Accurate And Fast Identification And Localization Of A Cooperative Target In Complex Background

Space robotic arm plays an essential role in space station.It enables all kinds of operations outside the space capsule,such as docking,parking,fueling,maintenance,upgrades,transportation,rescue,and removal of orbital debris.Normally,a vision sensor(hand-eye camera)and an arresting device are fixed on the robotic arm;a cooperative target(an auxiliary measuring marker with a fixed shape)and a to-be-arrested device are installed on the to-be-captured good.The vision sensor takes images of the cooperative target,calculates the relative pose(position and orientation)between the target and itself using visual method,and then transfers this pose to the one between the arresting and the to-be-arrested devices,which is used as the basis for the robotic arm to plan its moving trajectory.However,the background of the visual camera is complex,the lighting condition is poor,the position is variable,and the image may have slight motion blur.In addition,positioning accuracy and real-time requirements are high,and the resource of aerospace hardware is limited.Therefore,the accurate and rapid identification and localization of a cooperative target in complex scenes are crucial techniques for the robotic arm to successfully perform tasks.It consists of several main research contents as follows: target design,target identification,marker center positioning,pose measurement and hand-eye calibration.Cooperative target design is an essential prerequisite for recognition and localization.It determines the strategy and difficulty of recognition,and influences the accuracy of pose measurement.Combining circle and line,and using dot as markers,a cooperative target is designed.Based on the design,the pose measurement accuracy,camera movement range and velocity are analyzed.The proposed target design has distinctive characteristics from the space background,simplifies its recognition process,is suitable for both human eyes and machine vision,and ensures high-accuracy pose measurement.Cooperative target identification is a necessary condition for high precision target positioning.However,it is not easy to identify the pose changeable target accurately and fast in complex scenes with as little storage occupied as possible.Firstly,an adaptive threshold and local optimal edge point selection method based on gradient histogram are used to draw out single-pixel-width,smooth and continuous edges of interest in the image.Then,using the information of isophote curvature and the Helmholtz principle,circles are quickly extracted from edges.Next to each circle,according to the circle center and radius,and the physical size and motion range of the target,an adaptive squared area is calculated.Within the area,straight lines are detected to identify the target.A large number of experiments under a variety of complex environments showed that the recognition accuracy of cooperative targets was always higher than 97.5%.The algorithm has high accuracy,fast speed and is robust to pose illumination and motion blur.The accuracy of the relative pose measurement is seriously affected by the center positioning accuracy of the target markers on images.In this paper,two steps,i.e.marker identification and center positioning are used to locate the marker centers with high accuracy.The former first extracts the reference information,such as the center of the target,the size of the ring,the gray-scale value of foreground and background.Then,based on the information,the region growing method is used to find the initial marker candidates from reasonable edges.The latter uses methods like the improved non-maxima suppression,partial area effect based sub-pixel edge location and ellipse fitting,to calculate the marker central coordinates with sub-pixel accuracy.Under the conditions such as illumination non-uniformity and motion blur,the proposed algorithm accurately identifies the markers;the central positioning average error is less than 0.03 pixel.It is robust and accurate,and ensures high accuracy pose measurement.According to the pixel coordinates of marker centers,the target physical size and the camera intrinsic,this paper uses the P3 P algorithm to measure the relative pose between the cooperative target and the camera.To convert the pose to the one between the arresting and to-be-arrested devices,it is necessary to solve the conversion relation between the camera and the robotic arm,which is called hand-eye calibration.Applying matrix rearrangement,this paper proposes an on-orbit hand-eye calibration method using the cooperative target.It is verified that its calibration error was less than 0.004 mm,which met the precision requirement of the robotic arm system.The advantage of this method is that it enables on-orbit calibration,and has the merits of simple operation steps,small calculation amount and high precision.The localization accuracy of the proposed algorithm was tested by a large number of simulation and real images.The results showed that,when the target was 0.3 m away from the camera,the rotation errors of the three axes(Axis X,Y and Z)were all less than 0.015° and the translation errors were all less than 0.2 mm;at the distance of 1.5 m,the rotation errors of the three axes were less than 0.15°,and the translation errors were within 2 mm.The proposed algorithm is very suitable for high precision positioning of cooperative targets in aerospace projects.