【作者】 黄伟；

【导师】 江万寿；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2021， 博士

【摘要】 输电线路作为电力部门的关键设施,其安全运营对电力工业有着至关重要的影响。随着国民经济快速发展,电网规模逐渐扩大,输电线路分布愈加广泛,且跨越地形更为复杂,传统人工巡线方式无法满足日常巡检的需求。无人机作为一种新型的遥感数据获取平台,在摄影测量和电力行业等领域得到广泛应用。利用无人机来代替人工巡检方式,已成为电力部门日常巡检的重要手段。利用无人机影像对电力场景进行三维重建,有利于检修人员及时了解电力走廊三维场景真实情况,快速定位隐患位置。然而,虽然无人机能够提高巡检效率,但也给输电线路无人机影像三维重建带来新的挑战。具体表现在以下几个方面:1)在无人机电力巡检中,为了提高巡检效率,通常采用矩形或者S形飞行方式进行数据采集,这种长航带退化结构在无人机影像相机自检校时容易出现“碗状”效应;2)在电力场景三维重建中,电力线是输电线路的重要组成部分,同时也是电力维护人员重点关注的设施。基于Patch Match的逐像素可见影像选择密集匹配算法虽然能够重建出电力线的密集点云,但面临效率低下的问题,如何在保证电力线重建完整度的情况下来提高密集匹配的效率,也是输电线路三维重建急需解决的问题之一;3)电力杆塔结构较为特殊,主要有线状的钢材连接组成,在无人机影像中随着视角变换,电力杆塔的背景颜色变化较大,且电力杆塔不同部件之间互相遮挡严重,给电力杆塔线段匹配造成困难。针对上述问题,本文从三个方面展开针对性的讨论,具体内容如下:1.针对长航带结构中相机自检校容易出现“碗状”效应的问题,提出了一种GNSS（Global Navigation Satellite System）约束的长航带无人机影像自检校方法。首先研究相机畸变模型中经典的物理模型和最新的数学模型;然后在增量式Sf M（Structure from Motion）空三框架下,设计了一种联合无人机影像相机检校参数初始化和高精度差分GNSS位置信息辅助的相机自检校方法。利用两个电力走廊实验区域不同采集模式下的四组无人机影像,进行无控制约束以及单个控制点约束的相机自检校实验。结果表明,本文提出的相机自检校策略在无控制点约束时,可以有效缓解长航带结构自检校的“碗状”效应,减轻模型的弯曲程度,提高自检校空三的绝对精度;在单个控制点约束自检校时,水平和高程精度分别优于0.04 m和0.05 m。与当前主流开源和商业软件对比,本文算法能够得到相当或更高精度。2.针对Colmap中密集匹配效率低下的问题,本文在此基础上进行改进,提出了一种顾及电力线的快速Patch Match密集匹配算法。首先,提出一种新的随机红黑棋盘模式的传播方式,通过利用颜色最相似的邻域像素来传播平面参数,提高传播效率。同时为了让Colmap中逐像素可见影像选择的方法与随机红黑棋盘传播方式相结合,本研究改进了Colmap中可见概率推理的更新策略;其次,通过减少非必要的深度值和法向量匹配代价计算次数,来进一步提高密集匹配效率;最后,在GPU（Graphics Processing Unit）中实现了一种高效的深度图融合算法,采用基于重投影误差的加权函数进行深度图融合。通过三种不同飞行方式的无人机影像数据进行实验分析,表明本文算法能够在保障电力线完整度的情况下提高密集匹配的效率。同时,为了验证本文算法的精度,利用两组标准数据集进行实验,结果表明本文算法在₁F-score上均优于Colmap,且速度能提高4倍以上。3.针对电力杆塔线段匹配的问题,提出了一种基于共面约束的电力杆塔三维线段重建方法。首先从无人机影像所提取的电力杆塔二维线段交点中,分析出分布比较稳定的虚拟交点;其次,设计了一种共面约束匹配代价,对虚拟交点及对应线段进行匹配,并统计出电力杆塔的主要面元信息;最后,利用这些主要面元信息引导匹配剩余线段,并进行粗差剔除,最终实现电力杆塔的三维线段重建。通过与Line3D++进行实验对比,本文算法能匹配更多的电力杆塔三维线段,同时取得较好的精度,充分表明本文算法的有效性。更多还原

