Research On Pose Estimation Algorithm For UAV Landing Based On IR Cooperative Object

UAV (Unmanned Aerial Vehicle) landing on the basis of computer vision is a hot issue presently.Within the problems of the issue, pose estimate is the key for UAV’s landing. The pose estimationalgorithm of UAV based on infrared cooperative object was studied.Firstly, the relationship of coordinate transformation for the UAV landing was establishedaccording the research of the camera projection model. On the foundation, the algorithms for poseestimation of UAV landing based on computer vision were designed. According to the existingdeficiencies of IR image segmentation methods, an adaptive region growing method was used. Usingcooperative object’s feature of high brightness, the growing seed was found within the cooperativeobject. Then, the growth rule was designed to get the target. Experiments showed that it was feasiblein various contexts for segmentation. Compared with double otsu and maximum entropy algorithms, itwas more flexible and had less noise.Then, according to the deficiencies of the Pose Estimation algorithm based on feature line withinthe feature line extraction, roll angle and yaw angle estimation, an algorithm based on the projectionmatrix was proposed to make the relevant improvements. The edge lines of the object and its cornerswere determined through the classification of the distances between edge points to a certain line. Therelationship between the two possible locations of the vanishing line and the roll angle was studied.The yaw angle was the angle between the center line of cooperative object, instead of the principalaxis, and the center line of image. Experiments showed that the modified method was more robust andimproved the accuracy significantly.Finally, to the problems within the above method that were requiring focal length known, thepoor accuracy of the pitch angle and height, and the considerable computational complexity, a poseestimation algorithm based on geometric model of cooperative object was proposed. After the analysisof the relationship between the various geometric features of cooperative object and the UAV’ pose,the geometric model was established. The pose and attitude of the UAV were acquired throughsolving this geometry model.The comparison experiments of the two algorithms showed the algorithm based on the geometricmodel reduced the error of attitude estimation by20%and declined the error of position estimation by40%.In addition, the experimental results of infrared image showed: the estimation accuracy of location parameter based on the geometric algorithm was0.8m in the range of30m, which was2.67%.The estimation accuracy of height and pitch angle was0.7m and3°respectively. Compared with thealgorithm proposed by Lee of Korea Aerospace Research Institute, the average estimated accuracy ofthe algorithm based on geometric model was higher. Also, it could achieve all the pose parametersbased on computer vision system without the aid of inertial navigation equipments. Furthermore, thealgorithm based on geometric model was adapted to the situation with the changing focus distance.