基于多源特征增强的交通标志识别方法

Traffic sign recognition method based on multi-source feature enhancement

针对现实场景中交通标志识别在照明不佳环境下和应对扭曲、旋转、平移等空间变化时准确率较低的问题,在优化亮度、对比度特征的基础上结合图像的空间位置信息,提出一个新颖的基于多源特征增强的交通标志识别方法.首先设计亮度与对比度增强模块以凸显低照度图像的特征信息,降低低照度图像识别难度.然后结合空间转换单元构建轻量化特征处理网络,通过弱化图像携带的无关信息聚焦数据中的感兴趣区域,有效分离背景噪声的同时也改善了输入数据的空间不变性.最后主分类网络对图像特征图进行细粒度的识别并输出预测的类别标签.实验结果表明:本文提出的模型在公开基准数据集GTSRB上的准确率达到99.52%,能有效解决交通标志现实场景下识别率较低的问题.

To address the challenge of diminished accuracy in traffic sign recognition under real-world condi⁃tions of poor lighting environment and dealing with spatial changes such as distortion,rotation,and transla⁃tion,this paper presents an innovative traffic sign recognition approach based on multi-source feature enhance⁃ment that integrates spatial positional data with optimized brightness and contrast features.Firstly,aiming to mitigate recognition difficulties in low-light images,a brightness and contrast enhancement module is de⁃signed to highlight the feature information of low-light images.Then,a deep image classification network is constructed by combining the spatial transformation unit.Furthermore,a lightweight feature processing net⁃work is constructed combining the spatial transformation units.By weakening irrelevant information present in the image and focusing on the region of interest within the data,the background noise is effectively segre⁃gated,thereby enhancing the spatial consistency of the input data.Finally,the main classification network performs fine-grained identification of image feature maps and outputs the predicted class labels.Experimental results show that the proposed model achieves an accuracy rate of 99.52%on the public benchmark dataset GTSRB,effectively addressing the problem of low recognition rate for traffic signs in real-world scenarios.