Research On Registration Method Of Augmented Reality Point Cloud Model Based On Natural Features

Virtual and real registration technology is one of the key technologies in augmented reality,which plays an important role in enhancing the real-time and running speed of the system.The identification is performed by means of artificial markers,and the calculation amount of the algorithm is small,the registration accuracy is high,and the speed is fast.However,scene recognition and registration based on natural features has a wider application,but the requirements for real-time performance are higher.Traditional technologies are difficult to meet the requirements.It is necessary to focus on how to improve the accuracy of registration in natural scenes,speed up registration and improve Robustness and other issues.In this paper,based on the natural scene recognition and registration problem,a point cloud registration method based on genetic algorithm is proposed.The point cloud registration technology is improved,the accuracy and speed of registration are improved,and the method is applied to the enhanced system.The test and evaluation of the model displayed in the augmented reality,the main research work is as follows:(1)Data collection and processingIn this thesis,the point cloud data is collected,the KD tree is used to denoise the point cloud,and the planar feature is used to detect the edge cloud edge feature.The point cloud data is simplified and the required model data is modeled.(2)Feature point extraction and matching researchIn this paper,the principle of each algorithm of SIFT,SURF and ORB is expounded,and the experimental analysis is carried out.Among them,ORB has better effect of feature point extraction and matching.In order to meet the real-time requirements of the AR system,the ORB is used for feature extraction and matching.(3)Improvement of point cloud registration methodIn this thesis,the genetic algorithm and KD tree are used to improve the point cloud registration method.The genetic algorithm is used to perform global iteration.In the iterative process,new chromosomes are generated through crossover and mutation.The fitness function is used to evaluate the results of each generation.In the evolution direction of the population,the KD tree is used to divide the point cloud,and the K-nearest neighbor search is performed to improve the search speed,thereby improving the distance calculation speed of the point cloud in the iterative process and achieving the desired registration accuracy.(4)prototype system implementationIn this thesis,the internal parameters of the mobile device's camera projection are calibrated.The model detection module of the AR system is completed by Matlab.The qualified model is imported into the Android Studio project,and the virtual and real registration modules of the system are developed through Android Studio.Generate Android mobile app to implement augmented reality prototype system for mobile.The system can display the model in real time in natural scenes without the need for markers,and achieve the effect of enhancing the reality.