无人车行驶环境图像的几何测距  

Geometric Ranging of Unmanned Vehicle Driving Environment Image

作　　者：代金坤 罗玉涛[1] 梁伟强 Dai Jinkun;Luo Yutao;Liang Weiqiang(School of Mechanical & Automotive Engineering, South China University of Technology, Guangzhou 510640;GAC Automotive Engineering Institute, Guangzhou 510640)

机构地区：[1]华南理工大学机械与汽车工程学院,广州510640 [2]广汽集团汽车工程研究院,广州510640

出　　处：《汽车工程》2020年第8期1034-1039,1049,共7页Automotive Engineering

基　　金：广东省科技计划项目(2016B010132001)资助。

摘　　要：本文中提出了一种无人车行驶环境图像的几何测距方法。首先,利用迁移学习的方法改进Tiny-YOLOv2网络模型,对需识别的物体进行训练与检测,标记物体并定位物体在图像中的位置。其次,提出了一种通过物体分类、边缘检测及边缘拟合的方法,进一步提取物体的图像信息。最后,建立了一种基于空间几何理论的测距模型,结合物体先验信息实现了物体的距离测量。该方法在4 m以内88%以上的测量值误差不超过0.2 m,同时测量误差并没有随着距离的增加而有较大变化。A geometric ranging method for the image of the driving environment of the unmanned vehicle is proposed.Firstly,the migration-learning method is used to improve the Tiny-YOLOv2 network model so as to train and detect the object to be identified,mark the object and locate the position of the object in the image.Secondly,a method of object classification,edge detection and edge fitting is proposed to further extract image information of the object.Finally,a ranging model based on spatial geometry theory is established,and the distance measurement of the object is realized by combining the prior information of the object size.With this method,more than 88%of the measurement error within 4 m is less than 0.2 m,and the measurement error does not change much with the increase of distance.

关 键 词：几何测距 单目视觉 卷积神经网络 迁移学习 图像分割 

分 类 号：G63[文化科学—教育学]