Modelling pedestrian-vehicle interaction behaviours at non-signalised crosswalks using a game theory approach incorporating the risk perception

Uncontrolled crosswalks, lacking signal guidance and regulations, are common locations for pedestrian-vehicle interactions (PVIs). The dynamic interaction could result in fluctuating risk perception levels for pedestrians and vehicles, which poses higher risks to pedestrians. Therefore, it is essential to examine the association between pedestrian and vehicle behaviours at street crossings, especially in unsignalized ones, to enhance long-term road safety. This research centres on the risk perception of pedestrians and vehicles at unsignalized crosswalks, analysing their interaction behaviours while considering risk perception. Specifically, to recognise the acceptable levels of risk perception for pedestrians and vehicles, we established quantitative models for the risk perception utilising cognitive psychology theory and the Long Short-Term Memory (LSTM) network. Then, using game theory, the correlation between quantitative risk perception values, acceptable risk perception levels, and the movement states of pedestrians and vehicles are explored, followed by the construction of interaction behaviour models for pedestrians and vehicles. Finally, data collected from field observation on the movements of pedestrians and vehicles is applied to examine the performance of the proposed the PVIs behaviour models, and the results indicated an accuracy rate of 93.33% for the developed behaviour models, demonstrating excellent predictive efficiency. Also, the model’s adaptability was tested on other unsignalized crosswalks.