Research On Pose Estimation Method Of Texture-Less Object In Complex Scene

With the rapid development of the manufacture of high-end equipment such as aerospace spacecraft,large telescopes and accelerators,the demand for precise measurement of three-dimensional geometric parameters such as morphology,size,position and attitude of large components is increasingly strong.The three-dimensional(3D)geometry parameters of large components are obtained by solving the coordinates of specific points of the measured cooperative object through a spherical cone prism.The pose estimation of the measured cooperative object is the basis of high precision measurement.Accurate pose estimation can effectively improve the accuracy of measurement.Spherical pyramid prism is texture-less object.Its surface is smooth,reflective,and has no specific visual features.In industrial application scenes,there are disturbances such as chaotic background,dynamic illumination change,defocus of imaging system,relative movement of imaging system and object,specular highlight and occlusion.The pose estimation of spherical cone prisms in industrial applications is an urgent problem to be solved.Therefore,it is of great significance and application value to study the pose estimation method of texture-less object in complex scenes,which provide accurate pose information for the aiming unit of the instrument.On the basis of analyzing the pose estimation method of texture-less object,the pose estimation method of texture-less object in complex scenes is studied in this paper.mainly completing the following research work:1 、 To solve the problem of texture-less object detection in complex scenes,a method of texture-less object detection based on warped hierarchical template tree was proposed.Firstly,Open GL is used to simulate the images of texture-less objects from different viewpoints,and the template set of cumulative gradient direction coding is established;Then a warped hierarchical template tree based on prior knowledge and feature similarity is established by combining the viewpoint prior knowledge of different templates;Finally,template matching and sparse representation are integrated to detect object.The object region is provided for image restoration,and the initial pose is provided for pose estimation.The method has higher robustness and accuracy in complex scenes.2、Aiming at the hybrid degradation caused by defocus blur,relative motion blur and specular highlight of texture-less measured object image in complex scene,a hybrid degradation model was proposed to restore texture-less measured target image.Firstly,the binary reflection model and defocus degradation model are integrated to establish the hybrid degradation model.Then,the gradient direction prior constraint function was constructed to estimation blur kernel.Guided filtering is introduced to suppress ringing effect to achieve texture-less object image deblurring.Finally,specular highlight is detected in the object region,and the adjacent brightness constraint is used to achieve specular highlight separation.The restored image is obtained,providing high-quality image for pose estimation.3 、 Aiming at the pose estimation problem of texture-less object in complex scenes,a pose estimation method of texture-less object based on multi-constraint search was proposed.Firstly,candidate correspondences are searched on the vertical line of the projected contour.Then,considering the edge direction and region histogram constraints,the multi-objective constraint is constructed to search the correspondences.Finally,weighted iteration is used to estimate the pose.The method has high accuracy in complex scenes.4、The pose estimation experimental system of texture-less cooperative object was established.Texture-less object detection and pose estimation algorithms and modules are developed.And the proposed method is verified.Texture-less cooperative object detection and pose estimation algorithm module can be applied to the aiming unit integrated into the precision measurement instrument of 3D geometric parameters.The proposed method can realize the pose estimation of the measured cooperative object with spherical corner prism under the clutter background,dynamic light,motion blur,defocus blur,specular highlight and partial occlusion.The texture-less object detection and pose estimation algorithm modules provides a support for the development and engineering of the prototype of 3D precision measurement instrument for large components.