Accuracy Assessment of Three-Dimensional Vehicle Edge Features Generated with Aid of Photogrammetric Epipolar Lines 2018-01-0530
Photogrammetry is widely used in the automotive and accident reconstruction communities to extract three-dimensional information from photographs. Prior studies in the literature have demonstrated the accuracy of such methods when photographs contain easily-identifiable, distinct points; however, it is often desirable to determine measurements for locations where a seam, edge, or contour line is available. To exploit such details, an analyst can control the direction that the epipolar line is projected onto the camera plane by strategic selection of photographs. This process constrains the search for the corresponding 3D point to a straight line that can be projected perpendicular to the seam, edge, or contour line. Thus, the goal of this study was to evaluate the modeling accuracy for cases in which an analyst uses epipolar lines in a workflow. To do so, artificial images were created using a computer-generated camera within a computer-assisted drawing environment to allow for a known reference model to compare with results produced using photogrammetry. A systematic study was undertaken by modeling two-dimensional curves on a plane, three-dimensional curves on a curved surface, and then curved edges on a vehicle model. Each model was assessed for accuracy, and the sensitivity of the accuracy to camera placement was carefully examined and explained. Finally, the procedures were applied to an actual vehicle, for which the results were compared to a 3D laser scan of the vehicle. In conclusion, the average residual error between a photogrammetry model created with the aid of epipolar lines and 3D scanned points for a three-dimensional vehicle edge feature was 1.69 mm (SD = 0.55 mm).