Underwater video analysis for Norway lobster stock quantification using multiple visual attention features

Underwater video is being increasingly used to assess the impact of human activities in marine habitats, as a complementary tool for the assessment of commercial stocks. But, analysing video images manually to study and evaluate marine habitats is a lengthy and tedious task. This paper proposes an automatic method to detect the Norway lobster (Nephrops Norvegicus) an important east-Atlantic and Mediterranean wide-distributed commercial crustacean species, in order to reduce the time and effort it takes marine scientists to manually quantify them. Here, the detection procedure follows a human visual attention model. Three visual attention features are considered: intensity map (IM), edge map (EM), and motion map (MM). The work is composed of two main parts: first the three feature maps are extracted; then, all candidate regions are processed and categorized in view of lobsters detection. Experimental results show that the proposed methodology is able to reliably detect candidate regions after combining the partial results.