【作者】 童丸丸；

【导师】 潘翔；

【作者基本信息】 浙江大学 ， 信息与通信工程， 2010， 硕士

【摘要】 无人飞行器(Unmanned Aerial Vehicle, UAV)作为智能移动机器人之一因在航空摄影、灾情调查、紧急救援、国土资源勘察等方面的出色表现,以及军事上的特殊作用,倍受航空大国和军事强国的青睐。在UAV的各项关键技术中,视觉导航技术能重建UAV飞行环境,用于引导UAV在复杂环境下安全自主飞行,是近年来研究的热点。本文利用视觉传感器,瞄准UAV视觉导航技术发展方向,以双目立体视觉和基于单目视觉的运动恢复结构(SFM)为基础,侧重研究具有高精度的单目合成视觉基本原理,开发鲁棒的合成视觉算法,构建完整的软件处理系统。基于运动视差分解思想,提出了一种基于SURF特征的单目合成视觉三维重建算法。首先,通过基于SURF特征的运动补偿消除关于参考平面的视差部分；然后,创建拉普拉斯图像金字塔降低光照变化影响,利用残留视差算法在金字塔各层级上估计结构参数并逐级求精；最后,在给定摄像机参数和位置信息情况下完成三维重建。通过对模拟场景重建结果的定量分析和对室外光照变化场景重建结果的定性分析,证明该算法比双目立体视觉和SFM具有更高的精度,且具有较好的鲁棒性。更多还原

【Abstract】 Unmanned Aerial Vehicle (UAV) is one of automatic intelligence robots, due to the outstanding performance on aerial photography, disaster investigation, emergency and natural resource exploration, as well as the special role of the military, it attracts more attentions in military and air power. As one of the UAV key technologies, visual navigation was studied a lot in recent years, which could rebuild UAV flight environment for secure and autonomous flight in complex environments.Using vision sensor and aiming the development direction of UAV visual navigation, while stereo vision and structure from motion (SFM) are investigated, we focus on the research with the basic principle of high precision monocular synthetic vision, exploiting the robust algorithm and establishing the complete software system.On base of motion parallax decomposition, a synthetic vision three-dimensional (3D) reconstruction algorithm based on SURF (Speed up Robust Features) is proposed. First, remove the part of planar plane parallax by SURF characterization motion compensation. Then, the image Laplace pyramid is created to reduce the lighting variation, and the structure parameters are estimated on each pyramid level with residual parallax algorithm and refinement. At last, the 3D reconstruction is completed with the parameters given camera matrix and location information. With the quantitative analysis of the simulation scene and the qualitative analysis of the outdoor scene, the approach with higher precision and better robustness are obtained in comparison with stereo vision and SFM.更多还原