Abstract: To get the attitude of the Unmanned Air Vehicle(UAV) in the process of continuous flight, an attitude algorithm of Micro-unmanned Air Vehicle(MUAV) based on monocular vision is introduced in this paper, by capturing the sequence images with monocular vision of unmanned camera, obtaining the feature points information with using Scale Invariant Feature Transform(SIFT) algorithm, based on epipolar geometry and Random Sample Consensus(RANSAC) method, the attitude transform information and navigation information of MUAV are resolved. Experimental results show that in proper environment with good matching points state, the attitude transform error with monocular vision is less than 0.1 degree, and accumulated attitude error within 180 degree rotation range is less than 1 degree.

Key words: attitude, monocular vision, epipolar geometry, Scale Invariant Feature Transform(SIFT), Random Sample Consensus(RANSAC), error compensation

摘要： 针对无人机在连续飞行过程中的姿态求取问题，提出一种基于单目视觉的微型无人机姿态算法。基于无人机摄像机获得序列图像，利用图像尺度不变特性变换获取特征点信息，结合对极几何约束关系，运用随机采样一致性原理求解载体位姿变换信息，从而获得载体的导航信息。实验结果表明，通过单目序列图像获得的姿态角度变化精度优于0.1°，在180°旋转情况下的误差累加值小于1°。

关键词: 姿态, 单目视觉, 对极几何, 尺度不变特征变换, 随机采样一致性, 误差补偿

CLC Number: 

• TP301.6