Research On Natural Planar Marker Recognition And Tracking For Augmented Reality

With the advancement of times and the improvement in living standards,demands for convenient and efficient interactive methods grow rapidly.Augmented reality(AR)is such a promising approach which can bring people new experience beyond imagination.In this paper,combined with augmented technologies based on artificial markers or feature tracking,we make deep studies on recognition and tracking of natural planar markers,which offers feasible solutions to applications of augmented reality within small scenes.To deal with the recognition problem of markers,an original natural planar marker recognition method based on grid is introduced in this paper,which can handle the situation of small marker area in the image by refined image similarity measurement and exclude images without marker efficiently by defining distinctiveness degree.In addition,this method will localize the rough area of the marker in the image,which provides good prior knowledge for the following camera pose estimation.In terms of the marker-tracking problem,a tracking method based on key-frame is proposed to get stable and reliable results.To reduce failures caused by motion blur,tracking based on line-matching is adopted to compute more reliable homography matrix and then obtain camera pose,which utilizes the matched lines information to get reliable point pairs by optimization process of iterations.To cope with the error of camera pose estimation and the inevitable accumulated error of tracking,a camera pose optimization method based on marker edge information is adopted to optimize the camera pose,which is solved by Levenberg-Marquardt(LM)algorithm together with Gaussian pyramid strategy.Furthermore,an effective strategy based on key-frame is introduced,which reduces the accumulative error effectively and lays a solid foundation for more satisfactory AR experience.The experimental results and comparison demonstrate that the natural planar marker recognition method based on grid can robustly work in tough cases such as small marker area in image,motion blur,partial occlusion,large tilt etc.In addition,the proposed keyframe-based tracking strategy achieves reliable results even in tough cases mentioned above while meeting the real-time requirement.