Research On Methods For Detecting Vertical Corner Lines Of Buildings In Images

Reconstruction and digitization of single view urban sence images play an important role invarious applications of the modern society. In the fields of computer vision, image analysis andunderstanding, there are three main methods for single view image reconstruction: automatic method,interactive method and manual operation method. Contrary to traditional labor intensive but accuratereconstruction solutions, existing automatic algorithms are low-accuracy, and semi-automatic methodsneed the outlines of the buildings as the initial inputs.In this paper we study the detection method of vertical corner lines of buildings, i.e., theintersection lines of the vertical piecewise surfaces of buildings in images. The vertical corner linescan be used as the initial inputs to improve3D image reconstruction results, and can also be used asintermediate features to image-based localization of targets in urban environments. The searchingalgorithm proposed in this paper is a bottom-up and step-by-step pipeline processing. First, we get theline segments as the low-level features in the image using a fast and accurate line segment detectornamed LSD, which is with a false detection control scheme. Secondly, we cluster the detected linesegments to the corresponding vanishing points based on J-Linkage estimator, and the vanishingpoints can be used as mid-level features. Finally, according to some geometric constraints, thehigh-level features, i.e., vertical corner lines, will be detected through the new searching scheme. Theglobal parameters of the program are analyzed and set, and experimental results demonstrate goodperformances of our method in different scene images.In the paper we also propose a simple extracting method for the simple scene, i.e. using thecorners of buildings skyline to recover the vertical corner lines of buildings in vertical rectifiedimages. Simulation experiments are conducted to show its efficiency.