reference: Struck: Structured Output Tracking with Kernels
hot topic:
tracking-by-detection methods, treated as a classifiction task, use online learning techniques to update the object model
questions:
1) for these updates to happen one needs to convert the estimated object position into a set of labelled training examples, and it is not clear how best to perform this intermediate step.
2) label prediction is not explicitly coupled to the objective for the tracker
solutions:
1)we present a framework for adaptive visual object tracking based on structured output prediction.
2)allowing the output apace to express the needs of the tracker, avoid the need for an intermediate classification step
3)use a kernelized structured output support vector machine
4)for real-time application, we introduce a budgeting mechanism which prevents the unbounded growth in the number of support vectors which otherwise occur during tracking.