快速鲁棒的城市场景分段平面重建

发布时间：2018-05-10 05:53

本文选题：能量最小化 + 平面拟合　；参考：《自动化学报》2017年04期

【摘要】：对于基于图像的城市场景重建,由于光照变化、透视畸变、弱纹理区域等因素的影响,传统像素级与区域级的重建算法通常难以获得可靠的重建结果.为了解决此问题,本文提出一种快速、鲁棒的分段平面重建算法.根据城市场景结构特征与分段平面假设,本文算法首先利用基于连通域检测的空间平面拟合方法从初始空间点中抽取充分且可靠的候选空间平面,然后在MRF(Markov random field)能量最小化框架下将场景的完整结构推断问题转化为平面标记问题进行求解.由于候选平面集与融合灰度一致性度量、空间几何与可见性约束的能量模型的高可靠性,场景的完整结构因此可被有效地重建.实验结果表明,本文算法能较好地克服传统算法可靠性差、重建场景不完整等缺点,同时具有较高的计算效率.
[Abstract]:For image-based urban scene reconstruction, due to the influence of illumination variation, perspective distortion, weak texture region and other factors, it is difficult to obtain reliable reconstruction results by traditional pixel level and region level reconstruction algorithms. In order to solve this problem, a fast and robust segmented plane reconstruction algorithm is proposed. According to the structural features of urban scenes and the assumption of segmented plane, this algorithm firstly extracts sufficient and reliable candidate space planes from the initial spatial points by using the spatial plane fitting method based on connected domain detection. Then the complete structure inference problem of the scene is transformed into a plane marking problem under the framework of MRF(Markov random field energy minimization. Because of the high reliability of the energy model with candidate plane set and fusion gray level consistency and spatial geometry and visibility constraints, the complete structure of the scene can be reconstructed effectively. The experimental results show that the proposed algorithm can overcome the shortcomings of the traditional algorithm such as poor reliability and incomplete reconstruction scene, and has high computational efficiency.
【作者单位】：周口师范学院网络工程学院;中国科学院自动化研究所模式识别国家重点实验室;中国科学院大学;
【基金】：国家高技术研究发展计划(863计划)(2015AA124102) 国家自然科学基金(61273280,61333015,61472419) 河南省自然科学基金(162300410347) 河南省科技攻关项目(162102310589,162102210396) 河南省高校重点科研项目(17A520018,17A520019,15A520116,16B520034) 周口师范学院高层次人才科研启动基金(zknuc2015103)资助~~
【分类号】：TP391.41

【参考文献】