Related Experiment Videos
Handling uncertainty in optimization enabled the deep Q network for traffic sign and criticality detection
Soja Salim1,2, Jayasudha J S3, Soniya B1
1Computer Science and Engineering, Sree Chitra Thirunal College of Engineering, Thiruvananthapuram, India.
Traffic Injury Prevention
|August 7, 2026
Summary
This study introduces an advanced Traffic Sign Recognition (TSR) model with uncertainty quantification, achieving high accuracy for critical sign detection and ensuring road safety.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Traffic Sign Recognition (TSR) is crucial for autonomous driving systems.
- Existing TSR models often lack robust uncertainty quantification, impacting reliability.
Purpose of the Study:
- To develop an efficient TSR model incorporating uncertainty quantification.
- To improve the accuracy and reliability of critical traffic sign detection.
Main Methods:
- Utilized Gaussian filtering for pre-processing and Segmentation U-Net (SegU-Net) for sign localization.
- Employed a Gradient Descent-Teamwork Optimization Algorithm-based Deep Q Network (GD-TOA based DQN) for sign classification.
- Implemented Mahalanobis distance for confidence level computation and uncertainty assessment.
Main Results:
- Achieved high performance metrics: 97.9% accuracy, 98.2% precision, 98.3% recall, and 98.2% F-measure.
- Demonstrated effective uncertainty quantification for critical sign identification.
- The model successfully flagged signs with high uncertainty for human review.
Conclusions:
- The proposed TSR model with uncertainty quantification is highly effective for criticality detection.
- This technique shows significant promise for enhancing the safety and reliability of intelligent transportation systems.
