Algorithm for Recognizing an Underwater Pipeline from Stereo Images

Abstract

The problem of recognition of an underwater pipeline (UP) from stereo images using an autonomous underwater robot (AUV) is considered in relation to the initial position of tracking the UP, or to situations where the previous section of the UP is hidden by interference (submerged in the ground, hidden by algae, etc.). The final result of the identification of the UP section, visible by the stereo camera, is the calculation of its center line and the detection of the relative position of the AUV and UP in the camera coordinate system. The article proposes a recognition method based on the selection of visible UP boundaries (contours) on vectorized images of a stereopair. At the stage of vectorization, noise is eliminated, illumination is equalized, and the image is processed using the Canny method to obtain a binary image. The construction of UP contours is performed using the algorithm proposed by the authors, which is a modification of the Hough method. The main feature of the proposed algorithm is a relatively high performance due to a multiple reduction in the amount of information being processed. Reducing the volume of processed data is done by pre-sorting the line segments in the vectorized image, and by optimizing the computational scheme in the algorithm. The experiments also showed that the algorithm can detect the visible boundaries of the UP on blurry, non-contrasting images. The algorithmic basis of the method is described in detail, including: — search and construction of the most reliable UP boundaries using the method of the integral contribution of the line segments to the line formation; generation and selection of point features belonging to the surface of the UP (due to the constructed contours); calculation of the 3D direction of the center line; calculation of the center line of the visible section of UP; —calculation of the AUV position parameters relative to the UP required for the AUV control system. The centerline calculation is performed using the least squares method using point features belonging to the surface of the UP. The performed computational experiments on virtual scenes using the real texture of the seabed confirm the operability of the implemented approach and the possibility of its application for the inspection of underwater infrastructure.