基于部件的大类别集交通标志识别方法研究

项目名称： 基于部件的大类别集交通标志识别方法研究

项目编号： No.61271306

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 黄琳琳

作者单位： 北京交通大学

项目金额： 75万元

中文摘要： 交通标志自动识别技术在无人驾驶汽车和驾驶员辅助系统中具有重要的应用前景，部分技术已用于无人驾驶实验系统。然而，现有方法仅对少数几种类型几十种类别进行识别，难以推广到多种类型、大类别集交通标志的识别。本项目基于交通标志图形构成的特点，提出基于部件的大类别集交通标志的快速准确识别方法。基本思想是对构成交通标志的形状基元（如圆形、三角形、矩形、线条、箭头、文字、符号等）进行分割和识别，并分析符号之间的关系，得到交通标志的结构解释。研究内容包括：交通标志的部件和特征分析，基于颜色分割和区域的标志识别，基于符号检测的标志识别，基于基元和空间上下文融合的结构识别。本方法充分借鉴了计算机视觉领域目标识别的前沿理论与方法，有望提高交通标志识别的类别可扩展性和计算效率。实现的方法将在公开的交通标志图像数据集和计划采集的大类别集图像数据集上进行验证。

中文关键词： 交通标志图像；检测；识别；学习；扰动

英文摘要： Vision-based traffic sign recognition is important for unmanned driving and driver assistance systems. In the past years, many research works have been done in this field, and some techniques have been applied in unmanned driving systems. However, most of existing methods only consider a small number of traffic sign classes, say, several decades of classes. It is difficult to generalize these methods to as many as hundreds of classes in real applications. Based on the hierarchical structure of traffic signs which are composed of some common symbols organized spatially, we propose a component-based approach for large category set traffic sign recognition. This approach segments the primitives (circle, triangle, rectangle, line, arrow, character, symbol) and based on the primitive recognition and their spatial relationship, gives the structural interpretation of the traffic sign. The involved research issues include: (1) Analysis of characteristics of components and features of traffic signs, (2) Component segmentation and recognition based on color clustering, (3) Traffic sign recognition based on primitive and symbol detection, (4) Structural interpretation based on graph representation fusing primitives and spatial relationships. Our research is based on the advances in computer vision and pattern recognition,

英文关键词： traffic sign；detection；recognition；multi-scale classification；Perturbation