Research On Rapid Spatial Positioning And Modeling Technology Based On Multi-source Data Fusion

Visual-based environmental perception is one of the most important technologies in augmented reality equipment,including camera location,object recognition and location in real scene.The difficulty is how to get depth information only by monocular camera,and then locate the camera and calculate the relative position between the object and camera from single image.The paper studies this problem,and develops a positioning and perception system on Microsoft Augmented Reality Device Hololens,which uses monocular camera to locate,identify salient objects and calculate their relative positions with the camera.Firstly,we elaborate the principle of PnP algorithm based on artificial signs,and design a error detection system based on polar geometry for PnP algorithm is susceptible to noise,and prove that the positioning system can run reliably in the application environment through experiments.After that,we analyze the advantages of global features and local features in case of object recognition.We fuse SURF and multiple global features to train at the same time,and a selfadaptive weighted fusion algorithm is used to fuse several features to improve the recognition effect.In the aspect of object location,a fast pose calculation method is adopted.This method uses virtual camera and geometric model of object to generate 2D model for template matching,and uses genetic algorithm optimized by simulated annealing to quickly get the best matching pose.Finally,the system structure and operation process on Hololens are described.The paper introduces the method of multi-source data fusion on Hololens and the integration of each module,and carries out a number of test experiments on the system to prove that the performance effect in each scene can meet the requirements,the camera positioning accuracy meets the requirements of rough positioning,and the time and accuracy of object recognition can meet the standards.