As human factors contribute to more than 90% crashes, abnormal driving behavior has been intensively studied to improve traffic safety. Basic Safety Messages (BSM) transmitted between connected vehicles (CVs) are time series data with high-cardinality. Real-time anomaly detection on driving status using BSMs is challenging but promising. With the advent of large-scale connected vehicles, due to the tremendous volume and complexity, it is not realistic to store all the BSMs into the data center. This paper is motivated to explore what information imbedded in BSMs needs to be stored, how to extract it and how to process it for real-time safety diagnosis. We propose a real-time multi-dimensional anomaly detection system for abnormal driving status detection on individual level specifically configured for BSM data. The architecture of the proposed system is composed of two parts: in traffic management center (TMC) and the in-vehicle devices. In the TMC, the system collects and stores BSMs of the vehicles it covers for a short period of time, and determines in batch mode the thresholds of the selected key performance indicators (KPIs) representing normal status for each vehicle using unsupervised machine learning algorithms, and broadcast the thresholds through the BSMs; in the in-vehicle device, as new BSMs streaming in, the devise compares the new values with the received thresholds and determine if it is an outlier of each KPI; if detected, the outliers will be sent back to the TMC where the system will run topology learning to determine if the outliers combined deserve an anomaly flag against that vehicle according to the deviations and durations; finally, the system will determine the impact factors to update thresholds according to the significances of the outliers. This system of anomaly detection is a crucial component of our pipeline for the automatic safety digression system in the CV environment. This research is sponsored by the Southeastern Transportation Research, Innovation, Development and Education Center.     

Anomaly Detection on Driving Status Using Basic Safety Messages in Connected Vehicle Environment

ASCE International Conference on Transportation & Development (ICTD 2021) is organized by the Transportation & Development Institute (T&DI) of ASCE to facilitate the exchange of information, knowledge, and best practices among transportation and development practitioners and researchers, public infrastructure owners, policymakers, government engineers and planners, operations managers, and leading academics from around the world.

Paper Proceedings: https://doi.org/10.1061/9780784483534 | https://doi.org/10.1061/9780784483541

Conference DOI: https://doi.org/10.26226/m.631784867747920012e73503