摘要
实时、准确的轨道定位是轨道检测及维护的基本条件,轨道定位的常用方法为接触网标识牌定位方法,存在识别率低、效率低等问题,本文提出基于显著性检测的接触网标识牌识别算法弥补这些缺点。首先,利用马尔科夫链显著性检测算法定位接触网立柱,获取包含标识牌的感兴趣区域;其次,采用形态学与垂直投影的方法对标识牌单个数字进行分割;最后,利用多层感知机神经网络(Multi-Layer Perceptron,MLP)分类器完成标识牌数字识别。通过实验验证了该图像处理算法具有极高的可行性,接触网立柱定位及标识牌识别的成功率可达93%以上,为轨道定位提供新思路。
Real-time and accurate track positioning is the basic condition for track inspection and maintenance.The common method for track zone positioning is the catenary sign positioning method,but there is a problem of low recognition rate.Therefore,a catenary sign recognition algorithm based on saliency detection is proposed to solve this shortcoming.First,the Markov chain saliency detection algorithm is used to locate the catenary column and obtain the area of interest that contains the sign;then the morphology and vertical projection methods is used to segment the single number of the sign;finally the multilayer perceptron neural network(Multi-Layer Perceptron(MLP))classifier is used to complete the number recognition of the sign.Experiments have verified that the image processing algorithm is extremely feasible,and the success rate of catenary column positioning and sign recognition can reach more than 93%,providing new ideas for track zone positioning.
作者
蒋丽洁
柴晓冬
李立明
郑树彬
JIANG Lijie;CHAI Xiaodong;LI Liming;ZHENG Shubin(College of Urban Rail Transit Institute,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《智能计算机与应用》
2020年第9期95-99,108,共6页
Intelligent Computer and Applications
基金
国家自然科学基金(51975347)
上海市科委重点支撑项目(18030501300)。
