基于优化ICP算法的机器人双目相机点云配准

Point Cloud Registration of Robot Binocular Camera Based on Optimized ICP Algorithm

为了高效求解机器人在复杂工况下对物体的抓取位姿,在传统点云精配准基础上,基于KD-Tree优化ICP点云精配准。介绍机器人双目视觉生成视差图和深度图的算法,阐述点云配准的流程,论述以数据结构层面优化ICP算法的原理。采用双目相机作为视觉输入,重建场景点云,与物体三维模型精配准后,得到机器人末端执行器抓取位姿。结果表明:与优化ICP算法之前相比,该方法得到的物体目标点云在保证配准精度的同时能以更快的速度收敛至最优解。

To solve the grasping pose of the robot in complex working conditions efficiently, based on the traditional point cloud fine registrationthe, ICP registration is optimized on the account on KD-Tree. This paper introduces the algorithm of generating disparity map and depth map by robot binocular vision, expounds the process of point cloud registration, and discusses the principle of optimizing ICP algorithm on data structure level. The phase, used as the visual input to reconstruct the point cloud of the scene, is precisely matched with the three-dimensional model of the object so as to abtain the grasping pose of the robot end actuator. The results show that the object point cloud obtained by ICP registration based on KD-Tree can converge to the optimal solution at a faster speed while ensuring the registration accuracy.