Transformer Fusion and Residual Learning Group Classifier Loss for Long-Tailed Traffic Sign Detection
- 5 February 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (7) , 10551-10560
Abstract
Automatic traffic sign detection and recognition (ATDR) plays a crucial role as a sub-module of the intelligent transportation system (ITS) and has gained significant attention in recent years. However, ATDR faces challenges due to two main factors: the small size of traffic signs in images and the long-tailed distribution of sign categories in the real world. (i.e., only a few categories have sufficient samples, and most categories have insufficient samples.). These challenges result in a decline in the performance of existing detection frameworks. In this paper, we propose a novel framework for traffic sign detection, specifically designed to tackle the challenges posed by these issues. Our approach includes an efficient adaptive and transformed spatial feature fusion module for detecting small traffic signs. This module adaptively learns feature map associations to extract and fuse meaningful features through cross-space and cross-scale interactions at different levels. Additionally, to tackle the long-tailed traffic sign problem, we have designed a residual learning group classifier loss, by dividing the traffic signs into two levels, reserving a specific capacity for the long-tailed category from the perspective of the parameter space, the residual fusion mechanism is introduced to enhance and optimize the tail category image. Our experimental results, obtained from rigorous evaluations on the TT100K and GTSDB datasets, demonstrate the effectiveness of our approach. The improvements achieved in small traffic sign detection and long-tailed detection underscore the significance of our framework in advancing ATDR research.Keywords
Funding Information
- National Key Research and Development Program of China (2018YFB1600500)
- National Natural Science Foundation of China (51908388, 52002262, 52075353)
This publication has 39 references indexed in Scilit:
- Improved YOLOv5 network for real-time multi-scale traffic sign detectionNeural Computing and Applications, 2022
- Group multi-scale attention pyramid network for traffic sign detectionNeurocomputing, 2021
- TSingNet: Scale-aware and context-rich feature learning for traffic sign detection and recognition in the wildNeurocomputing, 2021
- The Mapillary Traffic Sign Dataset for Detection and Classification on a Global Scale2020
- Feature Pyramid Transformer2020
- A systematic study of the class imbalance problem in convolutional neural networksNeural Networks, 2018
- Traffic-Sign Detection and Classification in the Wild2016
- Traffic sign segmentation and classification using statistical learning methodsNeurocomputing, 2015
- Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning2005
- Traffic sign recognition and analysis for intelligent vehiclesImage and Vision Computing, 2003