基于深度属性学习的交通标志检测  

Deep attribute learning based traffic sign detection

作　　者：王方石[1] 王坚 李兵[2,3] 王博[2,3] 

机构地区：[1]北京交通大学软件学院,北京100044 [2]中国科学院自动化研究所,北京100190 [3]中国科学院模式识别国家重点实验室,北京100190

出　　处：《吉林大学学报（工学版）》2018年第1期319-329,共11页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(61370038;61472421)

摘　　要：为了弥补交通标志底层图像到高层语义之间的鸿沟,本文引入交通标志的形状、颜色、图案内容三种视觉属性,在卷积神经网络(Convolutional neural network,CNN)中加入属性学习(Attribute learning)约束,同时进行交通标志属性学习和分类学习,提出了一种基于深度属性学习的交通标志检测方法。并在公开数据集Sweden traffic sign detection dataset(STSD)和German traffic sign detection dataset(GTSD)上进行的实验结果表明,该方法能够有效地提高交通标志检测的准确率和召回率。A traffic sign detection method based on deep attribute learning was proposed. To make up the gap between raw image and high level semantics, three visual attributes, including shape, color and pattern, were introduced. Attribute learning was added to train Convolutional Neural Network （CNN）, where attribute learning and classification learning were carried out simultaneously. Experimental results on datasets Sweden Traffic Sign Detection Dataset （STSD） and German Traffic Sign Detection Dataset （GTSD） show that the proposed method can effectively improve the precision and recall in terms of traffic sign detection.

关 键 词：信息处理技术 交通标志检测 深度属性学习 卷积神经网络 

分 类 号：TN911.73[电子电信—通信与信息系统]