Research On Pose Estimation For Uav Based On Cooperative Object

For the purpose of implementing UAV's landing automatically and accurately under all-weather, computer vision navigation based on cooperative object on the runway is proposed. In this paper we did some research on pose estimation for the UAV.As the cooperative object just occupy a small percentage in the image, classical Otsu method cannot segment it. Considering the cooperative object emits specific infrared light that matches the infrared vision system, it must have the highest temperature and is brighter than any other part of the image. In this paper we use Otsu method twice. After first segmentation, make and operation between original image and separated image, at this time the target occupy a large percentage that can meet the requirements of the Otsu method. Then we use sobel operator to detect the edge of cooperative object, and make use of chain code to track the object contour. During point extraction period we use Harris method, and when extracting line we select a point on the chain code as a starting point, and then calculate the distance between point and the T-head straight line in the clockwise direction of the chain code. According to the threshold of distance we partition all the points of the chain code and then remove some points to fit line separately.During the period of guidance for UAV's automatical landing based on cooperative object and computer vision, a feature points method is proposed firstly. It is a calibration method using Tsai algorithm and its core idea is RAC. The pose parameters which we want to get can be obtained after calibration. Analysing the results we find this method has a disadvantage that the error increases by height. So we propose another method through making use of feature lines. The method uses the principle of vanishing line by taking full advantage of the actual shape of cooperative targets. The roll angel can be calculated from the vanishing line and yaw angel can be obtained from the image. Other parameters can be calculated by the relationship between world coordinate system and camera coordinate system. Experiment results show the feasibility of these two methods, and the line method has a higher precision and is more reliable.