首页|融合Faster-RCNN的双目视觉对路面三维纹理的重构测量

融合Faster-RCNN的双目视觉对路面三维纹理的重构测量

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为了增强双目视觉技术对极端光照条件的适应能力，改善其在路面三维纹理测量中的光照适用范围，引入Faster-RCNN目标检测网络，构建融合Faster-RCNN的激光约束改进双目视觉重构算法，开发路面三维纹理双目测量装置，并对改进前后的算法分别开展激光约束目标提取、路面三维纹理重构和实际场景中应用的效果进行分析。结果表明，Faster-RCNN的融合增强了算法对目标的提取能力，提高了算法对极端光照的适应能力，算法改进后在5 Lux和1 050 Lux两种极端光照下的高程差平均偏差分别减少了 68。6％和70。0％，达到0。11 mm和0。09 mm。同时，改进算法在实际场景中也能保持较好的测量精度与稳定性，其在测试区域内高程差的最大偏差和平均偏差均能满足路面三维纹理测量对精度的要求。
Reconstruction of Three-dimensional Pavement Texture based on Improved Binocular Vision Incorporated Faster-RCNN
In order to enhance the adaptability of binocular vision to extreme lighting conditions and improve its lighting applicability in the measurement of three-dimensional(3D)pavement texture,Faster RCNN target detection network is introduced into this study.Then the laser constrained improved binocular vision reconstruction algorithm incorporating Faster RCNN is constructed and the binocular measuring device of pavement 3D texture is also developed.In addition,the results of laser constrained target extraction,pavement 3D texture reconstruction and actual scene application are further analyzed before and after improvement of algorithm.The results show that the incorporation of Faster RCNN can enhance the ability of extracting the target and improve the ability to adapt to extreme light.Under the two extreme lighting conditions of 5 Lux and 1 050 Lux,the average deviation of elevation difference is reduced by 68.6％and 70.0％to 0.11 mm and 0.09 mm,respectively.Simultaneously,the improved algorithm can also maintain good measurement accuracy and stability in the actual scene,and its maximum and average deviation of elevation difference can still meet the accuracy requirements of 3D pavement texture measurement.

road engineeringpavement three-dimensional textureFaster-RCNN target detection networkbinocular visionlaser constraint

王元元、李仁杰、薛金顺、张华、刘勇

展开 >

湖北文理学院纯电动汽车动力系统设计与测试湖北省重点实验室 襄阳市 441053

中基华工程管理集团有限公司重庆高新区分公司 重庆市 401331

襄阳路桥建设集团有限公司 襄阳市 441002

道路工程 路面三维纹理 Faster-RCNN目标检测网络 双目视觉 激光约束

国家自然科学基金项目湖北省自然科学基金计划(联合基金项目)湖北文理学院研究生质量工程资助项目

521784222023AFD057YZ3202303

2024

公路
中国交通建设集团有限公司

公路

CSTPCD北大核心
影响因子：0.54
ISSN：0451-0712
年,卷(期)：2024.69(6)