【Abstract】 As one of the key facilities in the power department,the safe operation of transmission lines has a vital impact on the power industry.With the rapid development of the national economy,the scale of the power grid has gradually expanded,the distribution of transmission lines has become more widespread with a more complex geographic environment.The traditional manual inspection cannot meet the needs.As a new type of remote sensing data acquisition platform,UAV（Unmanned Aerial Vehicle）has been widely used in the fields of photogrammetry and the power industry.Using UAVs instead of manual inspection has become an important means of daily inspection in the power department.The 3D reconstruction of the transmission line corridor from UAV images is helpful for maintainers to understand the real situation of the transmission line corridor in time,and quickly locate the hidden danger locations.However,although UAV can improve the inspection efficiency,it also brings new challenges to the 3D reconstruction of UAV images in the transmission line.They are embodied in the following aspects:1)To improve the UAV inspection efficiency,the rectangle or S-shaped flight trajectory is usually employed for data acquisition.The long corridor degradation structure would lead a“bowl effect”with the camera self-calibration of UAV images;2)In the 3D reconstruction of the power corridor,the power line is an important part of the transmission lines,and it is also one of the key facilities that maintainers pay more attention to.The dense matching algorithm with pixelwise visible image selection in Colmap can reconstruct the dense point clouds of the power line,but it also confronts the challenge of inefficiency.How to improve the efficiency of dense matching while ensuring the completeness of the power line reconstruction is also one of the urgent problems of the 3D reconstruction in transmission lines;3)The pylons have special structures and are mainly composed of wired steel connections.As the view changes in the UAV image,the background color of pylons changes greatly,and the different components of the pylons are severely occluded with each other,which makes it difficult to match the line segments of the pylon.In view of the above problems,this paper makes profoundly discussion from three aspects as follows:1.To solve the problem of the“bowl effect”with camera self-calibration in the long corridor structure,the self-calibration method of GNSS（Global Navigation Satellite System）constrained BA（bundle adjustment）for weakly structured long corridor UAV images is proposed.First,existing camera distortion models are grouped into two categories,i.e.,physical and mathematical models,and their mathematical formulas are exploited in detail.Second,within an incremental Sf M（Structure from Motion）framework,a camera self-calibration method is designed,which combines the strategies for initializing camera distortion parameters and fusing high-precision GNSS（Global Navigation Satellite System）observations.The former is achieved by using an iterative optimization algorithm that progressively optimizes camera parameters;the latter is implemented through inequality constrained BA.Finally,by using four UAV datasets collected from two sites with two data acquisition modes,the proposed algorithm is comprehensively analyzed and verified,and the experimental results demonstrate that the proposed method can dramatically alleviate the“bowl effect”of self-calibration for long corridor UAV images that have weak structure,and the horizontal and vertical accuracy reach 0.04 m and 0.05 m,respectively,when using one GCP（Ground Control Point）.In addition,compared with open-source and commercial software,the proposed method achieves competitive or better performance.2.To improve the efficiency of the dense matching method in Colmap,an accelerated 3D reconstruction method for 3D reconstruction of UAV images with fine-scale power lines is proposed.Firstly,an efficient random red-black checkerboard propagation mode is proposed,which utilizes the neighbor pixels with the most similar color to propagate plane parameters.To combine the pixelwise view selection strategy adopted in Colmap with the efficient random red-black checkerboard propagation,the updating schedule for inferring visible probability is improved;Secondly,strategies for decreasing the number of matching cost computations are proposed,which can reduce the unnecessary hypotheses to verify;Thirdly,an efficient GPU（Graphics Processing Unit）based depth map fusion method is proposed,which employs weight function based on the reprojection errors to fuse depth map.With three datasets of UAV images of different data acquisition modes,experiments indicate that the proposed method can keep the completeness of power line reconstruction with more efficiency compared to other Patch Match based methods.Two benchmark datasets are applied to verify the precision of the reconstructed point clouds with the proposed method.The comprehensive experiment results demonstrate that the proposed method can achieve a better₁F-score and 4 times faster than Colmap.3.To solve the problem of line segment matching with pylon,the 3D line reconstruction of the line segment with pylon based on coplanar constraint is proposed.Firstly,the intersection points of the 2D line segment of the pylon from UAV images are extracted,and the virtual intersection points with stable distribution are analyzed from those intersection points;Secondly,a coplanar constraint matching cost is designed to match the virtual intersection points and the corresponding line segments.Then the main plane information of the pylon is analyzed;Finally,the main plane information is adopted to assist in matching the remaining 2D line segments and remove the wrongly matched line segment to complete the 3D reconstruction of the line segment with pylon.更多还原