Abnormal Driving Pattern Detection from GPS Trajectories Using Vision Transformer

Given GPS points on a transportation network, the Driving Pattern Detection (DPD) problem aims to classify drivers as normal or abnormal based on their driving behavior. The DPD problem is challenging due to the variability in trip lengths, routes, and spatial patterns, which complicates input standardization for deep learning models. In this paper, we introduce a novel spatial representation learning framework for the DPD problem by analyzing driving patterns using a real-world dataset. We propose using binary grid images to capture the spatial structure of driving trajectories and present a new driving behavior representation for input to a Vision Transformer (ViT) model for driver classification. The experimental results demonstrate the effectiveness of the proposed algorithm, achieving an F1 score of 94% that significantly outperforms the baseline models. The results indicate that binary grid representations can effectively encode interpretable spatial patterns in driving behavior, with direct relevance to improved driver classification, road safety, and cognitive health assessment.