A Cooperative Approach for Pattern Recognition in Underwater Scene Understanding by Multi-Sensor Data Integration

Surveying the oceans' floors represents at the same time a demanding and relevant task to operators concerned with marine biology, engineering or sunken cultural heritage preservation. Scientific researchers and concerned persons combine their effort to pursue optimized solutions aiming at the mapping of underwater areas, the detection of interesting objects and, in case of archaeological survey mission, the safeguard of the detected sites. Among the typical tools exploited to perform the cited operations the Autonomous Underwater Vehicles (AUVs) represent a validated and reliable technology. These vehicles are typically equipped with properly selected sensors that collect data from the surveyed environment. This data can be employed to detect and recognize targets of interest, such as manmade artefacts located on the seabed, both in an online or offline modality. The adopted approach consists in laying emphasis on the amount of regularity contained in the data, referring to the content of geometrical shapes or textural surface patterns. These features can be used to label the environment in terms of more or less interesting areas, where more interesting refers to higher chances of detecting the sought objects (such as man-made objects) in the surveyed area. This thesis describes the methods developed to fulfill the purposes of mapping and object detection in the underwater scenario and presents some of the experimental results obtained by the implementation of the discussed techniques in the underwater archaeology field.