Armstrong Aboah
North Dakota State University
Biography
I am an Assistant Professor at the North Dakota State University. An ingenious and resourceful Transportation Data Scientist with a proven track record of success in research and hands-on experience developing cutting-edge database solutions, statistical modeling, data products, and computer vision systems aimed at improving transportation system management and operations. Has worked as an architect and application developer on a variety of projects that required the use of data mining and machine learning models to solve large-scale, complex, and difficult transportation problems. I am broadly interested in computer vision and machine learning. My research involves visual reasoning, vision and language, image generation, air taxis, naturalistic studies, and autonomous vehicles.
SRP Project Title
PRIME: A Foundational Predictive Real-time Intersection Monitoring Engine
NAIRR Project
PRIME: A Foundational Predictive Real-time Intersection Monitoring Engine
Topical Areas
Applied Computer Science; Artificial Intelligence and Intelligent Systems; Civil Engineering; Computer Science
Abstract
Traffic intersections remain critical points of vulnerability in transportation infrastructure, accounting for 20% of vehicular accidents and over 7,000 annual fatalities in the United States. Current intersection monitoring systems, relying on basic motion detection or single-class object detection, struggle with complex scenarios involving multiple road users and varying environmental conditions. This research proposes PRIME (Predictive Real-time Intersection Monitoring Engine), a foundational traffic intersection monitoring algorithm that integrates advanced deep learning techniques for multi-class video object detection and trajectory prediction. This project directly aligns with NAIRR priority areas by creating open-source foundation models for specific applications and utilizing experimental data from sensors and detectors. The proposed system employs a novel object detection architecture with enhanced feature pyramid networks and combines transformer encoders with graph neural networks to achieve robust object detection and accurate 5-second trajectory predictions. Our technical objectives include developing a multi-class detection system with over 95% accuracy, implementing proactive trajectory prediction, and generating anonymous traffic patterns for urban planning. The project will be executed utilizing TACC Frontera GPU resources for model development and training. We will also release our codebase and make our annotated dataset publicly available with the aim of establishing a common foundation for intersection monitoring systems.
Desired Skills
A good programming background, like Python. Familiarity with computer vision.
Lightning Talk Title
Towards a safe and smart intersection
Keywords
Machine learning, Safety, Intersection, trajectory