Line Segment Grouping and Linking: A Key Step Toward Automated Photogrammetry for Non-Contact Site Surveying

Abstract

The surveying technique of photogrammetry has been proven to be safe, efficient, and inexpensive to extracting spatial data (i.e. shape, size, position) of infrastructure from collected photos. These data are useful in many infrastructure and construction applications such as quality control and quantity take-off. However, photogrammetry has not been widely used in construction projects because much effort is still needed to convert the image data into a three-dimensional (3D) geometric model in the physical space. More specifically, it demands manually marking object vertices and edges and referencing them across photos. To alleviate this situation, this paper proposes a novel method to group and link line segments in order to automate the object vertices and edges marking process. In the proposed method, small, discontinuous line segments are first detected in an image, and classified into different sets based on the vanishing points they belong to. Within each set, a novel algorithm is created to link the line segments into long line segments. Next, the corner information is derived from the image by a classic corner detector, and utilized to assure the actual locations of endpoints of each obtained long line segment. By removing the remaining ungrouped or unlinked small line segments, the final result is a set of complete and accurate line edges readily to be used in the ensuing step of photogrammetric process. So far, the grouping and classification of line segments have been implemented. The test result from outdoor and indoor images indicates effectiveness and promise of the method in facilitating the automation and broad applications of photogrammetry in construction.