Analysis of Disparity Accuracy Factors in Disparity Image Generated Using Moving Images Taken From A Rotating Crane Kazama Kojiro (a*), Susaki Junichi (a), Ishii Yoshie (a), Nakaoka Shohei (b)
a) Department of Civil and Earth Resources Engineering, Graduate School of Engineering, Kyoto University
C1, Kyoto-Daigaku Katsura, Nishikyo-Ku, Kyoto, Japan
*kazama.kojiro.34n[at]st.kyoto-u.ac.jp
b) Advanced Technology Development Department of Tadano Ltd., Nittamachi-cho, Takamatsu-shi, Kagawa, Japan
Abstract
Rapid generation of three-dimensional (3D) maps is essential for automating construction sites. One promising technique involves generating disparity images from moving images captured by a monocular camera mounted vertically downward on the tip of a crane boom. Our proposed method calculates disparity by applying Semi-Global Matching (SGM) to image pairs rectified using the camera^s rotation angle, based on the assumption of a vertically oriented camera. However, the precise conditions under which the accuracy of these disparity images varies remain poorly understood. The factors influencing accuracy are significantly more complex for a rotating crane camera compared to standard translational stereo systems. For instance, while a larger baseline to height ratio typically improves accuracy in translational setups, with a rotating crane, this benefit can be counteracted by increased vertical disparity caused by the crane^s rotational movement, which in turn degrades SGM performance. Therefore, this study aims to establish the optimal conditions for acquiring high quality disparity images considering the practical application. To establish these optimal conditions, this study employs an approach combining experiments with actual images and comprehensive simulations. First, an experiment with actual images captured at a specific radius and height revealed a key trade-off, whichi is, as the baseline-height ratio increases, the height map^s Root Mean Squared Error (RMSE) initially decreases before increasing, indicating an optimal value exists. Building upon this finding, our simulation-based analysis systematically investigates a wider range of parameters including radius, rotation angle, and height. This analysis clarifies the general conditions for achieving high accuracy by assessing their impact on RMSE, disparity completeness, and vertical disparity. A thorough understanding of these influential factors is expected to facilitate the consistent and reliable generation of highly accurate 3D maps, advancing the automation of construction processes.
Keywords: photogrammetry, disparity image, quality assessment, three dimentional mapping
