Recovery of the 3D Structure of a Sewer Pipeline Using Computer Vision

摘要

Recovering the exact three-dimensional shape of a pipeline allows quantitative analysis of possiblerndeformations, degree of intrusion by laterals, amount of sediments in the pipe and clear or occupied crossrnsectional area. By employing the latest advances in computer vision technology, it is possible to estimaternthe 3D structure using a single, moving, off-the-shelf camera instead of very expensive and specializedrnequipment like laser or ultrasound measurement devices. In the structure-from-motion analysis the coordinatesrnof a 3D point can be estimated by tracking the point in several images and by calculating therncorresponding camera matrices. Using a sparse cloud of points it is possible to construct a densernreconstruction using a suitable interpolation method to fill in the missing sections. The interpolation usesrninformation in the images to produce a model, which matches the collected images as well as possible.Thernprocess includes 5 key steps: The extraction of interest points using a suitable detector allows therncomputation of the epipolar geometry, which is used as a constraint while searching for more points to bernused in the reconstruction phase. As the final step a suitable method of visualization is used to producernmodels from which the requested statistical data can be extracted. The camera used should be calibrated tornproduce precise and robust results. Because most of the existing calibration methods used are based onrnnear laboratory conditions, a new flexible and easy-to-use method is developed. The new method allows fastrncalibration of the camera using a very simple calibration object, which makes the process easy to apply inrnthe field after a maintenance or repair of the camera system. Preliminary results suggest that the accuracy ofrnthe method is well below one pixel, which translates into a 1-2 pixel resolution in reconstruction (a few mm inrn300 mm pipe using a 640x480 camera). The main aim of this paper is to present a brief theoreticalrnbackground on the methodology and to review some preliminary results from the measurements in Helsinkirn2001.