Georgia Institute of Technology
Comparative Analysis of Traffic Sign Tracking Methods: From Classical Trackers to Temporal YOLO Extensions and Trajectory Optimization
Abstract
dc:description.abstractEfficient monitoring of roadway infrastructure is essential for safety, maintenance planning, and regulatory compliance. Traditional traffic sign inventory methods rely on manual field surveys or static imagery, which are costly and difficult to scale. Mobile video acquisition enables continuous capture of roadway environments but introduces the challenge of maintaining consistent identities for traffic signs across motion blur, occlusions, and large viewpoint changes. This thesis investigates traffic sign tracking in dash-cam video as part of a larger detection–tracking–classification–localization pipeline deployed for the Pima County roadway asset inventory project. The study benchmarks classical motion-based trackers, modern association-based algorithms, and a proposed YOLOv11-based tracker augmented with an appearance-embedding head. Two post-tracking refinement modules—CoTracker for reconnecting fragmented trajectories and a position-based graph smoothing method for identity cleanup—are also evaluated. Experiments on a curated, frame-accurate annotated dataset show that the proposed method improves identity stability and reduces fragmentation relative to widely used baselines. The results demonstrate that lightweight appearance-aware tracking, combined with targeted post-processing, provides an effective balance between accuracy and computational efficiency for large-scale roadway asset inventory applications.
Degree
thesis:*- Name thesis:degree_name
- Computer Science, MS
- Department thesis:degree_department
- Computer Science
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Moutahir, Jed Seyed, Amine
- Advisor dc:contributor.advisor
-
- Pradalier, Cédric
- Committee members dc:contributor.committeemember
-
- Tsai, Yi-Chang
- Yang, Zhongyu
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/80240