Contextual and Behavior Factors Extraction from Pedestrian Encounter Scenes Using Deep Language Models.

This study introduces an NLP framework including deep language models to automate the contextual and behavior factors extraction from a narrative text that describes the environment and pedestrian behaviors at the pedestrian encounter scenes. The performance is compared against a baseline BiLSTM-CRF model trained for each factor separately. The evaluation results show that the proposed NLP framework outperforms the baseline model. We show that the proposed framework can successfully extract nested, overlapping, and flat factors from sentences through the case studies. This model can also be applied to other descriptions when physical context and human behaviors need to be extracted from the narrative content to understand the behavioral interaction between subjects further.