基于双目视觉的弱纹理凸多面体三维建模  

Research on 3D Modeling of Weakly Textured Convex Bodies Based on Binocular Vision

作　　者：坎标[1] 郝金鹏 王烨[1] 马顺喜 KAN Biao;HAO Jin-peng;WANG Ye;MA Shun-xi(School of Mechanical Engineering and Rail Transit,University of Changzhou,Changzhou Jiangsu 213164,China;Jiangsu Linyang Machinery Co.,Ltd,Changzhou Jiangsu 213311,China)

机构地区：[1]常州大学机械与轨道交通学院,江苏常州213164 [2]江苏羚羊机械有限公司,江苏常州213311

出　　处：《计算机仿真》2025年第3期428-431,496,共5页Computer Simulation

基　　金：国家自然科学基金(51105046)。

摘　　要：弱纹理平面由于像素变化小,利用其自身特征点来进行视觉识别和测量有较大难度。针对由多个弱纹理平面构成的凸多面体,依次采用多个角度光照使相邻平面之间形成明暗差,确定凸多面体的面域和棱边。以相邻棱边的交点作为建模的角部特征点,再根据角部特征点邻近像素点构成的匹配向量的相似性进行角部特征点匹配,形成单视角下的局部三维线框模型。根据凸多面体顶面的关联点拼接多个局部模型,获得完整的三维线框模型。结果表明,处理的数据量仅为激光扫描仪的千分之一,角部控制点的位置精度平均相对误差为2.43%,棱边长度平均误差为2.7%,与激光扫描仪结果接近,能满足灵活、快速三维建模的需求。Due to the small pixel changes of weak-texture planes,it is quite difficult to use their own feature points for visual recognition and measurement.For convex polyhedra composed of multiple weak-texture planes,illumination from multiple angles is sequentially applied to create a light-dark difference between adjacent planes,so as to determine the face regions and edges of the convex polyhedron.The intersection points of adjacent edges are selected as corner feature points for modeling.Matching of corner feature points is performed based on the similarity of matching vectors formed by neighboring pixels of the corner feature points,resulting in a local three-dimensional wireframe model from a single viewpoint.By combining multiple local models using associated points on the top surface of the convex polyhedron,a complete three-dimensional wireframe model is obtained.The results indicate that the data scale for processing is only one-thousandth compared to that of a laser scanner.The average positional accuracy error for corner control points is 2.43%,while the average length error for edges is 2.7%.These results closely approximate those obtained by a laser scanner,meeting the demands for flexible and rapid three-dimensional modeling.

关 键 词：双目视觉 弱纹理 凸多面体 角部特征点 线框模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]