A Novel Framework of Visual Detection, Tracking and Landing for UAVs Utilizing AR Markers

An Unmanned Aerial Vehicle (UAV) visual landing is an example of a robot visual navigation application scenario. State estimate is a crucial component of landing missions. The advancement of computer technology and algorithms has dramatically enhanced the speed and precision of visual processing techniques. We present an approach to predict the states (attitude, location, and velocity) of unmanned aerial vehicles (UAVs) based on an onboard camera during the landing phase, with the goal of improving the autonomy of intelligent multi-rotor UAVs. First, the control points of a visual landing target are derived by detecting an ArUco marker in vision. The attitude and position of UAVs, also known as their pose, are then determined. To calculate dense optical flow, a solution based on corner point interpolation of the Lucas-Kanade algorithm is presented due to mission speed and precision requirements. The integral of optical flow is then derived from the optical flow field and certain UAV attitude limitations. Lastly, the UAV's velocity is computed based on the optical flow integral and the UAV's estimated pose. This estimate approach has apparent advantages because it just uses the visual sensor. Integrating optical flow may eliminate the influence of noise spots in the optical flow field.