State of the Art:
Perception comprises automatic target recognition (ATR), tracking, and Simultaneous Localization and Mapping (SLAM) for localization from forward looking sonar and video. For feature extraction there is a significant published bibliography. Dealing with significant background noise and extracting supplementary characteristics regarding their size and shape or even their associated shadows are often required to improve the discriminability. However this kind of features can result in less reliable landmarks that could induce errors.
New generation of high frequency forward looking sonars produce acoustic images with considerably more resolution and higher refresh rates. Harris corner detector is used in combination with a multiscale Gaussian approach to detect features after equalizing the inhomogeneous insonification of the images. However this approach does not provide a description of the feature that guarantees its distinctiveness across other images.
Progress beyond the State of the Art:
The principle effort will be to develop improved feature extraction and localization using Probability Hypothesis Density (PHD) filters. The aim is to find a reliable method for automatic feature extraction in acoustic data to make features more discriminable. Features will be used to map geometry in the World Model, and to feed a SLAM algorithm for localization. SLAM map will be initialized with the knowledge of the environment of the underwater oil‐field structures. SLAM is required to correct vehicle position. Together with the map‐based features the on‐board sensors will image natural/artificial unknown features that will feed the SLAM. So, SLAM should be based in the Probabilistic Hypothesis Density filter which it is not constrained to an accurate data association, leading to a persistent autonomous AUV that can be self‐localized even in the presence of false/spurious features.