Research On Runway Detection And Recognition Method For Visual Landing Of UAV

At present,due to the increasing progress of computer technology,computer vision is widely used in the field of Unmanned Aerial Vehicle(UAV)navigation.The application of cheaper and more flexible visual sensors makes the vision-based navigation method show great advantages in the field of UAV navigation.Computer vision technology began to emerge in dealing with the problems in the direction of UAV navigation,and this technology quickly occupied an important position with its advantages in the research of related topics.The application of visual navigation in the landing of fixed wing UAV has been studied in this thesis.Meanwhile,difficulties existed in visual landing of UAV under foggy weather have been further studied.Firstly,the images taken by UAV in foggy weather were enhanced and restored.Then,in order to ensure the safety of UAV visual landing,research on detection and recognition of runway in visual landing of UAV was carried out on the basis of clarity technology of foggy image.The followings are what have been done in this thesis:1.The defogging algorithms of the images taken by UAV landing in foggy weather have been studied.Core ideas of representative histogram equalization algorithm and defogging algorithm of theoretical model of dark channel prior have been discussed in detail from the two aspects of image defogging algorithm based on non-physical model and image defogging algorithm based on physical model,where the algorithm steps and principles are analyzed.The effect of the algorithm is verified through experimental results of image processing.The shortcomings and disadvantages of the algorithm have been deeply studied as well.2.This thesis focuses on the most advanced Retinex series defogging algorithms based on non-physical models,including Single Scale Retinex(SSR)and Multi-Scale Retinex(MSR).According to the advantages and disadvantages of this kind of algorithm,an improved MSR algorithm suitable for UAV image is proposed,which can effectively meet the real-time defogging with high quality.Firstly,Red-Green-Blue(RGB)components of the foggy image taken by UAV is enhanced by homomorphic filtering algorithm,and then converted from RGB space to Hue-Saturation-Value(HSV)space.While keeping the hue H unchanged,the brightness component V is processed by bilateral Retinex,the incident component is enhanced by gamma correction,the reflection component is histogram equalized to correct the contrast stretching and the edge of the reflected image is enhanced by log enhancement operator.Meanwhile,the saturation S is adaptively nonlinear stretched.Then the image is converted to RGB space.By designing and compiling the QT simulation platform of defogging model of image,the simulation results of various defogging algorithms and the analysis of gray histogram results are displayed and compared.Finally,the defogging effect of various algorithms is evaluated according to the image of objective quality-evaluation-index to show the superiority of the improved algorithm model.3.On the basis of defogging effect of UAV image,the preprocessing of runway image of UAV landing is studied.Compared with the advantages and disadvantages of common edge detection operators,the more advantageous Canny operator is selected to complete edge detection of image.Combined with runway texture features,the recognition of airport runway is realized by line detection method of Hough transform.Finally,the simulation experiment of runway detection is designed.The camera calibration is used to obtain the internal and external parameter matrix and distortion coefficient of the camera,so as to simulate the real airport runway environment.In turn,the accuracy of runway detection and recognition is verified through the experiment.