A vision-based target following and landing control for quadrotor UAVs

Abstract

This thesis describes target following and landing control system of a quadrotor UAV by using onboard vision sensor. The target defined in this paper is a land target which has artificial color or patterns in order to detect landing site with the vision sensor. Vision processing algorithm was implemented on smart devices to achieve onboard processing.The visual targets are defined as a colored circle, which is detected by using simple processing algorithm. This algorithm include not only finding red objects on the image frame, but also selecting the most circular object among them. The target was followed by a commercial quadrotor platform with mounted forward facing camera.A location of the target in an image frame is combined with inertial sensor data to spot a metric position of the target on a navigation frame. The omnidirectional camera model were provided to generate geolocation algorithm while assuming flat earth. Also, a horizontal dynamic model of the target in the local coordinate frame, which is a constant velocity model, is carried out to construct nonlinear Kalman filter. Noises from detection algorithm are suppressed and nonlinearities from state update equation are resolved by using Unscented Kalman filter.Two control strategies are proposed in this thesis. Target coordinates in the image frame can be directly put into the quadrotor as a control input, which is frequently called ‘Direct Visual Servoing’. Also, the other method is considered, relative metric distance and velocity based target following with using outputs of the geolocation filter. The two individual methods are constructed and implemented to test their performances. Also, landing on a moving pad with nearly constant velocity was achieved consequently.The whole system described in this paper uses the UAV’s own onboard inertial and vision sensors without external motion capture system. The image processing, geolocation filter and control algorithms are integrated and validated by p...