Virtual-Real Registration Of Mobile Augmented Reality Based On Natural Features

Virtual-real registration is the principle issue of Augmented Reality, which determines the performance of real-time, robustness, running speed and storage space, etc. of Augmented Reality applications. The current development of mature AR applications relies mainly on identifying markers to overlay virtual information, which has good performance in speed and real-time, but limited usage scenarios. In AR system based on natural characteristics, the use of traditional method is difficult to meet the needs of specific terminal real-time reconstruction. So, how to ensure the accuracy of natural scene registration, efficiency and robustness, becomes the main problem to focus on for AR applications.This paper puts forward a new plan for the implementation of AR registration based on the natural characteristics, aiming to meet the real-time and lower memory consumption demand of those terminals, and building a prototype AR system based on the theory, to prove the feasibility of the scheme. Traditional AR system based on natural features extracts features from images and matches them, and the matched points can be used to reconstruct 3D point cloud. This paper analyzes the limits of mainstream feature point extraction and descriptor generation algorithm, and proposed FAST(Features from Accelerated Segment Test)-LDB(Local Difference Binary) algorithm to accelerate the process. The LDB algorithm reduces the storage space descriptor occupies by generating binary feature point descriptor, which is more suitable for AR application. System is divided into two stages, reconstruction is offline stage and registration is online stage. To ensure the real-time system operation, the system reduces the burden of device by different processing of the two phases.Experimental results show that in the similar cases of the environment, the speed of feature point extraction is 3 times faster than the original method, and the match is 4 times faster. The storage space occupied lower by 66%. In terms of the virtual-real fusion, the flaw of the prototype system is not implementing the occlusion handling between virtual and real worlds.