稠密化重建稀疏性特征多图像3D构建实现  

Implementation of Multi-image 3D Reconstruction with Densely Reconstructing Sparse Characteristics

作　　者：张海军 钟向阳[1,2] 陈映辉 ZHANG Haijun;ZHONG Xiangyang;CHEN Yinghui(Guangdong Provincial Key Laboratory of Conservation and Precision Utilization of Characteristic Agricultural Resources in Mountainous Areas,Meizhou,Guangdong 514015,China;School of Computing,Jiaying University,Meizhou,Guangdong 514015,China;School of Mathematics,Jiaying University,Meizhou Guangdong 514015,China)

机构地区：[1]广东省山区特色农业资源保护与精准利用重点实验室,广东梅州514015 [2]嘉应学院计算机学院,广东梅州514015 [3]嘉应学院数学学院,广东梅州514015

出　　处：《昆明理工大学学报（自然科学版）》2022年第1期46-53,共8页Journal of Kunming University of Science and Technology(Natural Science)

基　　金：国家自然科学基金项目(61171141,61573145);广东省自然科学基金重点项目(2014B010104001,2015A030308018);广东省普通高等学校人文社会科学省市共建重点研究基地(18KYKT11);广东省嘉应学院自然科学基金重点项目(2017KJZ02);教育部产学合作协同育人项目(201802153047);2019年广东省教育厅高校特色创新项目(2019KTSCX169);广东省嘉应学院自然科学基金项目(2021KJY05)。

摘　　要：给出了一种稠密化重建稀疏性特征多图像3D重建的方法.首先对采集的目标样本图进行特征提取,基于特征点通过绝对二次曲面约束等方法分层重建投影矩阵;其次对特征点基于自标定矩阵逆映射成稀疏3D点集,创建面片并扩展邻域生成稠密点集;最后经滤波生成精确3D目标模型.实验结果表明重建的物体可视性较高,稳定性好,具有总体性能优良等优点.The paper presents a method for the 3 D reconstruction of multi-image, which is realized by densely reconstructing sparse characteristics. Firstly, feature extraction is carried out for the collected target sample images, and the projection matrix is hierarchically reconstructed based on feature points by the methods of absolute quadric constraint and so on. Secondly, feature points are inversely mapped to sparse 3 D point sets based on the self-calibration matrix, and create the surface patch and extend the neighborhood to generate a dense set of points. Finally, the accurate 3 D target model is generated by filtering. The experimental results show that the reconstructed objects have the characteristics of high visibility, good stability and the excellent overall performance and so on.

关 键 词：射影及三维重建 绝对二次曲面 自标定 面片 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]