Fish population estimation and species classification from underwater video sequences using blob counting and shape analysis

Abstract

Fish population estimation and classification of fish species have been an integral part of marine science research. These tasks are important for the assessment of fish abundance, distribution and diversity in marine environments. We describe an efficient method for fish detection, counting, and species classification from underwater video sequences (UWVS) using blob counting and shape analysis. The video sequences were obtained with a moving camera resulting in rapid viewpoint changes thereby making it difficult to employ motion detection schemes in extracting fish images from background. Video preprocessing involved blackening out the corals from the underwater videos. This is done in order to effectively estimate fish count in the environment, though excluding those that are against a coral background. We then applied histogram comparison to initially blacken out the occlusions using blue and non-blue templates obtained randomly from the UWVS. We then introduced an erasure procedure to further aid in removing the coral background For fish detection, Canny edge detection was applied to extract fish contours. After the latter have been delineated, blob counting is then employed to in order to compute the fish count. Due to rapid frame changes, the average fish count per unit time is computed from the counts in each frame. For shape analysis, blob size is initially estimated and when a threshold is exceeded, Zernike moment-based shape analysis is performed on the blob for comparison with moment signatures of selected fish species stored in a database. The label of the best matching moments identifies the species of the fish blob. The shape-based classification algorithm is designed to identify the two most common species of fish found in the Tubbathaha reef in Sulu Sea, Philippines.