Optimizing Real-time Traffic Sign Recognition Algorithms For Mobile Devices.
Author(s)
Vial, Valentin
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
The effective management of road assets is essential for improving traffic safety and reducing crash rates. While recent advancements in artificial intelligence (AI) have been applied to traffic sign inventory systems, these systems often handle only a limited range of sign types. This is inadequate given that the Manual on Uniform Traffic Control Devices (MUTCD) outlines standards for more than 500 traffic signs and variants. This research aims to enhance traffic sign inventory systems by focusing on the Detection and Classification modules to accurately recognize a broader range of traffic signs. A novel pipeline leveraging CLIP architecture is proposed, featuring prompt-based classification. An optimized YOLO (You Only Look Once) detection model with an additional head for faster execution on a smartphone is also explored. The system was evaluated using real-world recordings, achieving a high recall rate (> 90%) and adaptability through zero-shot classification. These results demonstrate the feasibility of a scalable and efficient traffic sign inventory system, offering significant improvements in performance and processing time for large-scale applications.
Sponsor
Date
2024-12-06
Extent
Resource Type
Text
Resource Subtype
Thesis