基于机器人视觉引导的柔性振动盘上料系统研究

Research on a Robot Vision-Guided Flexible Vibratory Feeder System for Low-Texture Workpieces

摘要: 本文针对工业“柔性振动盘 + 机器人”上料系统中低纹理工件识别与抓取效率的挑战，设计并实现了一套基于机器视觉引导的自动化分拣系统。系统整合储料仓、柔性振动盘、工业智能相机及KUKA三轴机器人，通过TCP/IP通信实现多模块协同。鉴于低纹理工件表面特征稀疏的特点，提出以基于形状的模板匹配算法作为核心识别方案，并结合图像预处理、金字塔特征提取及多级相似度度量，有效克服了传统特征点匹配对纹理依赖性强、鲁棒性差的问题。为进一步提升实时性，本文引入贪心算法提前终止策略和图像金字塔分层搜索策略，对模板匹配的搜索过程进行优化，将单次检测时间从320 ms显著缩短至68 ms，同时识别准确率由89.2%提升至100%。在此基础上，利用九点标定建立高精度手眼映射关系，确保机器人抓取位姿精度。实验结果表明，该系统能够在复杂工业场景下实现低纹理工件的高效、稳定分拣，为柔性制造场景下的自动化上料提供了可行的技术路径。

Abstract: To address the challenges of low-texture workpiece recognition and efficient grasping in industrial “flexible vibratory feeder + robot” feeding systems, this paper designs and implements an automated sorting system guided by machine vision. The system integrates a storage bin, a flexible vibratory feeder, an industrial smart camera, and a KUKA six-axis robot, and coordinates these modules via TCP/IP communication. Considering the sparse surface features of low-texture workpieces, a shape-based template matching algorithm is adopted as the core recognition method, combined with image preprocessing, pyramid-based feature extraction and multi-level similarity metrics, which effectively overcomes the high texture dependency and poor robustness of traditional feature-based methods. To further improve real-time performance, a greedy early-termination strategy and an image pyramid multi-scale search strategy are introduced to optimize the search process, reducing the single-image detection time from 320 ms to 68 ms while increasing the recognition accuracy from 89.2% to 100%. In addition, a high-precision hand–eye calibration model is established using a nine-point planar calibration, ensuring accurate pose estimation for robot grasping. Experimental results demonstrate that the proposed system can achieve efficient and stable sorting of low-texture workpieces in complex industrial environments, providing a feasible technical solution for flexible automatic feeding in intelligent manufacturing.