Pose and Velocity Estimation Algorithm for UAV in Visual Landing

Abstract

Visual landing of Unmanned Aerial Vehicles (UAVs) is an application scenario of robot visual navigation. State estimation is an important part in the landing mission. With the development of computer hardware and algorithms, the speed and accuracy of visual processing techniques have been greatly improved. Aiming at the requirements of intelligent multi-rotor UAVs to improve their autonomous capabilities, we propose an algorithm to estimate the states (attitude, position and velocity) of UAVs based on an onboard camera during the landing phase. Firstly, the control points of a visual target are extracted by detecting an ArUco marker in vision as landing target. The attitude and position, named pose of UAVs are then calculated. According to the requirements to mission quickness and accuracy, a method based on corner point interpolation of Lucas-Kanade algorithm is proposed to calculate the dense optical flow. Then the integral of spherical optical flow is obtained by optical flow field and some constraints on the attitude of the UAV. Finally, the velocity of UAV is calculated based on the combination of the optical flow integral with the estimated pose of the UAV. This estimation algorithm has obvious advantages in that it only uses the visual sensor and can reduce the interference of noise points in the optical flow field by integral of spherical optical flow.