Research On Holo Lens Registration And Loop Closure Detection Technology In BIM Operations And Maintenance System

Building Information Modeling(BIM)can use digital technology to simulate the actual information of a building from design,construction to maintenance.The cost of a building in the use and maintenance stage can account for more than 80% of its total life cycle,which is the main cost investment period of the building.The traditional BIM operation and maintenance platform has many disadvantages,such as low visualization level,easy to lose the change information in the construction,static model can not interact with the site dynamically and so on.It is a trend to apply augmented reality(AR)technology to BIM system,which can make the guidance of maintenance site and virtual system connect seamlessly,more accurately assist workers to work on site,so as to greatly improve the operation and maintenance efficiency.At present,Microsoft HoloLens has the most industrial application potential in AR field,and its key technology is simultaneous localization and mapping(SLAM)technology.At present,HoloLens has two technical defects,one is that the 3D registration is not accurate enough and the other is that the error is accumulated.Aiming at these two technical problems,the main work and achievements of this study are as follows:Firstly,an RGB-D sparse direct method based on homogenization key points is improved.Firstly,the key points are extracted by quadtree homogenization,and then the camera pose is estimated by sparse direct method.This algorithm can make full use of all the information of the image.In the scene with insufficient texture information,it improves the defect that the frame is prone to mismatching in pose estimation.Secondly,a closed-loop detection method based on visual word bag is studied and improved to solve the problem that HoloLens is prone to generate error accumulation when it works in a large range for a long time.In this method,the improved k-tree is used to build the visual word bag model,and the key frame selection method is improved to make the judgment of the similarity between images more accurate,so as to improve the accuracy of the closed-loop detection algorithm.Thirdly,the validity of the above two algorithms is compared and analyzed through experiments.The results show that compared with the traditional RGB-D SLAM system registration and closed-loop detection algorithm,the improved method has better robustness and shorter time.Finally,a simple and effective BIM operationand maintenance demonstration system software is developed,and the effectiveness of the algorithm is verified by the actual HoloLens application.