the trackers based on on-line boosting

Visual object tracking is a fundamental topic in computer vision and was addressed by many researchers in the last decades. Many tracking algorithms with different assumptions (e.g., static background, accurate 3D model) have been proposed.

  • single object tracking

  • model-free tracking (i.e., only the initial position of the object is known)

Summarizing, the challenge is to track any object which might undergoes various appearance changes by using as less prior information as possible. The method should be robust in the sense of partial and full occlusions, changes in illumination and background clutter.

 

Three trackers with different characteristics are presented on this webpage:

 

on-line boosting tracker semi-supervised tracker

beyond semi-supervised tracker

     
To be adaptive, e.g., to appearance changes, on-line updates are performed in a supervised manner (filled circles). By formulating tracking as a semi-supervised learning problem, only unlabeled data is used during tracking (empty circles). Additional information is used to extend the semi-supervised tracking approach by an adaptive and object specific prior which is updated in a supervised manner.
     
This approach is very adaptive, but may drift. This approach limits drifting, but may restrict the adaptation too much This approach is a combination of the two former approaches to be adaptive without drifting.