This project addresses the critical issue of traffic fatalities, which have reached a 16-year high, by focusing on the identification of complex roadways and their impact on fatal crashes. Utilizing the MIT-AVT dataset, which includes extensive driving data from various vehicles equipped with advanced technologies, the research employs a computer vision model with panoptic segmentation to detect multiple objects in road scenes, thereby determining roadway complexity. The project aims to analyze this complexity in relation to driver behavior and the occurrence of fatal crashes. The research is divided into three phases: feature extraction, data integration, and model building. Feature extraction involves using the OneFormer panoptic segmentation model to label specific regions of images, while data integration involves annotating these images to create a road complexity dataset. The final phase includes building classification models to identify fatal crash hot spots and conducting correlation analysis to explore the relationship between road complexity, driver behavior, and fatal crashes. This research is significant for its potential to enhance automated vehicle safety and inform traffic safety policy.
Holistically Identifying Road Complexity and Relating it to Fatal Crashes
This project seeks to holistically identify complex roadways using computer vision and machine learning models, and analyze their impact on driver behavior and fatal crash occurrences.
Status
In Progress
Activity
Research
Principal Investigator
Project Dates
1/1/2024 - 12/31/2024
Approximate Project Cost
$120,000