Augmented reality system with planar homographies using building façade

Publication Type:
Thesis
Issue Date:
2017
Full metadata record
Files in This Item:
Filename Description Size
01front.pdf231.9 kB
Adobe PDF
02whole.pdf3.83 MB
Adobe PDF
In recent years, there has been widespread adoption of AR technology with various applications in the urban environment. The essence of an AR application includes the estimation of camera and object placement, which is still a challenging research problem. The first issue is that the structure of the real-world environment is usually very cluttered and involves many unforeseen uncertainties which cannot be generalised as a model. Secondly, the reprojection of the real-world scene to the camera’s viewpoint flattens 3D information down to a pixel level, which causes what is known as perspective projection. Fortunately, it is possible to recover the pixel level of the camera pose via planar homography. Methods commonly used in typical AR applications to recover the homography are based on feature-point matching, but many of them still suffer from the problem of perspective and camera movement. By taking characteristics of the urban environment, such as buildings which are widely seen on the street view, we propose a simple, yet efficient, approach to recover the camera pose from a single image. Our approach exploits the building façade of rectilinear structures portrayed by man-made structures (e.g. windows and brick structure). Using lines from these structures, our proposed algorithm can produce a rectified homography for each detected building façade, so we can then recover the fronto-parallel view of each building façade. Since these planar homographies are invertible, we can use them to recover the camera pose of the building façade for the object emplacement task. Finally, we implemented a tracker-based AR application that uses the idea of a fronto-parallel view that results in better matching effectiveness.
Please use this identifier to cite or link to this item: