Abstract

Underwater video is currently being used by many scientists within NMFS to study fish populations and to do a census of species. The Southeast Fisheries Science Center uses such systems to assess populations of reef fish in the Gulf of Mexico. The manual nature of this analysis is time consuming and labor intensive. The goal is to develop software tools and algorithms to automate fish counting in underwater video. The main step is to recognize the presence of fish in the images and track the locations of individual fish from frame to frame. Recognizing the presence of fish is achieved via a histogram thresholding technique. Tracking is performed using a simple linear motion model. This allows automated counting of the number of fish in a time segment and exporting information about each fish. The relative size and shape of each individual fish changes as its location and viewed aspect change. However, most of the fish imaged will be moving against a more or less stationary background, and that motion is used to aid in detection. Enumerating fish targets imaged during a given time period requires tracking the location of each individual through the subset of images where it appears so that it is not counted multiple times. After each region containing fish is isolated, region growing is used. This makes possible the accurate counting of the number of fish and rejection of isolated regions which are not fish. Results demonstrate the performance of the proposed fish counting technique.

PDF Preview