CONDENSATION—Conditional Density Propagation forVisual Tracking
International Journal of Computer Vision
Object Tracking Using Deformable Templates
IEEE Transactions on Pattern Analysis and Machine Intelligence
On sequential Monte Carlo sampling methods for Bayesian filtering
Statistics and Computing
IEEE Transactions on Pattern Analysis and Machine Intelligence
Democratic Integration: Self-Organized Integration of Adaptive Cues
Neural Computation
A tutorial on particle filters for online nonlinear/non-GaussianBayesian tracking
IEEE Transactions on Signal Processing
| Hi-index | 0.00 |
In Bayesian-based tracking systems, prediction is an essential part of the framework. It models object motion and links the internal estimated motion parameters with sensory measurement of the object from the outside world. In this paper a Bayesian-based tracking system with multiple prediction models is introduced. The benefit of multiple model prediction is that each of the models has individual strengths suited for different situations. For example, extreme situations like a rebound can be better coped with a rebound prediction model than with a linear one. That leads to an overall increase of prediction quality. However, it is still an open question of research how to organize the prediction models. To address this topic, in this paper, several quality measures are proposed as switching criteria for prediction models. In a final evaluation by means of two real-world scenarios, the performance of the tracking system with two models (a linear one and a rebound one) is compared concerning different switching criteria for the prediction models.