摘要
现阶段无人驾驶汽车主要依靠视觉技术来完成车辆的环境感知,在道路识别领域,视觉技术能准确识别出道路可行驶区域。为了增强无人驾驶车辆在非结构化道路对场景区域的判别能力,本文基于SegNet、ENet、UNet 3种在多分类任务中取得较好成果的分割网络架构,通过对模型网络或参数的调整和修改,提出一种能很好应用到非结构化道路区域分割问题的分割模型。通过拍摄并制作标签数据集,采取不同的评价指标进行分析,得到最佳的道路区域分割模型,用于预测非结构化道路的可行驶区域。实验证明,相比较于传统的非结构化道路分割的区域生长模型,本文提出的分割模型在分割精度上有明显提升。
At present,driverless vehicles mainly rely on vision technology to complete the vehicle environment perception.In the field of road recognition,the task of vision technology is to accurately identify the drivable area of the road.In order to enhance the discrimination ability of driverless vehicles in unstructured road for scene area,based on SegNet,ENet and UNet,this paper proposes a segmentation model for unstructured road image segmentation by adjusting and modifying the model network or parameters.By shooting and making label data sets,and taking different evaluation indexes for analysis,the best road region segmentation model is obtained to predict the driving area of unstructured road.Experimental results show that compared with the traditional region growing model of unstructured road segmentation,the segmentation accuracy of the proposed model is significantly improved.
作者
赵晋燕
罗素云
陈杨钟
ZHAO Jinyan;LUO Suyun;CHEN Yangzhong(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Dagong Technology(Shanghai)Co.,Ltd.,Shanghai 200000,China)
出处
《智能计算机与应用》
2021年第11期148-152,156,共6页
Intelligent Computer and Applications
作者简介
赵晋燕(1997-),男,硕士研究生,主要研究方向:机器视觉、图像处理;罗素云(1975-),女,硕士,副教授,主要研究方向:无人驾驶汽车环境感知及控制;陈杨钟(1983-),男,硕士,工程师,主要研究方向:控制理论与控制工程。
