Research On Pose Estimation And State Detection Of Base Part And Assembly Parts For Augmented Reality Aided Assembly

Augmented reality can be applied into the augmented reality aided assistance of complex industrial products,which improves the assembly efficiency.In the process of the augmented reality aided assembly,pose estimation is the key technique for registration of the augmented reality aided assembly guidance information.However,in a manual assembly scene where the base part is large,the assembly parts has a single texture and the parts are small,it is not easy to arrange the markers,and it is not suitable to use the pose estimation technique based on the markers.The base part and the assembly parts have different characteristics during the assembly process,and a unified natural feature-based pose estimation method cannot be used.At the same time,an assembly state detection method based on real-time pose information of parts is introduced in the augmented reality aided assembly technology,which is used to actively trigger the assembly guidance information in production,which can ensure the assembly efficiency and the work quality of the workers and reduce inspection time,thereby shortening the product manufacturing cycle.For the target above,studies were made on following points: the combination pose estimation of base part and assembly parts based on natural features,state detection method based on real-time pose information of parts and the system development.The working results are as follows:1)Product assembly information modeling for part tracking was proposed.The assembly information was divided into offline data information of base part pose estimation,assembly part pose estimation and assembly part identification,assembly guidance information and assembly status detection information.The specific information content included in each information category,the storage method of the information content,and the organizational relationship between the information were determined.2)A combination pose estimation of base part and assembly parts based on natural features was proposed.Different pose estimation methods were adopted to compute the relative position: the SURF feature point matching method and the PNP method were used to estimate the pose of the large base part.The LINEMOD algorithm was used to recognize the small and textureless assembly parts.The KLT tracking algorithm was introduced into the LINEMOD algorithm to estimate the assembly parts to solve the problems that the LINEMOD algorithm was slow in the process of the assembly parts pose estimation and could not estimate the pose under certain observation angles.Under the local visual constraint of the assembly base,the combined estimation of the base part,the pose of the assembly part and the part recognition in the same scene coordinate system were realized.3)An assembly state detection method based on real-time pose information of parts was proposed.Based on the result of combination pose estimation,the pose relationship between the real part and the 3D model was established.In the augmented reality scene,the bounding box of the assembly parts model was realized at the real-time position and the standard assembly position of the parts,and collision detection based on the bounding box in the augmented assembly scene was realized for preliminary detection of the assembly state.Based on pose matching of parts,accurate assembly state detection could be realized.According to the assembly state detection result,the virtual assembly guide information was triggered.4)On the basis of the above research,a prototype augmented reality aided assembly system was developed.The assembly processes of the key parts of the automobile door were used as examples to verify the effectiveness of the method.