Research On Tracking Technology Based On Natural Features In Augmented Reality

Augmented Reality (AR) is a kind of technology that enriches the way which usersexperience the real world by embedding virtual objects into the real scene. ARtechnology has been applied to many areas widely, such as military, medical, educationand entertainment. AR technology integrates image processing, pattern recognition,computer graphics and human computer interaction to complete the perception of virtualobjects in the real world.Tracking technology is used to accurately complete the registration in real time.Tracking methods in the system can be divided into two categories, artificial markerbased approaches and natural features marker based approaches. In this thesis, we focuson natural features marker instead of artificial marker, because natural features markersare not restricted by the scenes. After detailed studying SURF features extraction anddescription, the KD-tree structure is used to perform matching tasks more efficiently.Most of the tracking methods subjected to environmental constraints, cannot meet thesystem requirements. A tracking method based on natural feature is proposed. Firstly, therandom ferns algorithm is used to estimate the approximate location of target, and thenextracting and matching features with SURF algorithm to get precise location of target.A virtual glasses try-on system is designed and implemented by augmented realitytechnology. System allows users to choose their favorite type of glasses through a simpleinteractive. The system uses a Haar trainer to automatically recognize face as systeminitialization, and then the result will be transmitted to the tracking system for tracking.The control system is used to reduce the accumulation of errors. The construction ofvirtual glasses model adopts OpenGL technology, then fusion the virtual with reality tomake the scene more realistic.Experimental results show that the tracking algorithm can effectively reduce theimpact of illumination, noise, affine transformation, and partially occluded. Also thetarget reappear in the scene can be accurately recognized and tracked.