Underwater Tracking: A New Way to Follow Moving Targets
The Problem with Underwater Tracking
Tracking objects underwater is notoriously difficult. The water is filled with particles and debris that obscure visibility. Sonar videos, commonly used for underwater tracking, face issues like:
- Similar-looking backgrounds
- Changing visuals
These challenges make it hard for traditional tracking models to perform accurately.
Introducing OFDTrack: A New Approach
A new model, OFDTrack, aims to overcome these obstacles. It leverages optical flow to identify moving targets. Optical flow acts like a movement map, helping the model maintain tracking even when the target is momentarily lost.
Key Features of OFDTrack
Optical Flow Tracking
- Maps movement to keep track of targets.
Smart Template Updates
- Adapts to changes in the target's appearance, such as shape or size variations.
Propeller Disturbance Correction
- Includes an additional tracker to correct errors caused by underwater vehicle propellers.
Testing OFDTrack's Performance
To evaluate OFDTrack, videos of the same underwater vehicle were recorded in three different locations. The recordings were made using sonar and drone-mounted cameras. The results showed that OFDTrack outperformed other tracking models.
