Detecting planes and estimating their orientation from a single image

We propose an algorithm to detect planes in a single image of an outdoor urban scene, capable of identifying multiple distinct planes, and estimating their orientation. Planar surfaces are ubiquitous in man-made environments, and convey useful information about scene structure; detecting them from a single image can be very useful since it requires no parallax, and can be done when only one image is available. Previous methods have shown how this can be used for estimating camera placement, wide-baseline matching, and creating simple 3D reconstructions. Such methods fall into two broad categories: those that explicitly use the geometric properties of the image, such as vanishing lines [6] or texture [3], and those that use machine learning techniques to relate appearance to structure [5] – our method is motivated by recent examples of the latter.