Research On Visual Guided UAV Landing Technology Based On The Combination Of Visual Features And Cooperative Object Optimization

UAV (Unmanned Aerial Vehicle) landing on the basis of computer vision landing research isimportant on application value and theoretical significance.Within the problems of the issue,estimate UAV position and orientation by visual is the key technology for UAV landing.In thispaper,The pose estimation of UAV and Cooperative Object Optimization was studied.First, the relationship between UAV vision systems coordinate and the world coordinate systemwas studied. For the problem of Cooperative object extraction effect is not ideal for the existingmethods in the complex visible light environment image.According to same colors get together aftermean shift color clustering, a color clustering based on Mean Shift, OTSU segmentation andconnectivity area selection image segmentation method was proposed, compared with the existingextraction methods can be more effective cooperation objectives in the complex visible lightenvironment image segmentation.Then, in order to increase the scope of the vision-guided landing and improve system stability,the combination of the pose estimation method based on the horizon, runway and cooperative objectcharacteristics was studied.The Specific guidance process is:In the distance,solutions of the pitchangle and roll angle was calculated by the onboard vision system based on the image of the horizon,to ensure UAV is in normal flight attitude; In closer range, the pose was estimated based on therunway line features pose estimation method, combined with the horizon to get the most of the poseparameters of UAV;In close range, the cooperative object features in the image was extracted, thewhole pose parameters was estimated by estimation method based on cooperative object.Finally, to the problem of large pose estimation accuracy differences among the existingcooperation objectives in the case of different height and pitch angle and the pose paraments can notbe obtained even at a critical stage.By analyzing the sources of error and studying the impact ofcooperative object shape on pose estimation,the improved genetic algorithm was used.In thealgorithm,the shape of polygon was generated by six vertices in polar coordinates distribution asinitial solutions.A new two-dimensional cooperation objectives was optimized.In order to furtherreduce the differences in the accuracy of pose estimation,the relative pose of two planes cooperativeobjects was optimized.A new two planes cooperative object was optimized.verification tests show thatcompared to traditional T shaped and double-circle shaped cooperative objects,the optimized new plane cooperative object reduced the accumulative error of estimation by10.5%; compared to theoptimized single-plane cooperative object, optimized cooperative object of planes reduced theaccumulative error of estimation by13.5%.