摘要
设计并实现了一种基于工件模型的单目立体视觉方法,可以获得立体工件堆中顶层工件的空间分布。将立体工件的识别转化为对工件特征面体的识别,通过识别定位特征面体定位工件位姿,依据特征面体对工件的代表性和空间工件位姿聚集度评价识别工件真实性,设置选择阈值,从而得到顶层工件分布。为描述特征面体对工件代表性,以特征面体与立体工件在图像平面投影的可见轮廓线之比和特征面体识别可靠性来进行评价。同时,利用同一工件可能会由多个特征面体识别得到的特点,提出了一种以空间范围内相似位姿工件数量和姿态重合度来评价空间工件位姿聚集程度的模型。
A monocular stereo vision solution is designed to get the spatial distribution of the top workpieces of workpiece piles. The recognition object is transformed from the solid workpiece to the workpiece characteristic plane and the pose of workpieee is determined by the pose of the characteristic plane. According to the characteristic plane' s representation of workpiece and the aggregation of spatial poses, the existing reality of workpieces axe sort. By setting threshold, we can get the poses of the top workpieces. The characteristic plane' s representation of workpiece is evaluated by the ratio between the visible contour line of the characteristic plane and the three-dimensional workpiece in the projection plane, as well as the characteristic plane' s recognition reliability. The degree of workpiece spatial discretization is evaluated with the number and the attitude deviation of the workpieee in a specific space.
出处
《机械设计与研究》
CSCD
北大核心
2017年第4期41-44,49,共5页
Machine Design And Research
关键词
单目立体视觉
几何模型
分拣
机器视觉
姿态重合度
特征面体
monocular stereo vision
geometric model
bin picking
machine vision
pose coincidence
characteristic plane