Underwater Tracking: A New Way to Follow Moving Targets

Tracking things underwater is tough. The water is full of stuff that makes it hard to see clearly. Sonar videos, which are used to see underwater, have problems like similar-looking backgrounds and changing visuals. This makes it difficult for regular tracking models to work well. A new model called OFDTrack is trying to fix these problems. It uses something called optical flow to find moving targets. Optical flow is like a map that shows how things are moving. This helps the model keep track of the target even if it gets lost. The model also has a smart way to update its template. This means it can handle changes in how the target looks. For example, if the target changes shape or size, the model can still track it. Another cool thing about OFDTrack is that it can deal with the mess left by the propellers of underwater vehicles. It has an extra tracker that helps correct any mistakes caused by this mess. To test how well OFDTrack works, videos of the same underwater vehicle were taken in three different places. These videos were taken using sonar and cameras on drones. The results showed that OFDTrack works better than other tracking models.