Related Experiment Video
Updated: Jun 27, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
Semi-Supervised Traffic Sign Detection with Dynamic Pseudo-Label Selection and Gated Feature Fusion-Based Proposal
Chenhui Xia1, Yeqin Shao1, Meiqin Che1
1School of Transportation and Civil Engineering, Nantong University, Nantong 226019, China.
Sensors (Basel, Switzerland)
|June 26, 2026
Summary
This study introduces a novel semi-supervised method for traffic sign detection, significantly improving accuracy for rare signs and small objects. The approach uses dynamic pseudo-label selection and gated feature fusion, outperforming existing methods with minimal labeled data.
Area of Science:
- Computer Vision
- Machine Learning
- Autonomous Driving Systems
Background:
- Accurate traffic sign detection is crucial for autonomous driving safety.
- Fully supervised methods are limited by high annotation costs.
- Semi-supervised methods offer a cost-effective alternative but struggle with imbalanced data and small object detection.
Purpose of the Study:
- To develop a semi-supervised traffic sign detection method that addresses challenges of class imbalance and small sign detection.
- To reduce the reliance on extensive manual annotation in training autonomous driving systems.
Main Methods:
- Proposed a Class Distribution-based Dynamic Pseudo-Label Selection (CD-DPLS) module to improve tail class performance.
- Implemented a Gated Feature Fusion-based Proposal Refinement (GFF-PR) strategy for enhanced small traffic sign detection.
- Utilized an Adaptive-Weight Focal Loss (AWFL) to dynamically adjust pseudo-label weights based on classification confidence.
Main Results:
- The proposed method achieved superior performance compared to state-of-the-art semi-supervised approaches.
- Demonstrated significant improvements in mean Average Precision (mAP50) scores: 10.8% with 1% labeled data and 34.9% with 10% labeled data.
- Successfully improved detection rates for tail classes and small traffic signs.
Conclusions:
- The novel semi-supervised method effectively tackles class imbalance and small object detection in traffic sign recognition.
- The approach offers a practical solution for developing safer autonomous driving systems with reduced annotation effort.
- The CD-DPLS and GFF-PR modules represent significant advancements in semi-supervised learning for object detection.