Transport psychology based cognitive architecture for traffic behavior prediction

Prediction of extemporaneous events in traffic surveillance is crucial in the prevention or alleviation of the gravity of accidents. Modeling of normal/abnormal behavior and mental state inference of drivers help in the prediction of such events. Traffic psychology lends itself to the development of such models. Analysis of driver state, emotion and behavior are important components of traffic psychology. However, most models based on traffic psychology are rather abstract and lack neurobiological grounding. They are also disparate from computational models of traffic monitoring. In this paper, we extend and develop neurobiologically grounded computational models for driver state and behavior inference by mimicking the mirror neuronal architecture. The developed system uses a combination of modular cognitive neurobiological architecture combined with traditional computer vision techniques for traffic monitoring resulting in prediction and detection of extemporaneous events. Psychophysical as well as neurobiological criteria are used for evaluation on both simulated and real data. The model is shown to be robust to perturbations, with rapid convergence (less than 0.2 normalized time units) in most cases.