摘要
在实景采集图像数据进行三维重建时,往往难以避免环境噪声干扰,影响建模的质量和精度,因此需要处理相关噪声以准确提取文物主体特征。针对这一问题,基于改进U^(2)-Net神经网络对校园内青铜雕塑进行了多视角的三维重建分析。在U^(2)-Net神经网络中添加注意力机制以提高网络对文物主体的分割精度,从而获取只包含文物主体特征的图像,并且对3处不同环境下的文物进行分割;利用单目多视角三维重建法,对经过前期处理的图像进行建模。实验结果表明:改进U^(2)-Net神经网络对文物主体特征的分割精度更高,且适用于多样性的环境噪声处理,经处理的图像在三维建模过程中,重投影误差减少约25%,均方根投影误差减少到原来的2/3,建模时间共减少了25%。
When collecting image data for 3D reconstruction at real scenes,it is often difficult to avoid environmental noise interference,which affects the quality and accuracy of modeling.Therefore,it is necessary to process the relevant noise to accurately extract the main features of cultural relics.To address this issue,a multi-view 3D reconstruction analysis of bronze sculptures on campus was conducted based on an improved U^(2)-Net neural network.Firstly,an attention mechanism was added to the U^(2)-Net neural network to improve the segmentation accuracy of the network for cultural relics,thereby obtaining images that only contain the features of cultural relics,and segmenting cultural relics in three different environments.Finally,the monocular multi-view 3D reconstruction method was used to model the preprocessed images.The experimental results show that the improved U^(2)-Net neural network has higher segmentation accuracy for the main features of cultural relics and is suitable for processing diverse environmental noise.The processed images reduce the reprojection error by about 25%and the root mean square projection error by 2/3 during the 3D modeling process.The modeling time is reduced by a total of 25%.
作者
杨鹏程
王嘉豪
朱婧怡
崔嘉宝
YANG Pengcheng;WANG Jiahao;ZHU Jingyi;CUI Jiabao(School of Mechanical and Electrical Engineering,Xi’an Polytechnic University,Xi’an 710048,China;Institute of Archaeology,Chinese Academy of Social Sciences,Beijing 100101,China)
出处
《西安工程大学学报》
2025年第2期102-108,共7页
Journal of Xi’an Polytechnic University
基金
陕西省科技厅自然科学基础研究计划-面上项目(2022JM-219)。
