测绘通报 ›› 2023, Vol. 0 ›› Issue (3): 94-98.doi: 10.13474/j.cnki.11-2246.2023.0079

• 工程测量分会年会(2022年)优选论文 • 上一篇    下一篇

深度学习倾斜三维模型车辆去除算法

丁乐乐, 刘艳飞, 孟凡效, 张涛, 王震, 潘宇明   

  1. 天津市勘察设计院集团有限公司, 天津 300000
  • 收稿日期:2022-07-05 发布日期:2023-04-04
  • 通讯作者: 刘艳飞。E-mail:yanfeiliu@whu.edu.cn
  • 作者简介:丁乐乐(1986-),男,硕士,高级工程师,研究方向为大地测量学与测量工程、高分遥感智能解译。E-mail:kcydll@126.com
  • 基金资助:
    天津市重点研发计划(22YFYSHZ00120);住房和城乡建设部科学技术计划项目(2002-K-050)

Vehicle clearance based deep learning for oblique 3D model

DING Lele, LIU Yanfei, MENG Fanxiao, ZHANG Tao, WANG Zhen, PAN Yuming   

  1. Tianjin Survey Design Institute Group Co., Ltd., Tianjin 300000, China
  • Received:2022-07-05 Published:2023-04-04

摘要: 针对倾斜三维模型中车辆运动导致的车辆变形问题,本文提出了深度学习倾斜三维模型车辆去除算法。首先利用深度目标检测网络对三维模型中的车辆进行检测,定位车辆的位置;然后利用深度图像修复算法对车辆进行擦除,自动填充道路纹理。相较于基于Photoshop的手工车辆去除方法,该方法可有效提高作业效率。

关键词: 三维模型, 目标检测, 深度图像修复, 车辆去除

Abstract: To solve the texture mapping problem caused by the moved car in oblique 3D model, the vehicle clearance method based on deep learning (VCDL) is proposed in this paper. In the proposed method, the vehicles are detected based on deep-object-detection CNN, obtaining the location of the vehicles. And then the vehicles are removed from the model based on deep image inpainting network and fills road textures automatically. Compared with hand-craft vehicles clearance based on Photoshop, the method proposed in this paper can effectively improve the efficiency.

Key words: oblique 3D model, object-detection, deep image inpainting, vehicle clearance

中图分类号: