Date of Award

6-3-2026

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2026

Department

Computer Science and Engineering

First Advisor

David Anastasiu

Abstract

Identifying traffic trends is a critical step in urban planning and design for the purpose of creating safer and more efficient roads. One such problem in the field of traffic data is corridor counting, which is the process of counting the unique number of vehicles that pass through a set of predefined intersections in a given time. In this paper, we develop an automated solution to perform corridor counting in real-time using camera footage and computer vision. We build upon the architecture of a previous Senior Design project, replacing and refining key components and architectural choices to improve accuracy and enable real-time processing. Our solution employs YOLOv11m for vehicle detection, a model ensemble consisting of SeResNet101 and TransReID for vehicle re-identification, and RabbitMQ for internode communication across a distributed architecture. We evaluate our pipeline using the CityFlowV2 dataset, which is composed of forty-six synchronized, unique camera feeds. Our system achieved the dataset maximum of ten frames per second, satisfying our requirement for real-time processing. Most notably, this speedup was achieved while maintaining a 23.9% reduction in weighted root mean squared error compared to the baseline. These results demonstrate the feasibility of a real-time, multi-camera corridor counting system that does not sacrifice counting accuracy.

Download

Share

COinS