Dynamical Gaussian mixture model for tracking elliptical living objects

Authors:
Guanglei Xiong;Chao Feng;Liang Ji
Affiliations:
Department of Automation, National Laboratory for Information Science and Technology, Tsinghua University, Haidian District, Beijing 100084, China;Department of Automation, National Laboratory for Information Science and Technology, Tsinghua University, Haidian District, Beijing 100084, China;Department of Automation, National Laboratory for Information Science and Technology, Tsinghua University, Haidian District, Beijing 100084, China
Venue:
Pattern Recognition Letters
Year:
2006

Citing 4
Cited 5

Unsupervised Learning of Finite Mixture Models

IEEE Transactions on Pattern Analysis and Machine Intelligence
LAFTER: Lips and Face Real-Time Tracker

CVPR '97 Proceedings of the 1997 Conference on Computer Vision and Pattern Recognition (CVPR '97)
Maintaining Multi-Modality through Mixture Tracking

ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
SMEM Algorithm for Mixture Models

Neural Computation

Pearson-based mixture model for color object tracking

Machine Vision and Applications
Ellipsoidal decision regions for motif-based patterned fabric defect detection

Pattern Recognition
Online clustering of high-dimensional trajectories under concept drift

ECML PKDD'11 Proceedings of the 2011 European conference on Machine learning and knowledge discovery in databases - Volume Part II
Extracting robust distribution using adaptive Gaussian Mixture Model and online feature selection

Neurocomputing
Time domain single-phase reclosure scheme for transmission lines based on dual-Gaussian mixture models

Engineering Applications of Artificial Intelligence

Quantified Score

Hi-index	0.10

Visualization

Abstract

In this letter, we present a novel dynamical Gaussian mixture model (DGMM) for tracking elliptical living objects in video frames. The parameters, which inform object position and shape, are estimated by using a traditional Gaussian mixture model (GMM) for the first frame. Instead of simply using the parameters of one frame as the initial state for the next frame, DGMM employs Kalman filter to predict these parameters to initialize the evolution of GMM for the rest of sequel frames. Compared with the simple way, the estimate by Kalman filter is usually much closer to the true state, the computation load for tracking is reduced significantly. In addition, ''split'', ''merge'' and ''delete'' operations are integrated into DGMM to account for object splitting, merging and vanishing. This scheme can suppress the chance of getting stuck at a local maximum, led by a mismatch between the number of objects in the model and the actual number of objects. Results obtained from experiments on both simulated and real-world data demonstrate the effectiveness of our algorithm. DGMM has been successfully applied to track the early development process of urchin embryo in which the events of motility and cleavage occur.