基于SSD-MobileNet的投篮机器人目标识别算法  

Target Recognition Algorithm of Basket-Shooting Robot Based on SSD-MobileNet

作　　者：高洋 陈万米[1] 林城 Gao Yang

机构地区：[1]上海大学机电工程与自动化学院,上海200444

出　　处：《工业控制计算机》2021年第6期51-53,共3页Industrial Control Computer

摘　　要：目标识别是机器人领域中的研究热点,是机器人与外界交互的起点。在投篮机器人竞赛中,针对球类的识别,传统的机器视觉方法鲁棒性及准确性较差。提出了基于SSD-MobileNet的目标球识别算法,兼具了SSD识别算法的精确性以及MobileNet网络的轻量性,同时对SSD算法中的目标识别先验框进行了改进。通过实验验证了该算法在不同场景下的识别效果均优于传统的轮廓颜色识别算法,鲁棒性及准确性得到提升。Target recognition is a research hotspot in the field of robot,and it is the starting point of robot interaction with the outside world.In the basket-shooting robot competition,the traditional machine vision method for ball recognition has poor robustness and accuracy.In this paper,a ball recognition algorithm based on SSD MobileNet is proposed,which has both the accuracy of SSD recognition algorithm and the lightness of MobileNet network.At the same time,this paper improves the prior box of SSD algorithm.The experimental results show that the recognition effect of the algorithm in different scenes is better than the traditional contour and color recognition algorithm,and the robustness and accuracy are improved.

关 键 词：目标识别 SSD-MobileNet 卷积神经网络 投篮机器人 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242[自动化与计算机技术—计算机科学与技术] G841[文化科学—体育训练]