is studied.The main idea is to convert the image to HSV color space and segment the target mask through the corresponding color threshold,and obtain the center coordinates of the segmented mask by the circle fitting algorithm.The target detection algorithm Yolov5 based on deep learning is studied.By using the result of digital image processing as the annotation information of the model training data set,the tedious work of data annotation in deep learning is reduced,and the automatic extraction of pixel coordinates is completed.In this paper,the error sources of UAV’s camera combined with spherical targets and UAV’s camera combined with plane localization methods are analyzed.The effects of various errors on the two localization methods are simulated by Monte Carlo method,and the shooting pose range meeting the measurement requirements is determined by the simulation.This paper realizes the system functions through Python and Py Qt graphical interface framework.Experiments of UAV’s camera combined with spherical targets localization and UAV’s camera combined with plane localization were carried out.The experimental results show that the target localization results of the UAV’s camera combined with spherical targets have high accuracy,and all the results are within 0.7meters;the target localization results of the UAV’s camera combined with plane can improve the accuracy from about 8 meters to within 2.5m after error correction.They can meet the needs of localizing and measuring the landing target.