LSFP下的机器人视觉定位技术研究

Research on robot vision positioning technology under LSFP

针对当前点云的定位算法中,多平面分割效果差、误匹配以及定位精度不够高等问题,提出LSFP的定位思想和C-S分割算法,对目标物体进行定位。在深度相机场景下,利用点云处理算法初步获取目标的位姿,经手眼标定后,进行初步引导;再使用3D扫描仪进一步定位目标物体。结果表明:(1)C-S分割算法对多平面分割效果良好,能更准确地去除多余平面,减少点云数量。(2)在垂直于Z轴且Z轴坐标值为171 mm和-10 mm的平面上,仅PrimeSense的定位距离的均方根误差为7.3 mm和8.2 mm,LSFP下的定位距离的均方根误差为2.6 mm和3.2 mm。(3)LSFP定位思想能扩大场景定位的范围,提高物体的定位精度,具有良好的准确性。

At the problems of poor multi-plane segmentation,mis-matching and insufficient localization accuracy in the current localization algorithm of point clouds,the localization idea of LSFP and C-S segmentation algorithm are proposed to localize the target object.In the depth camera scene,the point cloud processing algorithm is used to initially obtain the bit position of the target,which is initially guided by hand-eye calibration;then the 3D scanner is used to further localize the target object.The results show that:(1)the C-S segmentation algorithm works well for multi-planar segmentation and can more accurately remove redundant planes and reduce the number of point clouds.(2)On planes perpendicular to the Z-axis and with Z-axis coordinate values of 171 mm and-10 mm,the root mean square error of the localization distance is 7.3 mm and 8.2 mm for PrimeSense only,and 2.6 mm and 3.2 mm for LSFP.(3)The LSFP localization idea can expand the range of scene localization and improve the object localization accuracy with good accuracy.