中圖分類號： TP391 文獻(xiàn)標(biāo)識碼： A DOI：10.16157/j.issn.0258-7998.211840 中文引用格式： 張成,，張瑞賓,，王曙道. 標(biāo)簽結(jié)合現(xiàn)實(shí)場景的交通標(biāo)志分類檢測研究[J].電子技術(shù)應(yīng)用，2022,，48(3)：27-31,，36. 英文引用格式： Zhang Cheng，Zhang Ruibin,，Wang Shudao. Research on classification and detection of traffic signs based on tags combined with real scenes[J]. Application of Electronic Technique,，2022，48(3)：27-31,，36.
Research on classification and detection of traffic signs based on tags combined with real scenes
Zhang Cheng,，Zhang Ruibin，Wang Shudao
School of Automobile and Traffic Engineering,，Guilin University of Aerospace Technology,，Guilin 541004，China
Abstract： There are a lot of researches on traffic signs in the safe driving and automatic driving of vehicles. Due to the wide variety of traffic signs and the influence of various factors, the classification and detection of traffic signs is also a challenging problem. To this end, a traffic sign classification and detection method combining tags with real road scenes is proposed. The method is divided into a data generation part and a target detection part. Experimental results show that the use of this method to generate training data can effectively train deep convolutional neural networks to achieve classification and detection of traffic signs in real scenes, and the optimized detection model has a smaller size and faster speed than the model mentioned in the article.
Key words : traffic signs,；automatic driving,；data enhancement；DCNN,；detection
0 引言
在車輛安全和自動駕駛領(lǐng)域,，交通標(biāo)志的檢測有著很大的實(shí)用價值。真實(shí)的交通場景復(fù)雜多變,，交通標(biāo)志易受到光照,、雨霧和遮擋等外在因素的影響。傳統(tǒng)的檢測算法根據(jù)交通標(biāo)志的形狀,、顏色等特點(diǎn)[1-6],，使用不同尺度大小的滑動窗口對待檢測圖片進(jìn)行潛在目標(biāo)區(qū)域提取，之后對潛在區(qū)域通過HOG(Histograms Of Oriented Gradient)[7],、Gabor[8],、Haar-like[9]等人工提取特征方法，結(jié)合支持向量機(jī),、BP(Back Propagation)神經(jīng)網(wǎng)絡(luò),、極限學(xué)習(xí)機(jī)和最近鄰算法等常用的機(jī)器學(xué)習(xí)算法完成分類的任務(wù)。這些方法若要完成細(xì)分類檢測問題,，工作量巨大,，且最后的效果也不盡理想。
深度學(xué)習(xí)方法不同于前面的方法，它利用深度卷積神經(jīng)網(wǎng)絡(luò)完成特征提取,，實(shí)現(xiàn)交通標(biāo)志的檢測任務(wù),。目前常用方法可分為候選區(qū)域和邏輯回歸。候選區(qū)域的網(wǎng)絡(luò)(如RCNN(Region-Convolutional Neural Network)[10],、Faster R-CNN[11])先提取出候選的區(qū)域特征,，之后根據(jù)候選區(qū)域的特征進(jìn)行位置和類別的學(xué)習(xí)，這種方法突出了出色的檢測精度,，犧牲了計(jì)算的時間和存儲資源,；邏輯回歸的網(wǎng)絡(luò)(如YOLO(You Only Look Once)[12]、SSD(Single Shot Detector)[13])直接將預(yù)測邊界框的坐標(biāo)和類別設(shè)置為回歸問題,，提升了網(wǎng)絡(luò)的檢測速度,，但是針對具體的任務(wù)網(wǎng)絡(luò)模型還需要進(jìn)一步調(diào)整，且完成交通標(biāo)志檢測的研究需要數(shù)據(jù)龐大的交通標(biāo)志數(shù)據(jù)集,。
