Occupancy Grid Mapping from 2D SONAR Data for Underwater Scenes

Abstract

Nowadays, creating a 3D reconstruction of the underwater environment is an important and non-trivial task, as it depends on the application environment and the main goal of the reconstruction, as well as on the information to be extracted on the final map. These reconstructions can allow robots to perform inspection and monitoring or cleaning tasks in different conditions, namely in dams. In these structures, it may sometimes be necessary to activate a cleaning system so that there is no risk of accidents. Therefore, in this work, a method is proposed to obtain a 3D reconstruction of the scene under the sensor based on a Mechanical Scanning Sonar (MSIS) and an external visual-based localization source. The localization is performed by a previously developed system based on the tracking of a visual ArUco marker. The main goal of this preliminary study is to evaluate the performance of MSIS in mapping tasks. Moreover, three different techniques are introduced to filter the measures and ensure that the final information is correct without affecting the systems in terms of memory requirements. Thus, the evaluation is carried out through some experiments on an experimental pool with some objects with different characteristics and building materials. The results show that a value of 50% of gain is the best for the tested materials. Moreover, the shape of the objects can be correctly identified and the obtained dimensions of the objects are adequately estimated. Moreover, the results of a discretization process in terms of x and y coordinates are presented, which allows to reduce the storage cost of the final map.