Triplet Contrastive Learning for Multiobject Tracking in Satellite Videos (opens in new tab)

Multiobject tracking (MOT) in satellite videos is pivotal for global and dynamic object monitoring. However, the movable objects such as airplanes, ships, and cars in satellite videos are usually small, with ambiguous margins and features. It is challenging to identify and associate the same objects between adjacent frames, especially for those of the same class. To capture the distinguishing features among objects across consecutive frames, we propose a triple contrastive learning tracker (T...