基于机器视觉的非特定物体的智能抓取系统的研究

针对传统机械臂局限于按既定流程对固定位姿的特定物体进行机械化抓取,设计了一种基于机器视觉的非特定物体的智能抓取系统;系统通过特定的卷积神经网络对深度相机采集到的图像进行目标定位,并在图像上预测出一个该目标的可靠抓取位置,系统进一步将抓取位置信息反馈给机械臂,机械臂根据该信息完成对目标物体的抓取操作;系统基于机器人操作系统,硬件之间通过机器人操作系统的话题机制传递必要信息;最终经多次实验结果表明,通过改进的快速搜索随机树运动规划算法,桌面型机械臂能够根据神经网络模型反馈的的标记位置对不同位姿的非特定物体进行实时有效的抓取,在一定程度上提高了机械臂的自主能力,弥补了传统机械臂的不足.

Research on Intelligent Grasping System for Non-specific Objects Based on Machine Vision

Aiming at traditional manipulators that are limited to mechanized grasping of specific objects at fixed positions and postures according to a set process, an intelligent grasping system for non-specific objects based on machine vision is designed. The system locates the object on the image captured by the depth camera through the convolutional neural network, and predicts a reliable grasping position of the object on this image. The system transmits the information of grasping position to the manipulator, and the manipulator grasps the target object. The system is based on the robot operating system, and the necessary information is transmitted between the hardware through the topic of the robot operating system. The final experimental results show that through the improved rapid-exploration random tree algorithm, the manipulator can effectively capture non-specific objects in different positions and postures in real time, which improves the autonomy of the manipulator and makes up for the shortcomings of traditional manipulators.