基于双目立体视觉和轻量化神经网络的交通标志分割和识别

Traffic Sign Segmentation and Recognition Based on Binocular Stereo Vision and Lightweight Neural Network

为了降低交通标志图像分割运算量,提出一种基于双目立体视觉和轻量化神经网络的交通标志分割和识别方法。使用已标定的双目立体视觉相机采集交通标志图像,并将其作为轻量化卷积神经网络的输入,通过卷积运算和池化运算提取交通标志的特征。在全连接层中,采用极限学习机和权值修正方法修正输出权值,从而得到交通标志的分割结果。实验结果表明,所提方法能够有效采集高精度的交通标志图像,并降低图像分割运算的复杂性,从而提高交通标志图像的应用性。

In order to reduce the calculation of traffic sign image segmentation,this study proposes a traffic sign segmentation and recognition method based on binocular stereo vision and lightweight neural network.A calibrated binocular stereo vision camera is used to collect traffic sign images,which are used as inputs to a lightweight convolutional neural network.The features of traffic signs are extracted through convolution and pooling operations.In the fully connected layer,extreme learning machine and weight correction methods are used to correct the output weights,thereby obtaining the segmentation results of traffic signs.The experimental results show that this method can effectively collect high-precision traffic sign images and reduce the complexity of image segmentation operation,thereby improving the applicability of traffic sign images.