作者：刘桂雄1, 肖天歌1, 陈国宇2, 黄坚1,3
作者单位：1. 华南理工大学机械与汽车工程学院，广东 广州 510641;
2. 广州能源检测研究院，广东 广州 511447;
3. 广州计量检测技术研究院，广东 广州 510663
关键词：储罐;体积测量;深度学习;多视图立体视觉;MVSNet
摘要：
针对目前储罐在位体积测量需求大、移位测量困难的问题，该文提出一种基于MVSNet多视角立体深度学习的储罐在位体积测量方法。首先，提出面向在位储罐体积测量的MVSNet深度预测改进多视图立体视觉方法，结合基于增量式运动恢复结构的储罐显著特征稀疏重建与相机姿态计算技术、基于MVSNet深度学习深度预测技术，获得储罐体积测量关键结构的稠密三维点云；然后，提出基于立体几何拟合在位储罐体积测量方法，旋转储罐点云与地面配准，并基于法线信息双阈值约束点云拟合储罐圆形拓扑结构，实现储罐体积测量。在2种储罐上进行初步实验，结果表明：该文方法提取到的高质量储罐点云数量比经典COLMAP框架分别增加15.6%、13.2%，点云提取时间分别缩短34.7%、39.2%，满足储罐在位体积测量需求。
An in-situ volume measurement method for storage tanks based on MVSNet multi-view stereo deep learning
LIU Guixiong1, XIAO Tiange1, CHEN Guoyu2, HUANG Jian1,3
1. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510641, China;
2. Guangzhou Institute of Energy Testing, Guangzhou 511447, China;
3. Guangzhou Institute of Measurement and Testing Technology, Guangzhou 510663, China
Abstract: For the current storage tank in-position volume measurement demand is large, but the measurement efficiency is low. This paper proposes an in-situ volume measurement technique for storage tanks, MVSNet-based multi-view stereo deep learning. Firstly, the MVSNet depth prediction improved multi-view stereo vision technique for in-situ tank volume measurement is proposed. It combining the incremental structure from motion-based tank salient features sparse reconstruction and camera pose calculation technique, and the MVSNet depth learning based depth prediction technique, the dense 3D point cloud of the key structure of the tank volume measurement is obtained. Secondly, an in-situ tank volume measurement technique based on stereometric fitting is proposed. It rotates the tank point cloud to make the ground alignment, and fit the tank circular topology based on the normal information double threshold constrained point cloud, in order to achieve the tank volume measurement. Preliminary experiments are conducted on two types of tanks, the results show that , the number of high-quality tank point clouds, which is extracted by this paper, increases by 15.6% and 13.2% respectively compared with that of the COLMAP framework. The point cloud extraction time was reduced by 34.7% and 39.2%, respectively. This method meets the in-situ tank volume measurement requirements.
Keywords: storage tanks;volume measurement;deep learning;multi-view stereo;MVSNet
2023, 49(1):26-30,49 收稿日期: 2021-08-29;收到修改稿日期: 2021-10-11
基金项目: 广东省自然科学基金-面上项目(2020A1515010947)；广州市科技计划项目(202002030439)；广东省重点领域研发计划项目(2019B010154003)
作者简介: 刘桂雄(1968-),男,广东揭阳市人,教授,博导,主要从事先进传感与仪器研究
参考文献
[1] 魏凯, 宋述古, 刘子勇. 基于三维激光扫描原理的球形罐容量计量方法研究[J]. 计量学报, 2015, 36(6): 607-609
[2] 屈玉福, 章平. 单视旋转显微镜三维测量技术研究[J]. 中国测试, 2018, 44(3): 1-5
[3] 张青春, 王旺, 杨广栋. 基于多目立体视觉的机械臂智能控制系统设计[J]. 中国测试, 2020, 46(12): 79-85
[4] 胡鹏程, 郭焱, 李保国, 等. 基于多视角立体视觉的植株三维重建与精度评估[J]. 农业工程学报, 2015, 31(11): 209-214
[5] 宋时德, 李淼, 张健, 等. 面向以多视角立体匹配获取的植株三维点云的去噪方法[J]. 计算机应用, 2017, 37(S2): 141-145
[6] 赵志文, 孙雪岚, 谷蕾蕾, 等. 运动恢复结构多视角立体重构在河工模型地形测量中的应用[J]. 泥沙研究, 2019, 44(5): 14-20
[7] 周静静, 郭新宇, 吴升, 等. 基于多视角图像的植物三维重建研究进展[J]. 中国农业科技导报, 2019, 21(2): 9-18
[8] 阮琼瑶, 李文达, 张尚弘, 等. 基于无人机和SfM的天津港堆场散料体积测量[J]. 水利水电技术(中英文), 2021, 52(6): 198-205
[9] 张越, 翟福琪, 蔡孙宝, 等. 基于点云数据的植物叶片特征提取及三维重建[J]. 中国测试, 2021, 47(8): 6-12
[10] 庄仁诚, 陈鹏, 傅瑶, 等. 列车车轮三维结构光检测中的点云处理研究[J]. 中国测试, 2021, 47(2): 19-25
[11] YAO Y, LUO Z, LI S, et al. Mvsnet: Depth inference for unstructured multi-view stereo[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018.
[12] 韩婧. 基于深度学习的多视图三维重建[D]. 武汉: 武汉大学, 2019
[13] 陈秋敏. 基于深度学习的多视图物体三维重建研究[D]. 成都: 电子科技大学, 2020.
[14] GALLIANI S, LASINGER K, SCHINDLER K. Massively parallel multiview stereopsis by surface normal diffusion[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015.
[15] 孙翔. 点云数据边缘提取及几何特征测量算法研究[D]. 秦皇岛: 燕山大学, 2018.
[16] SCHONBERGER J L, FRAHM J M. Structure-from-motion revisited[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016.
[17] SCHÖNBERGER J L, ZHENG E, FRAHM J M, et al. Pixelwise view selection for unstructured multi-view stereo[C]//European Conference on Computer Vision, 2016.
