Short-term traffic safety forecasting using Gaussian mixture model and Kalman filter

In this paper, a prediction model is developed that combines a Gaussian mixture model (GMM) and a Kalman filter for online forecasting of traffic safety on expressways. Raw time-to-collision (TTC) samples are divided into two categories: those representing vehicles in risky situations and those in safe situations. Then, the GMM is used to model the bimodal distribution of the TTC samples, and the maximum likelihood (ML) estimation parameters of the TTC distribution are obtained using the expectation-maximization (EM) algorithm. We propose a new traffic safety indicator, named the proportion of exposure to traffic conflicts (PETTC), for assessing the risk and predicting the safety of expressway traffic. A Kalman filter is applied to forecast the short-term safety indicator, PETTC, and solves the online safety prediction problem. A dataset collected from four different expressway locations is used for performance estimation. The test results demonstrate the precision and robustness of the prediction model under different traffic conditions and using different datasets. These results could help decision-makers to improve their online traffic safety forecasting and enable the optimal operation of expressway traffic management systems.