Abnormal Driving Behavior Detection System

The detection of abnormal driving behavior is important for safety. However, driving is a combination of both internal (e.g., skills) and external (e.g., road type, traffic conditions) factors that make abnormal driving behavior detection largely subjective. On the other hand, this subjectivity could be handled through the analysis of driving data at multiple levels including the low-level driving action recognition up to high-level inference of a road section. In this paper, we tackle this problem and propose a Hierarchical Abnormal Driving Behavior Detection System (H-ABDS) that is not only mining individual behavior of vehicles but also runs a statistical learning -based technique to understand human driving data at multiple levels. We design a hierarchical architecture to enable abnormal driving behavior detection at the city -scale and demonstrate its benefits through extensive simulations conducted on simulated traffic data. Our preliminary result has shown that H-ABDS can identify all driving anomalies by about 75% accuracy under a certain degree of connected vehicle penetration rates.