Estimating the abundance of specific populations of commercially significant fish is critical in oceanography and fisheries science. Automated fish-counting provides necessary estimates, although fish are often caught using bottom and midwater trawls via fishing nets pulled by boats, meaning that automated methods face the challenge of being underwater. To overcome this problem, video data can provide much of the information that is typically collected from fish caught using traditional trawl methods, such as tracking and counting of fish, length/size measurement and species recognition. Methods from this project are being extended to more general challenges for automation, including multiple-target tracking and target shape segmentation from complex scenes captured by moving cameras.