The method for recognition underwater artificial objects

The article is devoted to the current problem of detecting underwater presumably artificial objects (AO) within the framework of a seabed monitoring mission using an autonomous unmanned underwater vehicle (AUV) equipped with a stereo camera. Such objects include, along with underwater industrial infrastructure objects, sunken objects, military objects, etc. In particular, a pressing task is to detect objects near pipelines and other communications. Detection of these objects involves their automatic localization and recognition of characteristics indicating the artificial nature of the object’s origin. In contrast to the traditional approach based on comparing input data with an a priori given model of the object, this paper proposes a method for direct recognition using computer vision methods of elementary geometric elements (GE) characteristic of the object in the input 3D point cloud obtained by processing a stereo pair of images. Such GEs include spatial segments (parallelepiped edges), spherical surfaces, as well as color, texture and other features. According to the proposed methodology, direct identification of the detected object with the adoption of decisions on subsequent actions will be carried out by the operator in offline mode based on the analysis of the received images and data from automatic algorithmic processing on board the AUV. The computational experiments performed with model data confirmed the prospects of the proposed method for detecting underwater objects.