Abstract
This paper presents a new solution for the visual servoing of the translational kinematics of vertical takeoff and landing vehicles. While previous works mainly focused on the servoing from known objects under controlled conditions, the present paper deals with the servoing from objects of unknown three-dimensional (3D) geometry. The proposed technique is a hybrid controller that combines the global stability of position based visual servoing with the robustness of image based schemes. This is achieved using a switching strategy, which favours the advantageous controller based on the disparity between the reference and actual image. Although using a monocular camera, our technique supposes no knowledge about the 3D location of the features of interest. Only an approximation of the average depth of these features is required to realise a good tradeoff between performance and robustness. We validate the proposed control strategy on a quadrotor unmanned aerial vehicle in a mission of power assets inspection. To achieve real time performance, we consider a framework that combines optical flow based tracking with a periodic reinitialisation using scale invariant feature transform descriptor. Results from real world experiments as well as from a realistic simulator are provided to validate the proposed solution.