Description

Near-Miss Collisions are events that, with a slight change in position or timing, could have resulted in a collision, which could have caused severe injury or property damage. Understanding near-miss collisions can help identify risks and potentially improve road safety. In this project, we developed an effective end-to-end system based on advanced artificial intelligence (AI) models and computer vision algorithms to detect and report near-miss collisions as an important indicator to identify and measure safety risks, especially in specific circumstances such as a right turn on a red light. The main objective is to improve the safety of pedestrians and bicyclists, by applying automated AI-powered systems to detect accident risks for pedestrians and cyclists. The developed system includes algorithms for detecting and tracking all traffic objects including pedestrians and bicyclists, as well as algorithms for estimating collision risks and detecting near misses. We evaluated the developed system on real videos captured by actual traffic cameras in the city of Los Angeles. Despite the low quality of some of the videos, our results demonstrate high accuracy of the developed models in identifying traffic collision risks and detecting near-misses. The information generated by the developed system allows us to enhance safety measures for pedestrians and bicyclists while simultaneously optimizing traffic flow.

Publication Date

10-2024

Publication Type

Report

Topic

Active Transportation, Miscellaneous

Digital Object Identifier

10.31979/mti.2024.2350

MTI Project

2350

Keywords

Artificial intelligence, Machine learning, Pedestrian counts, Pedestrian safety, Traffic safety

Disciplines

Science and Technology Studies | Transportation

Additional Files
2350-RB-Pourhomayoun-AI-Machine-Learning-Pedestrian-Safety.pdf (738 kB)
Research Brief
Download

Additional files available below

Share

COinS