摘要
无论是在智能驾驶系统中,还是在智能导盲系统中,道路标线的检测都是一项重要内容。针对传统斑马线识别方法精度低、速度慢的问题,提出了利用深度可分离卷积网络改进SegNet模型的语义分割方法,通过网络爬虫以及手动数据标注,经过Tensorflow深度学习框架训练,其模拟检测达到了较好的结果。试验结果表明,由自行构建的斑马线数据集,训练后的模型每帧运算速度在59 ms内,对斑马线区域分割的像素精度达98.1%,交并比达91.6%。此运算速度以及分割精度满足大部分智能导航系统的需求,为斑马线识别的机器视觉识别提供了技术支持。
Road marking detection is an important part in not only intelligent assistant driving system but also intelligent guidance system. Aiming at the problems of low accuracy and slow speed of traditional zebra crossing recognition methods, we proposed a semantic segmentation method based on SegNet model improved by using deep separable convolutional network. Through web crawler and manual data annotation, and through Tensorflow deep learning framework training, the simulation detection achieves good results. The experimental results show that the operation speed of each frame of the trained model is within 59 ms, and the pixel accuracy of zebra crossing region segmentation is 98.1% and the intersection and merging ratio is 91.6%. The calculation speed and segmentation accuracy meet the needs of most intelligent navigation systems, and provide technical support for the machine vision recognition of zebra crossing recognition.
作者
程换新
蒋泽芹
成凯
Cheng Huanxin;Jiang Zeqin;Cheng Kai(College of Automation and Electronic Engineering,Qingdao University of Science and Technology,Qingdao 266061,China)
出处
《电子测量技术》
2020年第23期104-108,共5页
Electronic Measurement Technology
基金
国家海洋局重大专项(国海科字[2016]494号No.30)资助。
