Hadi Sadoghi Yazdi - Department of Engineering, Tarbiat Moallem University of Sabzevar, Sabzevar, Iran
Mojtaba Lotfizad - Department of Electrical Engineering, Tarbiat Modarres University, Tehran, Iran
Mahmood Fathy - Faculty of Computer Engineering, Iran University of Science and Technology, Tehran, Iran
Ehsan Kabir - Department of Electrical Engineering, Tarbiat Modarres University, Tehran, Iran

In this paper, an HMM/GMM statistical framework is presented for traffic scene analysis. Using this system, offending drivers is distinguished automatically. Three main parts of the proposed system are vehicle's tracking, features extraction from obtained trajectory and constructing stochastic model. Eight HMM/GMM are constructed for eight area of different traffic scene. More than 90 minutes training film is used for generating of statistical model. The parameters of HMM are obtained using Baum-Welch method for each area then obtained log likelihood of HMM output is used for generating of template by GMM. Obtained results show that creating template of HMM by GMM can determine offending drivers efficiently.

Hidden Markov Model, Gaussian Mixture Model, and traffic scene