随机工件的点云场景CAD模型的快速识别与定位

工业机器人精准抓取工件的前提是能够获取到目标工件的位置与位姿的信息。提出一种在三维点云场景对随机摆放目标工件的快速、精准的识别与定位的方法。在复杂点云场景中随机采样并提取关键点组成随机点对表征局部特征,再与由物体CAD模型建立的三维匹配模板进行配准,获得一系列候选位姿。利用表面相似点占比多次优化位姿,确定目标工件最佳位姿。最后利用虚拟相机投影二维视图到图像坐标系直观呈现结果。选择汽车座椅连接件作为验证实验的实验对象,实验结果表明:目标工件平移方向上拟合误差可以达到0.3 mm左右,旋转的拟合误差可以达到0.8°左右,具有较高的稳定与高效性能。

Rapid identification and positioning system of point cloud scene CAD model of random work-pieces

The premise of an industrial robot to accurately grasp work-pieces is to obtain information on the position and posture of the target work-pieces.A fast and accurate method for identifying and locating target work-pieces in a three-dimensional point cloud scene is proposed.Randomly sampled and extracted key points in a complex point cloud scene to form a random point pair to represent local features,and then registered with the three-dimensional matching template established by the object CAD model to obtain a series of candidate poses.Using multiple pose optimization based on the proportion of similar points on the surface to determine the optimal pose of the target work-pieces.Finally,using a virtual camera to project a two-dimensional view on the image plane to visualize the results.Select the car seat connector as the experimental object of the verification experiment.The experimental results show that the fitting error of the target work-pieces in the translation direction can reach 0.3mm,and the fitting error of rotation can reach 0.8°,which has high stability and high performance.