基于低成本多视角立体视觉的辣椒苗三维重建与策略优化  

Three-dimensional Reconstruction and Strategy Optimization of Pepper Seedlings Based on Low-cost Multi-view Stereo Vision

作　　者：朱磊[1,2] 程瑞英 丁一民 孙振源 郭政 江伟 ZHU Lei;CHENG Rui-ying;DING Yi-min;SUN Zhen-yuan;GUO Zheng;JIANG Wei(School of Civil and Hydraulic Engineering,Ningxia University,Yinchuan 750021,Ningxia,China;Ningxia Hui Autonomous Region Key Laboratory of Digital Water Control of Yellow River Water Network,Yinchuan 750021,Ningxia,China)

机构地区：[1]宁夏大学土木与水利工程学院,宁夏银川750021 [2]宁夏回族自治区黄河水联网数字治水重点实验室,宁夏银川750021

出　　处：《中国农村水利水电》2025年第4期123-129,共7页China Rural Water and Hydropower

基　　金：宁夏自然科学基金优秀青年项目(2023AAC05013);国家自然科学基金项目(52269015,52209059)。

摘　　要：基于立体视觉的三维重建技术有效降低了植株表型无损测量中的影像获取成本,但算力消耗较大。开展三维重建策略优化研究,对于提高植株表型信息的采集效率具有重要意义。研究构建了以手机为影像获取手段的多视角成像系统,获取了30株辣椒苗的影像数据,进一步结合运动恢复结构与多视图立体视觉(SFM-MVS,Structure From Motion-MultiPle View Stereo)算法,在2160个不同分辨率和不同数量影像组合情景下,分别重建了辣椒苗三维点云。通过重建速度、精度、稳定性以及植株表型参数(叶长、叶宽)准确性的评估,优化植株三维点云重建方案。结果表明,当分辨率为480 p,图片数量大于45幅时,重建成功率达到80%,点云间平均距离误差均小于0.05 cm,提取的表型参数值与实测值的R2均到了0.96以上。同时,该情景下单株平均重建时间为344 s,仅为参照情景(分辨率为1080 p、图片数量为120幅)耗时的10%。综上,在辣椒苗的重建中可将分辨率和图像数量分别设置为480 p和45,从而优化表型测量效率。研究结果可为基于三维重建的辣椒苗表型参数快速无损测量提供参考。The three-dimensional reconstruction technology based on stereo vision effectively reduces the cost of image acquisition in non-destructive measurement of plant phenotype,but the computing power consumption is large.It is of great significance to carry out research on optimization of 3D reconstruction strategy to improve the extraction efficiency of plant phenotypic information.In this study,a multi-view imaging system using mobile phones as the means of image acquisition was constructed,and the image data of 30 pepper seedlings were obtained,and the three-dimensional point clouds of pepper seedlings were reconstructed under 2160 different resolutions and different number of image combinations by combining the motion recovery structure and multi-view stereo(SFM-MVS)algorithm.Through the evaluation of the reconstruction speed,accuracy,stability and accuracy of plant phenotypic parameters(leaf length and leaf width),the 3D point cloud reconstruction scheme of plants was optimized.The results showed that when the resolution was 480p and the number of images was greater than 45,the reconstruction success rate reached 80%,the average distance error between point clouds was less than 0.05 cm,and the R2 of the extracted phenotypic parameter value and the measured value was more than 0.96.At the same time,the average reconstruction time of a single plant under this scenario was 344 seconds,which was only 10% of the time taken by the reference scenario(resolution of 1080 p and number of images of 120).In summary,the resolution and number of images can be set to 480 p and 45,respectively,in the reconstruction of pepper seedlings,so as to optimize the phenotypic measurement efficiency.The results of this study can provide a technical reference for the rapid and non-destructive measurement of phenotypic parameters of pepper seedlings based on three dimensional reconstruction.

关 键 词：三维重建 多视角立体视觉 点云模型 表型参数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S24[自动化与计算机技术—计算机科学与技术]