Real-Time Traffic Status Classification Based on Gaussian Mixture Model
Xiong Liu,Li Pan,Xiaoliang Sun
DOI: https://doi.org/10.1109/dsc.2016.39
2016-01-01
Abstract:Traffic status evaluation is the most important part of Intelligent Traffic System (ITS). However, nowadays, most models used to identify traffic performance are either based on traffic flow theories or machine learning algorithms and data mining algorithms. On that base, a new traffic performance evaluation model which combines traffic theory and machine learning algorithm is presented. In order to establish such model, characteristics of traffic data is discussed at first. The result of statistical analysis of traffic data shows that the probability density of a large amount of traffic data could be synthetized by a few Gaussian distributions. Therefore, a suitable traffic status evaluation model which combines Gaussian mixture model (GMM) and linear regression (LR) is proposed in this paper. In this model, firstly, by using Gaussian mixture model, historical traffic data will be clustered into three groups with respect to three traffic states, which are free flow, quasi-free flow and crowd flow. Then, the data with respect to crowd flow will be used to train linear regression model, and linear regression model will generate a straight line which analogues the line J in Kerner's traffic theory to classify crowd flow into two parts, which are related to synchronized flow and traffic jam. Through the two steps, traffic performance is described into four states, which are free flow, quasi-free flow, synchronized flow and traffic jam. The result of designed experiment shows that the model could evaluate traffic performance effectively.