An Algorithm Of Vision-based UAV Pose And Attitude Estimation

The landing of the Unmanned Aerial Vehicle (UAV) is crucial in the process of UAV flight. Vision-based landing of the UAV, being a hot research spot in the field of UAV control, is the main direction of landing control for UAV. In the process of landing control for UAV, the image processing algorithm directly influences the precision of the landing control.This paper focuses on the image processing and position and attitude estimation in landing control of UAV. In this paper, by using the perspective projection model and the translation coordinate frames relations, linear normal equations containing 6 DOF of rigid body can be acquired. And then, least square method can be used to solve the equations by linear optimization algorithm.The precision of the estimation can be improve greatly by improving the precision of the feature image processing. This paper proposes a new algorithm combined the scanning and state in the low image processing of the estimation for UAV. When scanning the image, we get the feature borders, extracting feature corners based on the convexity of 4-sided polygons and labeling all features last. Computer simulation experiment can effectively simulate the capability of the algorithm for vision-based landing processing while lacking condition of physical experiments. This paper generates simulation image sequences by the sequences of given position and attitude, detects corners by using the image features processing algorithm, estimates the relative position and attitude of the UAV by the estimation model, compares the estimation value with the given values at last.It is illustrated by simulation experimental results that the precision of the feature image processing and position and attitude of presented algorithm and the speed of computation can meet the requirements of precision and real-time for flight control system.