基于目标识别与定位技术的机械手研究  

Research on Robotic Arms Based on Target Recognition and Positioning Technology

作　　者：于佳弘 张骥祥[1] 张军[1] 

机构地区：[1]天津职业技术师范大学,电子工程学院,天津

出　　处：《人工智能与机器人研究》2023年第3期181-188,共8页Artificial Intelligence and Robotics Research

摘　　要：针对生活场景中使用机械手进行物体抓取的操作,本文提出了基于Yolov5算法的多目标检测系统,此算法相较于低版本的Yolo算法及R-CNN算法具有计算量小、准确率高的特点。市面中多采用单目相机赋予机械手二维视觉模块,本文在单目相机的基础上,相较于双目相机在获取三维数据时成本高、受环境影响时特征点无法进行匹配易产生误差的问题,结合结构光模块,生成单目结构光系统,更好地赋予机械手获取物体的深度信息的能力,最终得到物体的世界坐标值。通过实验现象表明,获得的平均精度均值为96.5%,定位精度相较于双目系统更为精准,最终能够较好地满足机械手完成对于物体在三维空间中定位与抓取的需要。This article proposes a multi-objective detection system based on the Yolov5 algorithm for object grasping using robotic arms in real-life scenarios. Compared to earlier versions of the Yolo algorithm and R-CNN algorithm, this algorithm has the characteristics of low computational complexity and high accuracy. Monocular cameras are often used in the market to endow the manipulator with a two-dimensional vision module. This paper, based on the monocular camera, compared with the binocular camera, which has a high cost in acquiring three-dimensional data, and the problem that the feature points cannot be matched when affected by the environment, is prone to errors. Combined with the structured light module, a monocular structured light system is generated to better endow the manipulator with the ability to obtain the depth information of the object, and finally obtain the world coordinate value of the object. Through experimental phenomena, it has been shown that the average accuracy obtained is 96.5%, and the positioning accuracy is more accurate than that of the binocular system. Ultimately, it can better meet the needs of the robotic arm for positioning and grasping objects in three-dimensional space.

关 键 词：Yolov5 目标识别 单目结构光 定位 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]