Planar Object Detection And Its Application Based On Local Feature Descriptors

The augmented reality technique is one of the most popular techniques in recent years and it has good prospects.Augmented reality technique based on planar objects is an important part of it as it's more stable and efficient and you may have seen it in many situations of our life.However,as the higher and higher requirements of speed,stability,generality,its popularization has been hindered.So there are great opportunities and huge challenges in this field.In this thesis,I proposed a series of techniques about planar object detection and track-ing.It includes a local feature descriptor based on color and rotation-invariant,a workflow of planar object detection based on local features,a framework that can detect and track multiple targets simultaneously,a solution that can make augmented reality applications cross multiple platforms.The descriptor above is proved to be better than SIFT in the field of planar objects detection.The workflow above contains a series of algorithms such as keypoints selection based on octree,fast approximate nearest neighbor,RANSAC,inverse inliers expansion,etc.And it's more stable and faster than the traditional planar object detection workflow.Finally,by integrating these techniques,an entire pipeline that can detect and track multiple planar targets in real time on mobile platforms is implemented.These techniques,in fact,can be used in even more situations of augmented reality based on planar objects.