| 摘 要: 根据目前常见目标识别算法检测较为复杂、识别时间较长并容易出现错检漏检等方面的不足，利用目前机器视觉技术，提出一种适用智能移动小车系统的目标识别算法YOLOv3(一种深度学习目标检测方法)，并搭建智能移动小车的目标识别仿真系统。其目的是研究目标识别算法在ROS(机器人操作系统)中结合智能小车运动控制功能进行正确实现，改善目标检测算法的漏检率、错误率等。分析目标识别算法在ROS系统中的精确度，使用数据集进行检测实验。实验结果表明，运用YOLOv3深度学习目标识别算法在提高识别目标速度的同时有较高的检测精度，当目标丢失在视野中时利用ROS系统控制移动机器人重新使目标快速识别，降低了识别目标的时间，提高了移动机器人运动识别的效率。 |
| 关键词: ROS系统；深度学习；YOLOv3识别算法；运动控制 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| Research on Target Recognition of Intelligent Mobile Car based on Robot Operating System |
|
TIAN Yu
|
(School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China)
893570276@qq.com
|
| Abstract: Currently, common target recognition algorithms have shortcomings of complex detection, long recognition time and being prone to error detection and missed detection. In order to solve these problems, this paper proposes a target recognition algorithm YOLOv3 (A deep learning target detection method) that is suitable for intelligent mobile car system by using the current machine vision technology. Meanwhile, target recognition simulation system of the intelligent mobile car is built in order to study the correct implementation of target recognition algorithm in ROS (Robot Operating System) combined with intelligent car motion control function, thus reducing miss detection rate and error rate of target detection algorithm. Accuracy of target recognition algorithm in ROS system is analyzed and data set is used for detection experiment. Experimental results show that the proposed YOLOv3 deep learning target recognition algorithm can improve target recognition speed and detection accuracy. When a target is lost in the field of vision, ROS system is used to control the mobile robot to recognize the target quickly again, which reduces target recognition time and improves the efficiency of mobile robot motion recognition. |
| Keywords: ROS system; deep learning; YOLOv3 recognition algorithm; motion control |
