Research On Tracking Registration Method Based On Mobile Augmented Reality

Augmented reality(AR)technology enhances people’s perception of the real world by combining virtual and real objects in a real environment.In recent years,with the rapid improvement of the performance of mobile devices,the application of augmented reality systems based on mobile terminals has become a current research focus and is widely used in many fields.Augmented reality mainly includes three key technologies: target tracking and registration,virtual-real integration and human-computer interaction.Mobile augmented reality systems use these technologies to realize virtual-real integration,making mobile augmented reality applications more diversified and people’s user experience of AR applications more realistic.After investigation and research,it is found that due to the complex conditions of the real environment,there are still problems such as cluttered background in the tracking registration process and inaccurate pose estimation when the target is occluded,which will lead to poor stability of the AR system.Therefore,improving the accuracy of tracking and registration and the accuracy of attitude estimation is of great research significance and value for the application of mobile augmented reality systems.Based on the above analysis,the main research work of this thesis is as follows:(1)Aiming at the problem that the tracking method based on the Siamese network cannot accurately distinguish the target foreground and the semantic interference background,the target is prone to drift and cannot perform scale estimation,tracking registration method for augmented reality based on Siam FC and DSST is proposed.The DSST filter is introduced into the Siamese network tracking process,and uses the directional gradient histogram to make up for the deep features in the Siamese network to suppress the drift to similar targets;the candidate targets transferred to the DSST are screened more accurately by Siam FC,which can alleviate the boundary effect of DSST;the feature detection and matching of the target to be registered is performed to complete the registration,which ensures the real-time performance and improves the robustness of the traditional augmented reality tracking and registration algorithm.(2)Due to the cluttered environment and severe occlusion,inaccurate vector field prediction will affect the quality of the generated keypoint hypothesis and the small angle between the two direction vectors used by PVNet to judge voting may lead to large voting bias,affecting the accuracy of target object pose estimation,a framework of 3D registered target pose estimation based on improved PVNet is proposed.On the basis of PVNet network,the pixel weight layer is added to screen out more accurate pixel point vectors,and considering the voting distance between the pixel and the key point hypothesis,a distance filtering voting scheme considering the voting distance is proposed.Based on the more accurate voting results,the Pn P algorithm is used to accurately locate and 2D key points to solve the 6D pose,so as to improve the performance of pose estimation.(3)Aiming at the application of AR in the field of experimental teaching in colleges and universities,on the basis of the above research algorithms,an experimental teaching material application system based on mobile augmented reality is designed,and an interactive experimental teaching material based on mobile augmented reality is realized.This textbook is applied to the interactive book "Introduction to Railways",which has a certain positive impact on the experimental teaching in colleges and universities,making the experimental teaching and display more vivid and effective.