Traffic signs contain important traffic information. Timely and accurate perception and understanding of traffic sign could be significant for driving safety for intelligent autonomous vehicles. In this study, we prioritize the complex problem of cross country road sign understanding with limited data from only one country. Our proposal consists of unsupervised domain adaptation and attention module for Cross-country Traffic Sign Recognition. Most prevalent methods require large quantity of traffic sign data from different countries and states for cross-country traffic sign understanding which could be nearly infeasible due to difficulties in data acquisition, collection and data privacy policies. By applying Squeeze-and-Excitation attention module enhanced unsupervised domain adaptation (UDA), our work demonstrated quality performance in cross-country traffic sign understanding with only source domain data (Traffic sign samples in China). Experimental results suggest that our model can make accurate inferences on the understanding of cross-country traffic signs. We evaluated its performance using data from different countries, including China, Germany, Iran, etc.
Cross-Country Traffic Sign Understanding via Attention Enhanced Unsupervised Domain Adaptation
28.10.2023
900109 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Unsupervised Evaluation of Lidar Domain Adaptation
IEEE | 2020
|Multi-scale traffic sign detection model with attention
SAGE Publications | 2021
|