|
Abstract:
|
Detailed eye glance pattern embarks high-resolution information in people’s driving behavior while doing lane change process, benefiting varieties of application areas where a thorough understanding of safety-critical events is essential, such as safety research in Automatic Vehicles, Usage-Based Insurance. Current state of art approaches for evaluating the risk eye glance pattern mostly focus on an aggregated level, which failed to model the underlying dependence structure and can only investigate a very limited number of behaviors. To address these issues, we propose a new embedding-based kernel two-sample test to enrich more statistical power. Also, understanding the accurate time point where eye patterns play important role in safety criticality, we innovatively introduce a multiple time block approach that dynamically slices the lane change into several blocks. We implement this method in SHRP2 NDS data and the application results show convincing performance.
|
