A Study Of Markerless Registration Based On Mobile Augmented Reality

The Augmented Reality(AR)is an emerging technology of human-computer interaction based on the virtual reality.Compared with the virtual reality,AR is desighed to combine the virtual information with the real scence,so as to enhance the visual and sensory experience of the users.With the continuous development of mobile intelligent terminal,the computing performance continues to improve as well as the built-in camera,sensors and other additional devices become abundant,Mobile Augmented Reality(MAR)is constantly getting into the sight of ours.However,the mobile device is unable to compared with the PC either in the computing ability or storage capacity,it is necessary to pick the appropriate algorithm apply to MAR system based on the computation and storage of the mobile device.3D registration technology is one of the key technologies to realize augmented reality.So it is the focus of the research on this field that how to improve the calculation speed and reduce the computation amount of the algorithm.This paper gives a study on the problem of the 3D registration on the present stage and proposes the corresponding improvement method on the related algorithm.In this paper,a new method of markerless image recognition is proposed,which consists of three aspects,fearture detection,feature description and feature matching.Firstly in the feature detection stage,an Enhanced-BRISK(E-BRISK)algorithm is proposed based on the BRISK algorithm which makes the scale pyramid more abundant.Then a new descriptor named MA-FREAK upon mobile acceleration sensor based on the FREAK descriptor is proposed to make a better discrimination of the feature descriptor by increasing the gravity information obtained by the sensor.At last,a fast feature matching method based on MA-FREAK is proposed in this paper,it can give a priority matching to the descriptor with the coutour information,which can improve the speed of feature matching.The experiments after each improvement show the algorithm of this paper can keep the speed of 3D registration and improve the robustness.