Research on rainy day traffic sign recognition algorithm based on PMRNet

• Published in: Mathematical biosciences and engineering : MBE Vol. 20; no. 7; pp. 12240 - 12262
• Main Authors: Zhang, Jing, Zhang, Haoliang, Lang, Ding, Xu, Yuguang, Li, Hong-An, Li, Xuewen
• Format: Journal Article
• Language: English
• Published: United States AIMS Press 01.01.2023
• Subjects: cot module; image deraining; multi-scale residual; traffic sign recognition; multi-scale residual; traffic sign recognition; image deraining; CoT module
• ISSN: 1551-0018; 1551-0018
• DOI: 10.3934/mbe.2023545

The recognition of traffic signs is of great significance to intelligent driving and traffic systems. Most current traffic sign recognition algorithms do not consider the impact of rainy weather. The rain marks will obscure the recognition target in the image, which will lead to the performance degradation of the algorithm, a problem that has yet to be solved. In order to improve the accuracy of traffic sign recognition in rainy weather, we propose a rainy traffic sign recognition algorithm. The algorithm in this paper includes two modules. First, we propose an image deraining algorithm based on the Progressive multi-scale residual network (PMRNet), which uses a multi-scale residual structure to extract features of different scales, so as to improve the utilization rate of the algorithm for information, combined with the Convolutional long-short term memory (ConvLSTM) network to enhance the algorithm's ability to extract rain mark features. Second, we use the CoT-YOLOv5 algorithm to recognize traffic signs on the recovered images. In this paper, in order to improve the performance of YOLOv5 (You-Only-Look-Once, YOLO), the 3 × 3 convolution in the feature extraction module is replaced by the Contextual Transformer (CoT) module to make up for the lack of global modeling capability of Convolutional Neural Network (CNN), thus improving the recognition accuracy. The experimental results show that the deraining algorithm based on PMRNet can effectively remove rain marks, and the evaluation indicators Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) are better than the other representative algorithms. The mean Average Precision (mAP) of the CoT-YOLOv5 algorithm on the TT100k datasets reaches 92.1%, which is 5% higher than the original YOLOv5.