【摘要】 汽车制造领域L4以及更高级别的自动驾驶技术已经成为当今汽车行业最前沿的发展方向。智能驾驶汽车在行进过程中对公路上的交通标志的识别可以有效帮助驾驶员及时做出决策,降低交通违法和交通事故的发生率。目前智能交通标志识别系统仅可在少量指定车型中使用,普及率较低。为使更多的驾驶员可以使用智能交通标志识别系统完成辅助驾驶,本文对中国标准交通标志数据库(CCTSDB)进行研究,提出了一种基于残差神经网络的智能交通标志识别算法。利用高斯平滑和Canny锐化对实验图像预处理,在Mxnet框架下引入残差神经网络模型Res Net-18将图像分类识别。结果表明该算法能够有效识别交通标志信息,在实验环境下对于交通标志识别率可达91.87%,具有识别速度快,可移植性好的显著特点,为智能交通标志识别系统的轻量化和大众化提出了新的可能性。更多还原

【Abstract】 Realizing L4 and higher level auto-driving technology in automobile manufacturing has become the most cutting-edge development direction of the entire automobile industry.The identification of traffic signs on the road by smart driving vehicles can effectively help drivers make timely decisions and reduce the incidence of traffic violations and accidents.At present,the Intelligent Traffic Sign Recognition System can only be used in a small number of designated vehicles with a low popularity.In order to enable more drivers to use the Intelligent Traffic Sign Recognition System to complete assisted driving,this paper studies the China Standard Traffic Sign Database(CCTSDB),and presents an intelligent traffic sign recognition algorithm based on residual neural network.First,we use Gaussian smoothing and Canny sharpening to pre-process the experimental image,and then introduce the residual neural network model ResNet-18 under the framework of Mxnet to classify and recognize the image.Under the experimental environment,the traffic sign recognition rate can reach 91.87%.The results show that the algorithm can effectively identify traffic sign information and has the distinguishing features of fast recognition speed and good portability.It proposes a new possibility for the lightweight and popularization of the intelligent traffic sign recognition system.更多还原