双目立体视觉在人形机器人三维重建中的应用

摘要

A three⁃dimensional(3D)reconstruction scheme based on binocular vision is proposed. Two VX6000 LifeCam cameras are installed in the NAO humanoid robot's face to collect the left and right images of the target. In the process of 3D re⁃construction,the camera is calibrated by utilizing the tessellated corner point positioning,the image is corrected to remove its distortion,and then the edge features of the image are extracted by Canny edge detector for stereo matching. Finally,the 2D image is triangulated by Delaunay triangulation,the image texture is mapped to the three⁃dimensional space,and then 3D target image is rendered to obtain the final 3D image. The experimental results show that the proposed scheme can accurately measure the depth of the scene,and can be used in the 3D reconstruction of the humanoid robot scene.%提出一种基于双目视觉的机器人三维（3D）重建方案，将两个LifeCam VX6000摄像机安装在NAO人形机器人脸部，分别采集目标的左右图像。在3D重建过程中，利用棋盘格的角点定位对摄像机进行标定，对图像进行校正去除畸变，利用Canny检测算子来提取图像的边缘特征，进行立体匹配，利用Delaunay三角剖分算法对二维图像进行三角剖分，并将图像纹理映射到三维空间，同时对立体目标图像进行渲染，获得最终的3D重建图像。实验结果表明，该方案能够精确的进行深度测量，实现场景的3D重建，能够很好的用于人形机器人应用场景。