Detection And Tracking Of Multiple Planar Objects For Mobile Augmented Reality

In the field of augmented reality,in order to make virtual objects interact with the real world,it's necessary to perceive the state of objects in the real world.Therefore,the detection and tracking of planar objects is an important part of the augmented reality system.The augmented reality based on these technologies has a wide range of application scenarios.The detection and tracking of planar objects recovers the 6-dof pose of the planar object At present,planar object detection and tracking in the case of large viewing angle,multi-scale,occlusion,illumination changes,etc.is still an urgent problem to be solved.In this regard,this thesis proposes an algorithm that can quickly detect multiple predefined planar objects in the above scenes and restore the pose of the object at the speed of 100FPS on the mobile end,which has reached the state of the artThe main contributions of this paper are as follows1)A top-down planar object detection algorithm is proposed.The region proposal method is used to determine the possible areas of the planar objects,thereby obtaining the approximate position and scale of the objects,which makes object retrieval and alignment easier.In addition,a verification algorithm is used to prevent the result from being incorrect.2)Based on the previous algorithm,a keyframe-based planar object tracking framework is proposed.The direct method is used to track the adjacent frames,the local map is established using the plane hypothesis,the system state is optimized using the re-projection error,and the local relocation function is added to cope with the tracking failure.Balanced in running speed,image continuity and accuracy,it achieves a high frame rate and can handle motion blur problems well.3)A tracking method for multiple planar objects is proposed.When multiple planar objects are tracked,3d registration is performed for relatively stationary planar objects,and the relatively moving objects are separately tracked.