A Bayesian Network for online evaluation of sparse features based multitarget tracking

Abstract

Online evaluation of tracking algorithms has received attentions in computer vision community to detect failures and apply correction methods for achieving better performances. In this paper, a novel online evaluation framework is proposed for a multitarget feature points based object tracking. An online partial least square regression and correlation model is constructed from short trajectory histories for the tracks. The model allows to estimate the state of one track from the other track states. The core idea for the method is creating a virtual reference data for evaluation from the learned model. The proposed self-evaluation mechanism is presented as a Dynamic Bayesian Network. The method is evaluated on a simulation data for tracking feature points from a pedestrian.

Temporal validation of Particle Filters for video tracking

“…Empirical thresholding is broadly applied to detect changes, limiting their application to unseen data. Similar approaches also exist for non-Particle Filter-based approaches focused on feature accuracy [14], filter switching [24] or multihypothesis similarity [15] [16]. All these approaches are making use of data-dependent manually selected thresholds (computed offline) for change detection.…”

“…Such uncertainty model is approximated by sequential convolutions of mixtures of Gamma distributions whose number of mixture components is selected via modified informationbased criteria. By applying hypothesis testing over filter uncertainty models, the parameters required for detecting inconsistency are automatically determined, unlike empirical-based approaches [15] [12][13] [16]. The proposed approach is included in a framework for online performance evaluation of video tracking [13].…”