An Vison-Based Pose Estimation Algorithm For UAV Autonomous Landing

Unmanned aerial vehicle(UAV)has become one of research hotspots in recent years,according to relevant statistics show that more than half of the unmanned aerial vehicle accident had occurred in unmanned aerial vehicle landing phase.The causes of this phenomenon is that unmanned aerial vehicles in landing phase,its flight status,speed,height,Angle,its configuration has very big change.In extreme cases,the drones can not estimate the position and speed,when there is out of control,unmanned aerial vehicle ground staff performing the wireless remote control needs certain delay.Therefore,to enhance the security of unmanned aircraft landing process,it is necessary to improve the control ability of unmanned aerial vehicle.However the improvement of independent control ability,the information acquisition and application of airborne navigation equipment capacity are closely related.Traditional GPS/INS group and navigation and GNSS rely on satellite positioning signal in the presence of a shade or shielding complex environments,especially when there is a signal shielding or loss of environment,the traditional methods cannot play its proper role.Therefore the UAV autonomous landing arises at the historic moment.This paper puts forward a kind of landing target extraction and pose estimation algorithm based on HSV color histogram,ellipse fitting and polygon fitting algorithms,to perform pose estimation for UAV in landing period.In this paper,the main work includes the following content:1.This paper proposes a new model of the landing target,the target model is consist of outer hexagon,outer circle,second circle,inner round,and a triangle.Circle and triangle are used for pose estimation.Hexagon is used to determine the target area.Each size of the graphic proportion is fixed,so that it is able to distinguish with the surrounding environment,and estimate UAV position according to the image size.2.A target recognition method is proposed in this paper,through the graphic color features a preliminary judgment whether there is land target in the image is made.The ellipse's accurate position can be confirmed by the ellipse fitting and elliptical filtering.Polygon fitting determines landing target detection range.According to the proportion of concentric circles and landing target color characteristics,we can eliminate the interference of complex environment and avoid the matching target errors.3.This paper improved the previous ellipse-based location estimation algorithm,through coordinate transformation and equation solution,we confirm the drone position.We transform pose estimation into solution of 12 degree univariate polynomial equation as well as simplify the solving process of the formula and complexity.4.By aerial photo tests under the condition of different illumination,angle and distance,we detect the accuracy of target recognition algorithm and the pose estimation algorithm.Aerial tests from 1 to 10 m different heights,and statistics target recognition probability of the landing target.In this paper,the experiment can effectively identify the landing target,eliminate the interference of surrounding environment,and in the pose estimation period error can reach 10 cm range.