Smart Observer for Working Conditions in Distant Water Fisheries
Distant water fishing vessels go far out in the sea for at least half a year before returning. Fishermen can easily be exploited by being forced to work for long hours onboard, with high work intensity. Working on real CCTV videos on fishing vessels from Taiwan, in this talk, we demonstrate how an on-vessel system can be designed to monitor fishermen working conditions using a combination of AI and statistical methods. To replace manual record of fishermen work hour attendance, we designed an on-vessel face recognition based work hour attendance system with both offline caching and online updating of attendance records. Further, we show how real-life videos can be very difficult to analyze and thus the monitoring system needs to consider different AI and statistical techniques to produce feasible results. Compared to the global fishing watch (GFW) satellite data that is popularly used for vessel monitoring worldwide, our on-vessel monitoring system, called Smart Observer, gives a more accurate estimation of the working hours of fishermen. Our work hour estimation error is within 30 minutes in a span of 24 hours, while that of GFW is around 90 minutes. We encourage future researchers to work on how to integrate on-vessel video data with satellite data for a much more accurate estimation.