基于双通道卷积神经网络的交通标志识别算法

Traffic sign recognition algorithm based on double-channel CNN

针对交通标志图像易受复杂背景、光照、运动模糊等影响导致识别率低和识别速度慢的问题,提出了基于非对称双通道卷积神经网络的交通标志识别方法。通过不同网络结构的两通路提取丰富的特征信息,上层通路使用跃层连接提取的浅层局部特征和深层全局特征,与下层通路提取的精细特征在全连接层进行融合,并使用激活函数LReLUs代替脆弱的ReLU,提高准确率。将实验结果与其他算法进行比较,证明所提算法的识别率和识别速度均优于其他算法,具有一定的先进性和鲁棒性。

Aiming at the problem that traffic sign images are susceptible to complex background,illumination and motion blur,resulting in low recognition rate and slow recognition speed,a traffic sign recognition method based on asymmetric two-channel convolutional neural network is proposed.Rich feature information is extracted through two paths of different network structures.The upper layer path uses shallow local features and deep layer global features extracted by hop layer connection and the fine features extracted from the lower layer path are fused in fully connected layer,and the activation function LReLUs is used instead of fragile ReLU,which improves accuracy.The experimental results are compared with other algorithms,it is proved that the recognition rate and recognition speed of the proposed algorithm are prior to other algorithms,which has certain advancement and robustness.