基于镜面折反射全景成像的三维重建方法研究  

Research on 3D reconstruction based on catadioptric panorama imaging

作　　者：郭伟青[1,2] 汤一平[1,2] 夏少杰[1,2] 韩国栋[1,2] 胡克钢 

机构地区：[1]浙江工业大学之江学院,杭州310024 [2]浙江工业大学信息工程学院,杭州310023

出　　处：《高技术通讯》2015年第10期878-885,共8页Chinese High Technology Letters

基　　金：国家自然科学基金(61070134)资助项目

摘　　要：针对基于图像的三维重建过程中成像系统获得的双视点或多视点图像不能一次性完整建模、图像匹配过程中计算复杂度高、标定繁琐的问题,设计了一种基于镜面折反射全景成像技术的单目多视角立体视觉系统及三维建模算法。该方法根据立体视觉传感器获取被测物体表面信息图像,通过阈值分割将图像分割为5个基本视图,并建立空间几何约束关系;使用离散的体素来表示被重构物体的几何信息,采用光一致性约束进行物体表面点的判断;通过轮廓体素遍历获取物体三维模型。该方法能一次性获得物体不同视角全方位图像信息,根据空间几何约束、纹理颜色约束及物体曲线约束,快速重构出物体的三维几何模型。该方法降低了匹配计算的复杂度和硬件使用成本,其有效性已在实际系统中得到了验证。In consideration of the problems of image-based 3 D reconstruction that the imaging system-acquired dual-view or multi-view images can not be completely modeled for one time during reconstruction, and the computational com- plexity is high in image matching wish the complicated calibration, a monocular multi-view stereo vision system and a 3D modeling method were designed based on the catadioptrie panoramic imaging technique. The method uses a stereo visual sensor to obtain the panorama image of an object, divides the image into five basic views by threshold segmentation, and constructs the constraints of space geometry ; Then, it uses discrete voxels to express the geomet- ric information of the object, and judges the surface points of the object according to the optical consistency princi- ple; Lastly, it uses the traversing method based on contour voxels to get the 3D modeling of the object. This method can get the omni-direetional image data from different perspectives for one time, and uses the constraints of space geometry, texture color and object curves to reconstruct the 3D model of the object. This method can reduce the computational complexity and the hardware cost, and its effectiveness was verified by a practical system.

关 键 词：镜面成像 全景视觉 三维建模 体素遍历 光一致性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]