一种单摄像机视觉导引方法

Vision guiding method with monocular camera

本文提出了一种基于已知固定特征的单目摄像机视觉导引方法,建立了摄像机的运动和投影模型。设计了以正方形顶点为特征的参考平面目标,提出了复杂背景中全自动正方形顶点特征的图像提取新方法及标记特征点的方案。仿真试验表明,摄像机距离参考目标约10m,噪声偏差达到1.5像素时,位置RMS误差小于10cm,姿态RMS误差小于1.4°。因此,所提出的算法具有很强的抗噪声能力,能够满足实时性和实用性要求,为视觉导引的工程应用奠定了理论基础。

A monocular vision guiding method for estimating the position and orientation of the camera is proposed. The motion and projection model of the camera on-board a machine system was established. A new planar target with four black squares was designed to generate the control points with the 16 vertices of the four squares. The position and orientation of the camera can be calculated with the world coordinates of the 16 control points and their corresponding image coordinates. Simulation test result shows that the proposed vision-based state estimate is accurate to within 10cm in each axis of translation and 1.4 degrees in each axis of rotation when the distance between the camera and the reference target is about 10m and the Gaussian white noise level is 1.5 pixels. The proposed technique is robust to noise and reliable, it meets the demand of the real time measurement of the position and orientation for the machine system control. It will promote the engineering applications of vision guiding.