【作者】 刘涛；

【导师】 柳长安；

【作者基本信息】 华北电力大学 ， 计算机应用技术， 2013， 硕士

【摘要】 架空电力线路巡检过程中,在保证设备安全性的前提下,飞行机器人需要灵活的对电塔上的重点巡检区域采集大量的巡检图像。这要求飞行机器人能够实时的获得自身与电塔在世界坐标系中的绝对位置,或者估计出自身与电塔的相对位置关系。木文考虑到定位系统的可靠性、易用性与经济成木等因素,研究了估计单目摄像机空间位姿的问题。首先,分析了基于场景三维点云模型的摄像机位姿估计方法和基于物体线框模型的摄像机位姿估计方法的基本原理。结合电力线巡检的应用背景,将其应用于电力线巡检场景中的摄像机位姿估计过程,讨论了二者的优缺点与适用范围。其次,对场景三维点云获取过程中的序列图像匹配方法进行了改进,使其更适用于有序的图像集合,提高了图像匹配的速度。然后,针对基于物体线框模型的位姿估计方法中的手动初始化过程,提出了一种融合红外图像与可见光图像的自动化初始位姿获取方法,并通过简化的实验验证了该方法的可行性。最后,通过分析基于物体线框模型的位姿估计方法的不足,设计并实现了一种室内环境下,基于多个红外相机的全局位姿估计系统。利用该系统在每-帧里为摄像机位姿估计过程提供一个粗略的初始估计值,在此初始位姿的指导下,跟踪物体的线框模型,获得更精确的摄像机位姿估计值。该方法通过降低基于物体线框模型的位姿估计方法中由于帧间累积误差对位姿估计结果造成的影响,提高了摄像机6自由度位姿估计的精度。实验证明了该方法的有效性与可靠性。更多还原

【Abstract】 In the process of overhead power line inspection, under the premise of ensuring the safety of equipments, a flying robot is required to collect a large number of images of the key areas of power pylons flexibly. This requires the flying robot can access the absolute position of the power tower and itself in the world coordinate system, or acquire the relative positional relationship between itself and the electrical towers in real time. Considering the reliability, ease of use and economic costs, an in-depth research on monocular camera pose estimation has been carried out.Firstly, the basic principle of point cloud based pose estimation and model based pose estimation has been analyzed. Combined with the task of power line inspection background, by applied those two methods to the process of camera pose estimation, the advantages, disadvantages and respective coverage has been discussed.Secondly, the sequence image matching method in the acquisition process of the scene’s point cloud has been improved, to make it more suitable for the orderly collection of images.Thirdly, a novel approach for the initialization procedure of a model-based tracking system has been proposed. Experimental results show that this proposed method has the same precision compared with the manual initialization result.Finally, by analyzing the deficiencies of the model based pose estimation method, a multiple infrared camera system has been designed and implemented which is suitable for global pose estimation in the indoor environment. The system could provide a rough initial guess of a camera’s pose every frame. A better camera pose could be obtained by tracking the wireframe model of the object under the guidance of the initial pose. And it could improve the pose estimation result by reducing the cumulative error in the model based pose estimation process. Experimental results show that this method has high validity and reliability.更多还原