Pedestrian Crossing Intention Prediction Method Based on Multi-Feature Fusion

Pedestrians are important traffic participants and prediction of pedestrian crossing intention can help reduce pedestrian–vehicle collisions. For the problem predicting an individual pedestrian’s action where there is potential, a method that considers multi-feature fusion proposed in this study, which integrates information affecting pedestrians’ actions, such as environment. This study based on BPI dataset for training validation, test results show model has good data fitting generalization ability; set accuracy 89.5% model, with AUC 0.992. In specific scenario, predict when longitudinal relative distance between vehicle about 20 m 0.6 s before crossing, provide useful decision making intelligent vehicles.