Real-Time Siamese Multiple Object Tracker with Enhanced Proposals

Maintaining the identity of multiple objects in real-time video is challenging because it is not always feasible to run a detector on every frame. Motion estimation systems are often employed, but they either do not scale well with the number of targets or produce features with limited semantic information. SiamMOTION allows the tracking of dozens of arbitrary objects in real time through a proposal engine that produces quality features with attention, a region-of-interest extractor fed by an inertia module, and a feature pyramid network. Its comparison head efficiently matches exemplar and search-area pairs via a pairwise depthwise region proposal network and a multi-object penalization module. SiamMOTION is validated on five public benchmarks, achieving leading performance against state-of-the-art trackers.