MRF模型的车载全景视觉位姿估计最优化方法研究

项目名称： MRF模型的车载全景视觉位姿估计最优化方法研究

项目编号： No.41501504

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 天文学、地球科学

项目作者： 张正鹏

作者单位： 辽宁工程技术大学

项目金额： 20万元

中文摘要： 车载移动测量系统的位姿数据精度直接影响点云(或影像)数据融合的精度，是实现街景量测与三维重建的几何基础。目前位姿数据的获取主要采用GPS/INS组合式定位导航方式。但在城市建筑物密集区域，当GPS信号遮挡区范围较大时，仅依靠INS会使得车辆定位精度随时间迅速降低。目前国内外已开展了基于视觉的位姿估计方法，但位姿估计的参数精度不高，且在长序列大场景范围内估计得到的位姿点容易产生漂移。项目从提高大场景全景影像位姿估计的精度出发，采用马尔科夫模型研究全景序列影像的位姿估计问题。首先从全景模型入手，研究基于立方体投影的全景多视几何模型；利用几何约束条件，研究全景点特征匹配与建筑物垂直线灭点提取方法；针对全景影像的位姿估计问题，研究马尔科夫随机场域下的角元素和线元素局部能量优化模型以及最优求解方法；最后融合漂移检测算法的“增量式”平差策略进一步优化位姿数据。

中文关键词： 全景影像；车载测量系统；位姿估计；影像匹配；光束法平差

英文摘要： The fusion precision of point cloud or image data are influenced by the pose data precision of Vehicle-borne mobile measurement system, and the geometric basis to realize the street view measurement and 3d reconstruction. The pose data are acquired mainly by the GPS/INS modular navigation and positioning method. But the dense regions of buildings in the city, when GPS signals keep out zone is bigger,Only rely on INS makes the vehicle positioning precision rapidly decrease over time. The pose estimation method at home and abroad has carried out based on visual, but the parameter precision of position estimation is not high, and estimated pose some prone to drift within the scope of long sequence and big scene. This project will improve pose estimation accuracy problem from big scene panoramic image sequence using the Markoff model. First of all, starting from the panoramic model, geometric model of the multi view projection cubes panorama based on geometric constraint conditions will be studyed; and research on feature matching and vanishing point extraction method of building vertical line; Study of local energy optimization model of angle elements and line elements and the optimal solution method in Markov random field to solve pose estimation problem for the panoramic image; For the panoramic adjustment problem, using fusion drift detection algorithm and incremental adjustment strategy, The last experiment, and quantitative analysis results of the bundle adjustment are given.

英文关键词： panoramic image;Vehicle-borne measurement system;pose estimation;image matching;bundle adjustment