Reseach On Position And Attitude Estimation For UAV Landing Based On Vision

This paper derived from a kind of Airborne Vision-based Auto-landing System. According to theproject requirements, this paper obtained UAV’s position and attitude parameters estimated by air-borne vision-based system in the landing phase of UAV.Firstly, the quick method to extract feature line from landing images was studied in this paper.According to differences of blue component between sky and ground area, a rough horizon extractionalgorithm is presented by using Otsu classification and RANSAC method, which can obtain accuratehorizon parameters and the division of sky and ground. By using sampling pyramid technology, thelow-resolution image of original image was built and researched in this paper. This method can notonly reduce calculated amount of image processing algorithm, but also does not loss much accuracyof algorithm. By using improved Hough transform line extraction algorithm，which is based on phasegrouping，this method has not only expanded the scope of application of the algorithm, but also im-proved the real-time performance and accuracy of the algorithm.Secondly, the hardware and software constitution of airborne position and attitude visual estima-tion system was studied in this paper. According to the main function of this system, the hardwareplatform of visual estimation system was built; the software of visual estimation system was devel-oped, including mission planning, thread design, overall implementation process of applications,image processing and position and attitude parameter estimate algorithm function module, etc.Thirdly, in order to test and verify the performance of position and attitude parameters estimationsystem, the semi-physical simulation system of UAV visual landing based on DM642developmentboard was built, which can simulate the landing process of UAV and system performance of visualestimation system during dynamic process in laboratory.Finally, the semi-physical simulation experiments were studied in this paper. The results showedthat the position and attitude visual estimation system studied in this paper can accurately extract de-sired feature lines, estimate position and attitude parameters, which have higher accuracy and goodreal time performance.