基于深度学习的交通标志检测方法研究

提出了一种基于Faster R-CNN深度学习框架的交通标志检测方法,使用VGG16卷积神经网络自动提取交通标志图像特征,并将卷积特征图传入区域建议网络(RPN)中进行前景目标筛选及回归目标边框,将建议区域框映射到特征图上,经过RoI池化层后输出固定大小的建议框,利用分类网络对建议区域进行具体的类别判断,并精确回归目标的边框。并将算法在德国交通标志数据集GTSDB进行了实验验证,实验结果表明了所提出算法的有效性,该方法对不同光照、遮挡、标志模糊等具有鲁棒性。

Research of Traffic Sign Detection Method based on the Deep Learning Model

A traffic sign detection method based on the Faster R-CNN deep learning framework was proposed. In this paper, VGG16 convolutional neural network was used to extract traffic sign image features automatically. Then the convolutional feature map was sent to the region proposal network (RPN) for foreground object screening and the bounding box regression. The region proposal was mapped to the fixed size proposal output after RoI pooling on the feature map. Then the classification network was used to perform specific classification and the bounding box regression was further computed. The experiments were performed on the German Traffic Sign Detection Benchmark (GTSDB). Experimental results show that the method is effective and robust to different lighting, occlusion, motion blur and so on.