Implementation of a Flexible and Lightweight Depth-Based Visual Servoing Solution for Feature Detection and Tracing of Large, Spatially-Varying Manufacturing Workpieces
Implementation of a Flexible and Lightweight Depth-Based Visual Servoing Solution for Feature Detection and Tracing of Large, Spatially-Varying Manufacturing Workpieces
复制标题
DOI：
10.3390/robotics11010025
复制
发表时间：
2022-02
期刊：
影响因子：
3.7
通讯作者：
Lee Clift;D. Tiwari;C. Scraggs;W. Hutabarat;L. Tinkler;Jonathan M. Aitken;Ashutosh Tiwari
中科院分区：
文献类型：
--
作者：
Lee Clift;D. Tiwari;C. Scraggs;W. Hutabarat;L. Tinkler;Jonathan M. Aitken;Ashutosh Tiwari
这项工作提出了一种新的解决方案，用于检测和跟踪大型制造工件的空间变化的边缘，使用消费级RGB深度相机，只有工件的部分视图，而没有先验知识。所提出的系统可以直观地检测和跟踪各种边缘，具有广泛的角度阵列，精度为15 mm或更小，而不需要任何先前的信息，设置或规划。结合物理实验的设置和更复杂的模拟实验进行。该系统的有效性是通过模拟和物理实验进行的尖锐和钝的边缘，以及典型的航空航天结构，由各种材料制成，尺寸从400毫米到600毫米。模拟结果表明，与人工噪声添加，该解决方案可以检测航空航天结构的精度为40毫米或更少，这取决于存在的噪声量，而物理航空航天启发的结构可以以5mm的一致精度被跟踪，而不管基本方向如何。与当前的工业解决方案相比，缺乏所需的规划和边缘检测的鲁棒性意味着它应该能够比当前标准更快更容易地完成任务，并且比当前使用的技术具有更低的财务和计算成本。
This work proposes a novel solution for detecting and tracing spatially varying edges of large manufacturing workpieces, using a consumer grade RGB depth camera, with only a partial view of the workpiece and without prior knowledge. The proposed system can visually detect and trace various edges, with a wide array of degrees, to an accuracy of 15 mm or less, without the need for any previous information, setup or planning. A combination of physical experiments on the setup and more complex simulated experiments were conducted. The effectiveness of the system is demonstrated via simulated and physical experiments carried out on both acute and obtuse edges, as well as typical aerospace structures, made from a variety of materials, with dimensions ranging from 400 mm to 600 mm. Simulated results show that, with artificial noise added, the solution presented can detect aerospace structures to an accuracy of 40 mm or less, depending on the amount of noise present, while physical aerospace inspired structures can be traced with a consistent accuracy of 5 mm regardless of the cardinal direction. Compared to current industrial solutions, the lack of required planning and robustness of edge detection means it should be able to complete tasks more quickly and easily than the current standard, with a lower financial and computational cost than the current techniques being used within.