Search and rescue in a marine environment is a demanding, year round responsibility. On an average day the Canadian Coast Guard responds to 9 search and rescue incidents, assists 68 people and saves 18 lives. The ocean is a dynamic and unpredictable space to operate in for both machines and people. Saving lives depends on years of training and experience as well as reliable and effective technology.
Computer vision based AI models are beginning to be trialed in order to further enhance the capabilities of sea rescue organisations across the globe. These systems are helping to locate and protect individuals in danger, and even automating some aspects of search and rescue.
The continued improvement of sea rescue AI technology will require high quality datasets made with image and video annotation. This blog will look at three promising applications that are already being put to the test in our oceans. In each case we will see how data annotation services, like Keymakr, can help by getting developers the training data that they need.
AI powered searching cameras
In a chaotic, confusing ocean context it can often be extremely challenging for rescue services to locate people or lifeboats that are lost at sea. Low light conditions, poor weather visibility, and shifting waves can sometimes mean that those in need of rescue are not spotted in time. Cameras equipped with machine learning algorithms can help to extend the vision of rescuers as they cover huge areas at a time.
AI models can be trained with images of small boats and people in water. With the help of annotated training data these cameras should be able to see things in changing ocean states that might be invisible to pilots and crew members. Being able to locate and drop help faster will undoubtedly save lives.
Bias in training data can often lead to models functioning poorly. For example if AI search and rescue systems are trained solely with images taken on rivers, they may not work as well at sea. Professional annotation providers can help developers achieve varied datasets. Keymakr, for example, has experience with creating bespoke images using in-house production capabilities, as well as expertise with data collection.
Unmanned rescue vessels
Offshore wind turbines are a vital part of the energy infrastructure of many countries. However, installing and maintaining these structures can be dangerous, with workers operating many miles from land in often unforgiving conditions. AI powered autonomous rescue vehicles are now being developed that can come to the aid of energy workers should they issue a distress call.
A network of these vessels could create a safety net around offshore wind farms. Guaranteeing that any worker can be reached quickly in any conditions, and without risking the lives of search and rescue services.
Unmanned rescue vessels are required to navigate in rough seas and locate the people they are meant to retrieve. When safety is the goal it is essential that AI models are trained with image and video data that is precise. Keymakr ensures precision in training datasets by employing three levels of human verification, in addition to an automated sense check.
Automated rescue rafts
AI is also being used to develop self navigating rafts that can find people amidst the shipwrecks and debris. These vehicles can be dropped from helicopters and are then able to maneuver towards those in need of help. Upon reaching their target they can expand, effectively becoming a life raft. This system can reach people who may be drowning quickly and keep them protected until rescue helicopters and personnel arrive.
Keymakr’s in-house annotation teams work with an innovative annotation platform. This technology means that annotation jobs are given to those best equipped to complete them, as well as allowing multiple annotators to work on video data at the same time.