Skip to Main Content
This paper describes a robust photogrammetric system for 3-D surface reconstruction. We use a linear camera calibration model with 14 camera parameters, which can be optimized using the least-squares solution. To address the non-linear lens distortion, a localized calibration concept is presented whereby different image locations have different sets of calibration parameters. A new subpixel target detection algorithm is proposed for accurate target center determination (with accuracy better than 0.02 pixel). The algorithm first uses zero-crossings of the second derivatives to determine an initial threshold for separating a target from its background. The multiple threshold functions are selected to calculate multiple potential target centers. The weighted average of these potential centers results in a reliable final target position. Comparing with other subpixel algorithms, such as best-fit circle/ellipse, template matching based approach or single threshold gravity point method; our algorithm is more accurate and can be applied to any shapes of targets. To minimize the possibility of mismatch of targets among different views, we adopt a new reference point based matching approach. The target pattern from a projector contains a number of reference points. The stereo correspondences start from the reference points. The appeared and position constraints are implemented to achieve error free matching, before the triangulation function is called to get the final 3-D coordinates. The system is able to achieve an accuracy (3σ) of 2.5 μm ( using calibration target with dimension 35 mm × 35 mm, 50 mm lenses and Jai M1 CCD with 1300 × 1030 resolution).