Identifying individual fish in videos
Underwater video cameras are commonly used in the fish farming industry for inspecting the health and wellbeing of the fish, monitoring the feeding process, etc. However, for many applications it is desirable to do automatic analysis of the images, for instance for collecting statistical information of the fish size, fish distribution in the cage or fish behavior.
SINTEF has for the client AKVA Group investigated possibilities for automatic identification and segmentation of individual fish in schools of fish based on short video sequences with stereo cameras. Advanced Variational and PDE (Partial Differential Equations) based image processing and segmentation techniques have been used, utilizing the mutual information in the two stereo cameras and in sequences of video, and by including prior models for fish shape probabilities.
Related article:
S. Clausen, K. Greiner, O. Andersen, K.-A. Lie, H. Schulerud, and T. Kavli. Automatic segmentation of overlapping fish using shape priors. Lecture Notes in Comput. Sci., Vol. 4522/2007, pp. 11-20, 2007. DOI: 10.1007/978-3-540-73040-8_2
If you are interested in more information please contact
Tom Kavli
.
 |
|
Examples of fish segmentations with overlap between different fish. |
