Visual tracking using spatio-temporal context template set learning

摘要

Correlation filter (CF) tracker has many advantages in visual tracking. But the update strategies of most CF trackers today such as STC are so simple that it can not handle more complex situations. For this problem, the paper presents a novel CF tracker based on templates, spatio-temporal context template set (STCTS) tracker. The algorithm not only improves the tracking strategy, but also improves the strategy of model update. While tracking, the model is updated by the linear combination of the template set and the template set will be updated in each frame. Experimental results show that our tracker has a good effect, and can handle more complex cases than STC.