Ensemble Tracking

Abstract

We consider tracking as a binary classification problem, where an ensemble of weak classifiers is trained online to distinguish between the object and the background. The ensemble of weak classifiers is combined into a strong classifier using AdaBoost. The strong classifier is then used to label pixels in the next frame as either belonging to the object or the background, giving a confidence map. The peak of the map and, hence, the new position of the object, is found using mean shift. Temporal coherence is maintained by updating the ensemble with new weak classifiers that are trained online during tracking. We show a realization of this method and demonstrate it on several video sequences.

Semi-supervised ensemble tracking

“…In this experiment, we attempt to track a man through occlusion. In order to examine how can the proposed approach improve the tracking performance, we compare it with the tracking results of ensemble tracking [1]. For fair comparison, both trackers for the sequence are started with same initial detection results.…”

“…This is not expected in tracking domain. Motivate by the works in [1], we borrow the ensemble tracking idea to design the sequential classifier. Assume that we have a classifier at time instant k − 1:…”

“…[2] firstly proposed a method to adaptively select color features that best discriminate the object from the current background. Another important work is [1], which used an adaptive ensemble of classifier. [3] designed an on-line boosting classifier that selects features to discriminate the object from the background.…”

“…However, [6] does not exploit the temporal relation between video frames. In this paper, we will use the idea of [1] to enhance the semisupervised tracking. The presented tracking algorithm is developed under the framework of particle filter.…”

Application of binary descriptors to multiple face tracking in video surveillance systems

“…Как прави-ло, в таком случае для отделения отслеживаемого объекта от фона используется бинарный классификатор. Такой подход лежит в основе алгоритмов Ensemble Tracking [8] и Tracking-Learning-Detection (TLD) [9].…”