Prediction of pedestrian-vehicle conflicts at signalized intersections based on long short-term memory neural network

Pedestrian protection is an important component of road safety. Intersections are dangerous locations for pedestrians with mixed traffic. This paper aims to predict potential traffic conflicts between pedestrians and vehicles at signalized intersections. Using detection and tracking techniques in computer vision, pedestrians' and vehicles' features are extracted from video data. An LSTM (Long Short-term Memory) neural network is proposed to predict the pedestrian-vehicle conflicts 2 s ahead. The established model reaches an accuracy of 88.5 % at one signalized intersection. It is further tested at a new intersection, reaching the accuracy of 84.9 %, while the new data merely takes up 30 % of the training data set. This indicates that the proposed model is promising to be implemented at different locations. Moreover, the proposed model can also be applied to develop collision warning systems under the Connected Vehicles' environment.