Depth estimation by an adaptive depth from motion algorithm

The utility of autonomous underwater vehicles (AUVs) is increasing due to their capability to inspect the environment where they operate, especially in risk (e.g., deep) areas. Vision sensors, on board an AUV, can provide a large amount of information and are often used to extract important visual cues in a scene. The authors discuss the problem of depth extraction by using a depth from motion (DFM) procedure robust to the inherent ill structure (e.g., bad illumination) of the environment to be investigated. This algorithm is capable of extracting 3D information from a sequence of visual images, and can be useful to build up the 3D relief of a scene. The robustness is due to the implementation of an auxiliary regulation procedure aimed at improving depth reliability when the data quality is insufficient. Comparative results are shown, which demonstrate the goodness of the regulation approach