单目相机物体位姿估计方法研究

现有的机器视觉通常以边缘轮廓和角点作为特征,因此要求背景单一,对环境结构化依赖程度高。为了拓展机器人的应用范围,使其脱离结构化的环境,提出了一种基于SIFT特征点和PNP技术的单目相机估计目标物体位姿的方法。以BumbleBee双目相机为硬件基础,以C++为开发平台,结合了Eigen计算库、OpenCV图像处理库和Triclops库,开发了单目视觉位姿估计算法,实现在复杂背景下对表面纹理较为丰富的物体的位姿估计。利用试验对所提方法进行了验证,试验结果表明,该算法具有较高的估计精度,可以作为机器抓取的依据。

Study on Object Position and Pose Estimation Method of Monocular Camera

The traditional methods of machine vision are usually characterized by edge contour and corner points, so it requires a monotonous background and relies on the structured environment strongly. In order to expand the application scope of robot and be away from structured environment, a method based on SIFT (Scale-Invariant Feature Transform) feature points and PNP (Perspective-n-Point) technique for monocular camera estimation of object position and pose is proposed. Taking the BumbleBee binocular camera as the hardware basis, C++ as the development platform, combined with Eigen computing library, OpenCV image processing library and Triclops library, the monocular vision pose estimation algorithm is developed to realize the pose estimation of objects with rich texture under complex background. The proposed method is verified by experiments, and the experiments results shows that the method can estimate with high accuracy and can be used as a basis for robot grasp.