基于SFM与深度学习融合的水果体积测量算法  

Fruit volume measurement algorithms based on SFM and deep learning

作　　者：黎兆锴 宋亚男[1] 徐荣华[1] 李发义 郑耿忠 LI Zhao-kai;SONG Ya-nan;XU Rong-hua;LI Fa-yi;ZHENG Geng-zhong(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学自动化学院,广东广州510006

出　　处：《计算机工程与设计》2023年第6期1699-1705,共7页Computer Engineering and Design

基　　金：广东省科技计划基金项目(2014A020217015、2016A020222012);广东省自然科学基金项目(2018A030313775);广东工业大学青年基金项目(18ZK0021);2020年广东工业大学本科教学工程基金项目(广工大教字〔2020〕22号、广工大教字〔2019〕70号);2018年广东工业大学高水平大学建设研究生教育创新计划基金项目(2018JGMS-09)。

摘　　要：以生活常见的水果作为研究对象,结合SFM对图像序列的深度估计以及神经网络重建三维结构的优点,提出SFM算法融合深度学习三维重建的水果体积测量算法。对单目相机采集的水果多视角图像进行研究,分析图像重建以及估计大小的方法,搭建快速、便捷估计水果实际体积算法框架。使用神经网络快速推理水果结构,解决三维重建构建稠密点云耗时长的缺点,利用多视角图像获取稀疏点云,估计目标尺寸,提高采样的便利性。实验结果表明,该算法能快速重建水果三维模型,实现简单、快速、较精确的水果体积测量。The common monomer fruit in life was taken as the research object.Through the combination of SFM for image sequence depth estimation and the advantages of neural network reconstruction of three-dimensional structure,a fruit volume measurement algorithm based on SFM algorithm and depth learning three-dimensional reconstruction was proposed.Multi-view images of fruit captured by monocular cameras were studied,image reconstruction as well as size estimation methods were analyzed,and a framework for fast and easy algorithms to estimate the actual volume of fruit was built.The neural network was used to quickly infer fruit structure,the long time-consuming shortcoming of 3D reconstruction on building dense point clouds was solved.The convenience of sampling was improved by taking advantage of multi-view image to infer target size.Experimental results indicate that the algorithm can quickly reconstruct the three-dimensional model of fruit and realize simple,fast and accurate measurement of fruit volume.

关 键 词：水果体积 计算机视觉 三维重建 体积测量 深度学习 单目相机 多视角图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]