Abstract

Underwater video is currently being used by many scientists within NMFS to observe, identify, and quantify living marine resources. Processing of video sequences is typically a manual process performed by a human analyst. Partial automation of this time consuming and labor intensive analysis process will make data from underwater video more cost effective and available in a more timely fashion. This work introduces a technique for automatic fish classification in underwater video. The technique is based on a series of processing steps. Background processing is used to separate moving objects from the still background. Object tracking is used in order to associate different views of the same object found in consecutive frames. This step is especially important since successfully recognizing and classifying one of the views as a species of interest allows marking all views in the sequence as that particular species. Feature extraction using Fourier Descriptors is used to extract characteristic information from the shape of each identified object. Finally, a nearest neighbor classifier is used to classify identified objects as one of the species of interest. Results demonstrate the performance of the proposed technique in terms of correct classification and false alarms for three species, namely trigger fish, red grouper, and yellow tail snapper.

PDF Preview