Automatic Safety Diagnosis in Connected Vehicle Environment

Monday, August 22, 2022 - 20:15

Previous researchers found that the most important accident causation factor was the driver’s abnormal driving status, which was associated with driving volatility. And the driving volatility can be traced from the trajectories of the vehicles that were embedded in the BSMs. Based on these findings, we developed an automatic safety diagnosis system for the connected vehicle environment (ASDSCE), a real-time near crash warning tool with a multi-dimensional cloud-based driving anomaly detection (DAD) model and a conflict identification model (CIM) on the individual level specifically configured for BSMs. The architecture of the proposed system is composed of two components: one is in the cloud who collects and stores BSMs of the CVs and determines in batch mode the thresholds of each vehicle; the other is in the in-vehicle subsystem which determines the driving anomalies and detect conflicts. A near crash will be warranted when the traffic situation satisfies both of the following two conditions: (a) a conflict is identified and, (b) at least one of the drivers that is involved in the conflict is in abnormal driving status. The ASDSCE contains the following features: focusing on detecting abnormal drivers instead of normal drivers; using the trajectory data embedded in the BSM to study driving volatility; implementing on the individual drivers instead of the aggregate level; and reducing the model training time in order to leave sufficient time to the involved drivers to perform successful evasive actions. The presented computational pipeline of ASDSCE includes raw data collection, data preprocessing, data analysis, data communication and warning message generation. ASDSCE is built with Python on Visual Studio 2019 using the BSMs from the CV pilot studies and evaluated using the SHRP2 naturalistic driving study crash data.  The report is available online from the University of Florida at


Thank you to our Premier Sponsors