基于CNN算法的交通标志检测与识别

Detection and Recognition of Traffic Signs Based on CNN Algorithm

交通标志的检测与识别已经成为计算机视觉与智能交通系统的热点研究方向。在AlexNet网络的基础上进行检测,同时使用RPN网络结构进行候选区域的提取,在池化层后利用Batch Normalization层将每层的输入值进行归一化。在识别阶段提出多级分类识别的算法,先后将检测阶段产生的潜在目标区域通过SVM分类器和CNN网络识别,在设计CNN网络时引入Inception结构,利用多卷积核学习更为丰富的特征。实验证明,提出的交通标志检测与识别算法在时效性和正确率都有极好的表现。

The detection and recognition of traffic signs has become a hot research direction of computer vision and intelligent transportation systems.Detects on the basis of AlexNet network, and uses the RPN network structure to extract candidate regions. After the pooling layer, the input value of each layer is normalized by using the Batch Normalization layer. In the recognition phase, a multi-level classification and recognition algorithm is proposed. The potential target areas generated during the detection phase are identified by the SVM classifier and the CNN network. The Inception structure is introduced when designing the CNN network. The multi-convolution kernel can learn more abundant feature. Experiments prove that the proposed traffic sign detection and recognition algorithm has excellent performance in timeliness and accuracy.