Automated Image Analysis for Fisheries CCTV - Funded by The Scottish Government, Environmental and Forestry Directorate, Project No SYS/001/14
The project will develop a prototype automated image analysis system for determining the number (and, potentially, the length distribution) of all CCTV-observed fish of a given subset of the species caught by Scottish vessels. The output of the system will be used, together with data from other electronic sensors and log records, as a tool to establish compliance within a new regulatory framework known as the Catch Quota Scheme (CQS). The system will also provide data for fisheries science and biological research. CQS is a key Scottish Government policy objective in influencing the reform of the EC’s Common Fisheries Policy and managing fisheries more effectively in the future. A pilot project undertaken by Marine Scotland in 2009 established remote electronic monitoring and CCTV cameras as a mechanism for improving the data gathering on the observation of discards. But human interpretation of the CCTV footage generated by the trial was time consuming and as such represents an unsustainable burden for Marine Scotland which will only increase as the CQS develops and expands. The computer assisted video analysis system will be developed by a team at the School of Computing Sciences, University of East Anglia and produce summary statistics from the footage by fusing visual cues (i.e. colour, texture) and other data within a machine learning framework. Illumination correction forms an important stage in this process and is an area where the team have particular expertise. The work will investigate a number of strategies for dealing with noise, occlusions, pose variations and will produce estimates within a statistical framework. Sections of video footage where estimates exceed certain predefined confidence limits will be indexed and subject to further review by human experts. The project will explore a number of detection algorithms, initially focusing on haddock individuals and then, if time permits, broaden this approach to consider other commercially important demersal species such as cod, whiting, saithe, hake and monkfish.
Funding - £89,280
French, G., Fisher, M., Mackiewicz, M. & Needle, C. T. In Amaral S. Matthews, T. P. S. M. & Fisher, R. (Eds.) Convolutional Neural Networks for Counting Fish in Fisheries Surveillance Video, Proceedings of the Machine Vision of Animals and their Behaviour (MVAB), BMVA Press, 2015, 7.1-7.10. Note: Best paper Award.