| Summary: | A system which quantifies the number of food pellets eaten by salmon in a sea cage would be beneficial to both fish farmers, and researchers. Such a system could be used for reducing food wastage, determining time-related feeding patterns, ensuring fish receive the correct dosage of drugs, etc. We developed algorithms for detection and counting of food pellets from recorded video image sequences, which could be used to determine actual feeding rates in a sea cage. The number of food pellets eaten over time is determined by counting the number of pellets not eaten. The method involves counting the number of pellets of a known size, falling through the view area of an underwater video camera. The size of the view area of the camera varies with the size of the food pellets used. The pellets, which appear white underwater, are counted as they enter the view area of the underwater camera, and are tracked in this area to avoid recounting. For each video frame in the sequence, image preprocessing is done, followed by object detection, object classification, and object tracking and counting. Original algorithms were developed for this project to automatically threshold images, track objects in consecutive frames, and count the objects entering the view area of the camera. The algorithms were implemented on a personal computer based image processing system. Experiments were carried out to test the algorithms with pellet densities used in actual feeding situations. The utility of the algorithms was confirmed by the experimental results. The average count error for the tests performed was approximately +/-10%. The recommended improvements to the counting algorithms should significantly reduce this error. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|
|---|
