3D Object Recognition And Pose Estimation For AR-Assisted Maintenance Information Enhancement

3D object recognition and pose estimation are the basis of AR-assisted maintenance information enhancement.At present,3D object recognition technology is insufficient in accuracy,real-time and pose estimation accuracy.In order to meet the requirements of the target recognition real-time and pose estimation accuracy of scene recognition.This thesis researched the local reference frame estimation and descriptor generation and matching based on global feature of point cloud.Meanwhile,a fusion shape texture global orthographic object descriptor and the 3D object recognition method are proposed.The effectiveness of the proposed method is verified in the assembly of the reducer.The main research contents of this thesis are as follows:(1)The generation principle and process of point cloud are explained.The scene point cloud preprocessing method and specific steps based on point cloud global feature for object recognition are researched.The 3D object recognition route and process based on point cloud are given.The 3D object recognition framework based on point cloud global features is proposed.(2)The local reference frame estimation method of texture point cloud target is researched.By introducing the distance weight,a new local reference frame estimation method is proposed.By adding color texture information,a global orthographic object descriptor with fusion shape texture information is proposed.The 3D object recognition method based on the proposed descriptor is researched.A descriptor matching method based on Pearson correlation coefficient is proposed.Finally,the descriptor generation and matching methods are obtained.(3)Based on the Challenge dataset,the performance difference between the proposed method and the GOOD,GASD and ESF in 3D object recognition is compared and analyzed.The relationship between descriptor dimension,color addition mode and other factors on proposed method is researched,and the setting method of each factor is obtained.The final results show that the local reference frame estimation method has better stability to the local missing target,and the proposed 3D recognition method has the optimal comprehensive performance in target recognition rate,pose estimation accuracy and time efficiency.The actual maintenance scene recognition experiment was carried out.The results show that the proposed method basically meets the requirements of AR auxiliary maintenance information enhancement in recognition accuracy and real-time performance.