基于级联端对端深度架构的交通标志识别方法

A method of traffic sign recognition based on the cascade and end-to-end depth architecture

摘要:

交通标志的正确识别是智能车辆规范行驶、道路交通安全的前提。为解决智能车采集目标图像模糊、分辨率低，造成识别精度低且时效性差的问题，构建一种基于级联深度网络的交通标志识别模型，该模型级联超分辨率处理网络ESPCN与目标检测识别网络RFCN，ESPCN网络提高输入采集图像的分辨率，为低分辨率图像实现超分辨率处理，RFCN网络提取图像全局特征，实现交通标志的检测与分类识别。平衡采样及多尺度的训练策略结合数据增强的预处理方法，增强了网络模型的鲁棒性及扩展性。经实验验证，算法模型针对常见交通标志识别率达到98.16%，召回率达到96.2%，且鲁棒性较好。

Abstract:

The correct identification of traffic signs is a prerequisite for smart vehicles to regulate driving and road traffic safety. In order to solve the problem that the target image of the smart car is blurred and the resolution is low, resulting in low recognition accuracy and poor timeliness, a traffic sign recognition model based on cascading depth network is constructed. The model cascades the super-resolution processing network ESPCN and target detection. Identifying the network RFCN, the ESPCN network improves the resolution of the input captured image, achieves super-resolution processing for low-resolution images, and extracts global features of the image from the RFCN network to realize the detection and classification of traffic signs. Balanced sampling and multi-scale training strategies combined with data-enhanced pre-processing methods enhance the robustness and scalability of the network model. The experimental results show that the recognition rate of common traffic signs is 98.16%, the recall rate is 96.2%, and the robustness is good.