Fish Tracking Based on Sonar Images by Means of a Modified Optical Flow

Underwater studies are becoming one of the trending topics with the massive technological advances achieved in the last two decades. Livestock monitoring is without a doubt a very interesting field of research to be explored, therefore, intelligent systems are required to reduce the human efforts and intervention. Our paper presents a fish tracking method based on a local optical flow application, with corners and key-points extracted out of the estimated fish contours instead of the traditional local forward optical flow application, we use a forwards and backwards corners tracking, followed by a contours’ movements scoring phase, where we try to attribute the right track to the contours extracted even when some of the corners are lost and under concurrent tracking outcomes. The results show a more stable performance of the approach and more resistance to the smoothness constraint violation.