基于条件随机场与多尺度卷积神经网络的交通标志检测与识别

Traffic sign detection and recognition based on conditional random field and multi-scale convolutional neural network

交通标志的检测与识别主要可以分为候选区域的提取与识别两个阶段。在候选区域提取阶段,首先生成颜色先验特征图和颜色概率特征图,然后利用条件随机场(CRF)对道路场景图像的颜色先验特征图与颜色概率特征图进行融合,得到交通标志显著性图像,进而通过最大稳定极值区域(MSER)算法进行候选区域提取,克服光照强烈变化以及复杂背景对提取结果的影响。在候选区域的识别阶段,通过多尺度卷积神经网络来完成交通标志的识别。

Traffic sign detection and recognition can be divided into two stages:extracting Regions Of Interest(ROI)and recognizing candidate regions.In the first stage,a color probability map and a color prior feature map were generated,then Conditional Random Field(CRF)was used to fuse the color prior feature map and the color probability feature map of the road scene image to obtain a traffic sign saliency map,after that Maximally Stable External Regions(MSER)method was used on this saliency map to get candidate regions.It can overcome the influence of light intensity change and complicated background on extracted results.In the second stage,a multi-scale CNN was used to classify the candidate regions.