基于人工地标的双足机器人视觉自定位

针对家庭或者办公室等室内环境下双足机器人的自定位问题,提出一种通过对世界平面的单应矩阵进行分解来实现自定位的方法。首先设计一种新型AR-mark人工地标,利用颜色信息分割感兴趣区域提取出AR-mark的特征点,然后对特征点与空间3D点进行匹配。通过直接线性变化法计算出世界平面的单应矩阵后,利用世界平面的单应分解算法获得摄像机坐标系与AR-mark人工地标坐标系间的位置关系,最终实现双足机器人的局部精确定位。实验结果表明,AR-mark人工地标易于识别,信息内容丰富,同时利用世界平面单应矩阵分解算法定位精度高,能够很好地辅助双足机器人完成室内自定位任务。

Visual self-localization of biped robot based on artificial landmark

To solve self-localization problem of a biped robot in indoor environment such as home or office, a novel method which adopts homography decomposition from world plane was proposed. First, a new kind of artificial landmark, AR-mark, is designed, and feature points are mainly extracted by the color-based segmentation. Along with the feature points matching with the space 3D points in world space, homography from world plane is calculated by direct linear variation (DLT) method. Finally, the position and pose parameters expressed relative to the landmark frame of the biped robot are obtained using homography decomposition to realize robot self-localization. The experimental results show that the designed AR-mark artificial landmark can be easily recognized with rich environment information. Moreover, the proposed method has high positioning accuracy to assist the biped robot to accomplish the indoor self-localization task.