Research On Point Set Acquisition And Registration Methods In Non-cooperative Pose Measurement

As the development of aerospace technologies,space explorations have passed from satellites and spacecrafts launching to on-orbit technique services.The on-orbit technique services are expected to be highly autonomous.Among the tasks,the chasers must know the relative pose of the targets before docking with them.The cooperative pose measurement methods can only be used for predefined cooperative objects.Pose measurement for non-cooperative objects have been a focus among recent researches in many space agencies.To supply accurate pose results for spacecraft with known geometry from close distance,we would like to study three-dimensional point set sensing and registration methods for non-cooperative pose measurement,and apply these methods to a real non-cooperative pose measurement to demonstrate there advantages.We first studied the method of point set acquisition with stereo cameras.The key is the stereo matching,which estimates disparity maps based on two or more images that are captured from different views.To deal with the challenges and apply well to space objects,we proposed a three-dimensional(3D)aggregation technique for windows-based method,which is implemented by decomposing the adaptive support window into horizontal segments;and aggregated matching cost in or between segments with scan-line optimization.This technique can possess the complementary advantages of global/semi-global and local methods in slanted surfaces and depth discontinuities,respectively.Compared to traditional local stereo methods,the new algorithm has greatly improved the accuracy of results in slanted surfaces and occlusion areas while keeping excellent performance near depth discontinuities.The evaluation experiments have confirmed our achievements.We then studied point set registration method based on continuous distance field.The current method for the construction of continuous distance field is to calculate the Euclidean distance with one-order approximation,which implicitly assume the gradient of the distance field is constant from the 3D point to the model.This assumption decreases the probability of global convergence seriously.We decomposed the Euclidean distance into differential paths which can be accurately calculated with one-order approximation,and then summed these differential paths by adding a weighting value for each of them.The weight is the gradient,which can be regarded as the density of the path.The whole process successfully removed the approximation.With the new distance,we present a new registration method based on nonlinear optimization.The proposed method has both of high efficiency and convergence probability,which is more applicable to on-orbit technique services.Compared to the ICP(Iterative Closet Point),the new method has greatly improved the computation efficiency while keeping the accuracy.Compared to the registration method with other continuous distance fields,it has improved both of the global convergence and accuracy at the same time.Finally,we present an applied research of non-cooperative pose measurement based on the above two methods.We performed the non-cooperative pose measurement on a 1U size Cubesat with a commercial stereo vision system and compare it with high-precision turntable.Then we analyzed the effect of disparity deviation to pose result based on the experiment,and demonstrated the superiority of our stereo matching method.We also made an uncertainty analysis for pose result from disparity fluctuation,which validated the robustness of pose measurement method with point set.