基于连通域分割与SVM的城轨百米标检测与识别算法

Detection and Recognition Algorithm for 100-Meter Signage in Urban Rail Transit Based on Connected Domain Segmentation and SVM

城轨道路沿线的百米标志牌是统计行车里程、辅助建图定位的重要参考信息,对其的检测与识别将有助于轨道交通车辆的智能驾驶与维护。文章基于传统的检测与识别技术归纳总结了一种百米标检测识别通用算法,其基于收缩算法进行检测定位,再基于模板匹配进行百米标上的数字识别;同时,对通用算法进行了优化改进,提出了一种基于连通域分割检测与支持向量机(SVM)数字识别的百米标检测识别算法。首先,在对图像进行二值化、形态学填充等预处理后,基于连通域分割剔除噪点斑块并完成百米标定位;然后,对定位得到的百米标进行背景噪点剔除、倾斜校正及字符分割;最后,再将获得的字符输入SVM模型进行识别。实验结果表明,优化后的算法对于百米标的检测准确率提升了约34.4个百分点,识别准确率提高了约25.6个百分点。

Hundred-meter signage within urban rail transit systems offer important benchmark information for train mileage statistics and aid in mapping and positioning.The detection and recognition of these markers contribute to the intelligent operation and maintenance of rail transit vehicles.This study identified a general algorithm for detecting and recognizing 100-meter signage,based on traditional techniques in this field.This approach employs a shrinkage algorithm for detection and positioning,and adopts template matching for digit recognition on the 100-meter signage.This general algorithm was then modified and optimized,leading to the development of a 100-meter signage detection and recognition algorithm that utilizes connected domain segmentation for detection and a support vector machine(SVM)for digit recognition.The process begins with image preprocessing,which includes binarization and morphological filling.Following this,noise and patches are removed leveraging connected domain segmentation,to facilitate the localization of the 100-meter signage.The positioned signages are then processed for background noise removal,tilt correction,and character segmentation.Finally,the extracted characters are input into the SVM model for recognition.The optimized algorithm demonstrated significant improvements,with detection accuracies increasing by approximately 34.4 percentage points and recognition accuracies by approximately 25.6 percentage points for 100-meter signage in subsequent experiments.