自然场景下交通标志检测与分类算法研究  

Detection and classification of traffic signs in natural environments

作　　者：李伦波[1,2] 马广富[1] 

机构地区：[1]哈尔滨工业大学航天学院,哈尔滨150001 [2]南京理工大学自动化学院,南京210094

出　　处：《哈尔滨工业大学学报》2009年第11期29-33,38,共6页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目(60774062);高等学校博士学科点专项科研基金(20050213010)

摘　　要：在HSV彩色空间进行颜色分割的基础上,提出一种基于局部特征与模糊规则的交通标志检测算法.为了对检测出的标志进行分类理解,根据交通标志的颜色与形状特征信息进行分层分解,设计多层决策分类系统,并采用J-means聚类分析与PSO算法来优化设计PNN作为其子分类器.对晴天、多云和小雨天气状况下共3000幅图像进行了交通标志识别,该检测算法的检测率分别达到93.28%、90.25%与88.97%;所设计分类器不仅具有精简的结构,而且有较高的分类精度.A novel approach is proposed for the detection of traffic signs in natural environments. Each RGB image is converted into HSV color space, and segmented by the hue and saturation thresholds. A symmetrical detector of local binary features and a set of fuzzy rules are used to determine the shape of region of interests （ROI） in the detection algorithm. For the traffic signs classification, a classification module based on decision trees is designed, and PNN is adopted for the further classification which incorporates the J-means algorithm and Particle Swarm Optimization to optimize the networks. Experiments were conducted for the detection and classification of traffic signs, involved in 3000 images under sunny, cloudy and rainy weather conditions. Re- suits demonstrate that the proposed detection algorithm is capable of achieving the hit-rate of 93.28%, 90. 25% and 88.97% respectively, and the classification module has simple structure as well as high accuracy.

关 键 词：交通标志识别 辅助驾驶系统 局部特征检测模板 概率神经网络 J—means算法 粒子群优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]