Automated Fish Counting Using Image Processing

This paper presents a simple method of counting feeder fish automatically using image processing techniques. A video of a school of fish is taken and every frame is processed singly and independently. The first step is to obtain blobs marking the positions of the fish. Several ways of accomplishing this task are discussed. Noise and background objects are filtered from the image of the blobs. Area information of the blobs is used to count the number of fish in one frame, and the average number of fish over all frames is then recorded. Experimental results show that the correct number of fish can be obtained for a school of 5, 10, 15, and 50 fish. Errors within frames increase with the number of fish, mainly resulting from the fact that area thresholding can be quite sensitive. Finally, a discussion about the method's effectiveness and possible improvements are provided.