To address the current limitations of illumination effects and low accuracy of model,a faster region-based convolutional neural network(Faster R-CNN)traffic sign detection algorithm was proposed.It addressed uneven illumination between sky and non-sky areas using Gamma transformation to improve the feature representation of traffic signs.The use of the convolutional block attention module-based an efficient network(CBAM-EfficientNet)counteracted network depth degradation,improved shallow network feature extraction and reduced parameters.The introduction of a feature pyramid network facilitated the detection of different-sized objects,enhanced the network's ability to perceive traffic signs of different sizes.Experimental results demonstrated high accuracy,with an PmA of 99.79%on the GTSDB dataset and 87.62%on the CCTSDB2021 dataset,provided a highly accurate solution for the detection of unevenly illuminated traffic signs.
维普
万方数据