Abstract: Aiming at the problem of inaccurate calculation or even unrecognition of fruit and vegetable poses due to the presence of occlusion in complex environments, a research method for estimating the poses of bell peppers based on Kinect camera is proposed to estimate the threedimensional poses of bell peppers in space by fusing the point cloud information of bell peppers from multiple viewpoints combined with the PCA algorithm. This method is based on fusing the point cloud information of bell peppers collected by Kinect camera from two viewpoints separated by 60°, and introducing the dual constraints of the corresponding point pairs of local normal vector angle and mean curvature threshold to improve the accuracy of the point cloud alignment during the ICP fine alignment process, in order to solve the technical problem of poor point cloud alignment and low overlap between the two viewpoints, which results in the large deviation of the point cloud attitude estimated by the PCA algorithm. The experimental results show that the mean value of the error of the algorithm in this paper is 5.15° in the case of no occlusion, the integrated mean error in the case of occlusion in one of the dualview perspectives is 5.67°, and the integrated mean error in the case of occlusion in both perspectives is 7.86°, which meets the technical requirements of the actual operation. This study is not only applicable to the spatial threedimensional pose recognition of bell peppers, but also provides technical references for the construction of other intelligent picking robot attitude recognition systems.

Key words: sweet pepper, machine vision, pose estimation, point cloud registration, PCA algorithm, color threshold

摘要： 针对在复杂环境中由于存在遮挡导致果蔬姿态计算不准甚至无法识别这一问题，提出一种基于Kinect相机的甜椒姿态估计研究方法，通过融合多视角下的甜椒点云信息，结合PCA算法估计出甜椒在空间中的三维姿态。基于Kinect相机的多视角RGB-D信息融合的甜椒姿态估计研究方法是将两间隔为60°的视角下Kinect相机采集甜椒的点云信息融合，并在ICP精配准过程中引入对应点对局部法向量夹角与平均曲率阈值双约束项提高点云配准精度，以解决两视角点云配准效果差、重叠度不高从而造成的由PCA算法估计点云姿态偏差大的技术问题。试验结果表明在无遮挡情况下，本文算法的误差均值为5.15°，在双视角中其中某一视角有遮挡情况下的综合均值误差为5.67°，双视角均有遮挡情况下综合平均误差为7.86°，满足实际作业时的技术要求。本研究不仅适用于甜椒的空间三维姿态识别，同时为其他智能采摘机器人姿态识别系统搭建提供技术参考。

关键词: 甜椒, 机器视觉, 姿态识别, 点云配准, PCA算法, 颜色阈值