基于排水管道检测视频的相机姿态估计优化方法研究  

Research on Optimization Method of Camera Pose Estimation Based on Drainage Pipeline Detection Video

作　　者：郑家祥 张驰 彭寿海 王万琼 王福芝 王殿常 尹炜 成浩科 ZHENG Jiaxiang;ZHANG Chi;PENG Shouhai;WANG Wanqiong;WANG Fuzhi;WANG Dianchang;YIN Wei;CHENG Haoke(National Engineering Research Center of Eco-Environment in the Yangtze River Economic Belt,Wuhan 430014,China;China Three Gorges Corporation,Wuhan 430014,China;Yangtze Ecology and Environment Co.,Ltd.,Wuhan 430062,China)

机构地区：[1]长江经济带生态环境国家工程研究中心,武汉430014 [2]中国长江三峡集团有限公司,武汉430014 [3]长江生态环保集团有限公司,武汉430062

出　　处：《给水排水》2025年第3期130-138,共9页Water & Wastewater Engineering

基　　金：国家重点研发计划项目(2022YFC3203200);中国长江三峡集团科技项目(WWKY-2020-0593)。

摘　　要：管道三维重建的研究现状不能够满足工程基础条件和应用需求,关键问题在于相机姿态估计作为视觉三维重建的重要基础,其时间成本高且准确率低,易受视频质量影响从而导致场景重建困难甚至失败。面向运动姿态恢复提出一种基于神经辐射场新视角合成的评价方法,适用于缺少真值位姿参考的姿态估计效果评价。考虑管道检测视频场景视深大、视野小的复杂环境特点,提出一种基于字符识别和深度估计技术的近实时掩码初始化方法,用于优化姿态估计过程。实验结果表明,掩码初始化方法在GPU环境下平均减少35.99%迭代时间,姿态估计效果整体略微增加。所提出的相机姿态估计优化方法实现了加速运动姿态恢复的迭代过程,能够提高排水管道点云初始化的效率,为排水管道场景的三维重建提供重要基础支撑。The current research status of pipeline 3D reconstruction is not able to meet the engineering basic conditions and application requirements,and the key problem is that the camera pose estimation,as an important basis of visual 3D reconstruction,has high time cost and low accuracy,and is easily affected by the video quality,which leads to the scene reconstruction difficulties or even failures.In this paper,we propose an evaluation method based on the novel view synthesis of neural radiation field for the evaluation of the pose estimation effect in the absence of a truth position reference for the level of motion pose reconstruction.Considering the complex environmental characteristics of pipeline inspection video scene with large viewing depth and small field of view,the paper proposes a near real-time mask initialization method based on character recognition and depth estimation for optimizing the pose estimation process.The experimental results show that the mask initialization method reduces the iteration time by 35.99%on average in the GPU environment,and the pose estimation effect increases slightly overall.The proposed camera pose estimation optimization method achieves the iterative process of accelerated motion pose recovery,which can improve the efficiency of pipeline point cloud initialization and provide important basic support for the 3D reconstruction of the drainage pipeline.

关 键 词：三维重建 姿态估计 神经辐射场 排水管道 

分 类 号：TU990[建筑科学—市政工程]