Vison-based Pose Control And Autonomous Return For An Unmanned Aerial Vehicle

In recent years, rotor unmanned aerial vehicles(UAVs) have been highly valued in many fields because of their excellent maneuverability. To help manipulate the UAV more conveniently and control it more precisely, vision-based position and attitude control is always a hot research topic. Autonomous return is an important function for UAVs. This function is usually provided by the control system on many products based on satellite navigation signals, which are not reliable in complex environments. The purpose of this study is to control the pose of a UAV precisely and to propose an autonomous return approach based on vision.Firstly, this paper introduces a velocity controller to help improve the control accuracy of the UAV’s position and attitude, after the velocity is estimated by the IMU information and the optical flow of the image captured by a vertical looking-downward camera. The flight experiment of a quadrotor shows that the proposed algorithm can finish tasks like hovering and autonomous tracking, in which landmarks are regarded as reference. Meanwhile, the relatively accurate displacement can be estimated by integrating the estimated velocity.This paper also proposed a UAV autonomous return approach using a looking-forward camera based on the velocity controller. According to the multi-view geometry in computer vision theory, the 3D transformation between two perspectives can be obtained by matching the current image with the key frame. Then the UAV can be controlled to hover at the position of key frame. Navigation based on velocity estimation and hovering based on image matching are combined together to construct a robust autonomous return approach. On the outward path, key frames and displacement information are recorded when the UAV travels a constant distance. On the return path, after the UAV almost reach one key frame by the estimated displacement, hovering based on image matching helps the UAV reach the recorded position closely. Finally it will return to the start position by repeating these two steps. The experimental simulation result shows that this approach can solve the autonomous return problem. Also, a complex route with the change of yaw angel is conquered.