光学技术, 2018, 44 (1): 63, 网络出版: 2018-02-01  

遗传算法结合自适应阈值约束的ICP算法

Optimized ICP method combining genetic algorithm with adaptive threshold constraints
作者单位
1 江南大学 轻工过程先进控制教育部重点实验室, 江苏 无锡 214122
2 无锡信捷电气股份有限公司, 江苏 无锡 214072
摘要
根据传统ICP算法存在的缺点, 提出了一种遗传算法结合自适应阈值约束优化的ICP位姿估计方法。采用统计学滤波和区域增长分割算法对原始点云进行预处理, 去除离群点并得到各混乱工件的点云集; 针对ICP易陷于局部最优的问题, 利用遗传优化算法对点云进行粗匹配, 得到目标点集相对于参考点云的初始位姿; 针对迭代速度较慢的缺点, 提出了一种自适应阈值约束法, 利用点对距离约束和法向量夹角约束去除局部大变形点, 在保证实时性的同时提高了位姿估计的精度。实验表明, 该方法能够在84.5ms内定位一个工件, 位姿估计误差达0.39mm, 满足实时性和抓取精度要求, 能够为工业机器人随机箱体抓取提供理论依据与指导。
Abstract
In order to solve the problem of iterative closest point algorithm, an improved pose estimation method based on genetic algorithm combined with adaptive threshold constraints is proposed. Point clouds without noise are obtained by adopting an optimized preprocessing approach combining a statistical outlier removal filter and region growing segmentation algorithm. Considering the disadvantage of local optimization for ICP, genetic algorithm is applied which calculated target clusters’ probable pose to reference point cloud. Adaptive threshold constraints are proposed in consideration of slow speed of iteration. Local large deformation is removed by selecting eligible points according to Euclidean threshold and normal vector angle threshold, which ensured computation speed and improved the precision of pose estimation at the same time. Experiments show that the method can get a pose of a target in 84.5ms, and the pose estimation error is 0.39mm, which meets the real-time and accuracy demands and provides theoretical basis and guidance for the grasp of random bin picking with industrial robot.

石爱军, 白瑞林, 田青华, 李杜. 遗传算法结合自适应阈值约束的ICP算法[J]. 光学技术, 2018, 44(1): 63. SHI Aijun, BAI Ruilin, TIAN Qinghua, LI Du. Optimized ICP method combining genetic algorithm with adaptive threshold constraints[J]. Optical Technique, 2018, 44(1): 63.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!