Measuring the Shape of Sewer Pipes from Video

In this paper we address the problem of automatic measurement of the shape of sewer pipes. We describe a method for recovering the interior shape of a sewer pipe from a video sequence that is acquired by a fish-eye lens camera moving inside the pipe. The method is based on solving the general structure-from-motion problem by tracking interest points across successive video frames. Here the interest points are points where the image intensity changes rapidly due to irregularities in the surface texture of the pipe. The experiments with real videos of concrete pipes show that the shape of a sewer pipe can be recovered solely from the video. The proposed method can be additionally used in other applications to recover the scene structure from video sequences taken by a calibrated fish-eye lens camera.