Modeling And Pose Measuring Of Non-Cooperative Target Based On Point Cloud In Close Range

With the development of space technologies,the demand for orbit services including orbit maintenance and orbital garbage cleaning is becoming more and more urgent.The failure spacecrafts and orbital garbage are non-cooperative targets.The 3D reconstruction and close-range measurement of relative position and orientation of the non-cooperative targets are the key to realize on-orbit sservice.However,the 3D reconstruction and close-range measurement of relative position and orientation of the non-cooperative targets in the space features limited measuring equipment,large scale-variation and complex space environment.It brings great challenges to ralated research work.Based on an aerospace project included in the National Hight Technology Research and Development Program of China(also known as 863 Program),this dissertation researches the 3D reconstruction and measurement of relative position and orientation of high-orbit non-cooperative targets by taking communications satellites with on-orbit failures as research objects.Therefore,the research has an important theoretical and practical significance.According to the characteristics of geostationary orbit(GEO)service,the mission requirements are analyzed and the corresponding measurement system is designed.Proceeding from the mission requrements for GEO on-orbit services,the characteristics of the measured targets,the measurement difficulties and the limitations of the existing measurement devices are analyzed.Therefore,a plan is designed for the close-range measurement of non-cooperative target.In view that the failurese satellites are not equipped with the cooperative-target marker,the To F(Time-of-Flight)camera is used to conduct 3D reconstruction and close-range measurement of the relative position and orientation of non-cooperative targets.Beside,the performance of this plan is compared with the existing close-range space measurements,this plan have the advantages such as compact design and robust to illumination changes.To solve the problems of the errors caused by external environment in the close-range measurement of non-cooperative targets,this dissertation puts forward an error modeling based on particle filter support vector machine(PF-SVM).Limited by the imaging conditions and interfered by external environment,the data acquired by the To F camera have certain errors.The effects of typical interference factors,such as material,color,distance,illumination change and scene complexity,on the depth error of the To F camera are analyzed on the basis of the characteristi cs and the environment of the target to be measured.The relationship between the above factors and the measurement error is obtained,and the error model is established.The model shows that the depth error of the To F camera is less affected by the backgr ound illumination or the color of the target.However,it approximately increases lineraly with the reflectance of the measured material and the distance.For random errors of the To F camera,this dissertation selects the parameters of support vector machi ne error model by particle filter to achieve depth error correction.To solve the problems of low resolution and less detailed information of the depth image,a fusion method of high-resolution color image and low-resolution depth image is proposed to realize super-resolution reconstruction of depth image.Generally,limited by the number of detector arrays,the resolution of the To F camera is low and the depth information is not detailed,which restricts its applications to some extent.In this dissertation,the high-resolution color image and low-resolution depth image are fused to realize super-resolution reconstruction of depth image.The method first requires to align color image to depth image,then take high-resolution color image as the input to get guide image by using joint bilateral filter,and finally use guide filter to get high-resolution depth image.This reconstruction method can enhance the spatial resolution of the depth image and eliminate the artifact effectively.Based on the point cloud data of a non-cooperative target,an interative closest points(ICP)algorithm is presented based on the features of point clouds,which is used to builed the 3D model of the target spacecraft.The 3D point clouds of from the surface of an object is obtained directly by To F camera,and certain surface point cloud data obtained from different angles are integrated and registered.T his method uses the geometrical features of the point clouds to be registered,such as curvature,surface normal and point cloud density,to search the correspondence relationship between two point clouds and introduces the geometric features into the error function to realize the accurate registration of two point clouds,which avoids the ICP algorithm to fall into local extremum and has higher convergence speed because it doesn’t need to set a well-defined initial value.Aiming at the problem that the non-cooperative target lacks the prior knowledge and the difficulty of pose tracking,the dissertation proposes a method of automati c recognition and position and orientation tracking of non-cooperative targets based on point cloud features.In this method,the 3D point clouds of a non-cooperative are firstly to calculate the curvature,normal direction,density and other geometric characteristic parameters of the target so as to identify it accurately.Then the partical filter algorithm is used to obtain the position and orientatin of a non-cooperative target by calculating the similarity of the point cloud features of two adjacent fra mes.The method can effectively identify the features of a non-cooperative tumbling target and realize the tracking of its position and orientation.