A Multicomponent Cross-Integration Network for Traffic Sign Detection in Complex Conditions

The accurate detection and recognition of traffic signs are crucial for autonomous driving, advanced driver assistance systems, and road safety. In real-world road scenarios with complex backgrounds, traffic sign detection often faces challenges, leading to false and missed detections. This article introduces an innovative network MCCI-Net to address the challenges associated with recognizing and detecting traffic signs, considering their vulnerability to perspective distortion and illumination interference. First, a CNN-based backbone network with feature grouping was designed to improve the accuracy and efficiency of feature extraction. Second, a transformer encoder with an interactive fusion architecture was designed to enhance key features, refine and balance the correlation between semantic and location information, and gain the ability to mitigate feature conflicts. Finally, a transformer-decoder head is designed to optimize object queries for category prediction and box regression, enhancing detection accuracy and iteration speed reliability. Experimental results fully validate the effectiveness and robustness of the proposed MCCI-Net in traffic sign detection and recognition. The approach demonstrates strong capabilities in overcoming the challenges posed by perspective distortion and illumination interference. It exhibits superior performance compared with the existing state-of-the-art traffic sign detection methods. Our approach achieved a 97.2% in the mAP@0.5 evaluation on the public dataset TT100K, 97.5% on the CCTSDB dataset, and 94.1% on the GTSDB dataset.