Study On Three-dimensional Spatial Relationship Based On Single Image

With the rapid development of computer technology, the research of the computer vision is also constantly updated, in which the three-dimensional spatial relationships reasoning technology development, as one of the important research topics, is leaps and bounds at no matter at home or abroad. The three-dimensional spatial relationships reasoning technology from a single image has long been focused on. Unlike the three-dimensional spatial relationships reasoning from multiple images, it handles fast and convenient, saves time and effort, but so far there is not a complete set of solutions yet, it is necessary to carry out in-depth research for three-dimensional spatial relationships reasoning techniques based on two-dimensional images.This paper studies the three-dimensional spatial relationships from a single image in-depth. Using of the plural identities to detect the ground, the vertical plane and the sky, mainly including the following aspects:First of all, this paper proposed a region segmentation methods based on the edges-growth from a single, and labeled the segmented-regions. It detected the edge breakpoints through image edge detection, and then design a reasonable edge growth algorithm based on these breakpoints to make the image constitute a series of meaningful enclosed area. Then merge the unreasonable segmented regions, to achieve the final segmented image with labeling regions.Then, this paper analyzed the diverse features of the enclosed area of the image, including shading, color, contour feature, as well as the contour contained in a straight line, curve, the vertical line and parallel lines and other features of the areas, and established the feature vector of every area based on the above features.At last, we got the three-dimensional spatial relationships of the typical architectural scene based on the multi-visual features, and the collection and visual perspective projection invariants.