基于Faster-RCNN的投篮机器人目标球识别算法  

Target Ball Recognition Algorithm of Shooting Robot Based on Faster-RCNN

作　　者：李帅 陈万米[1] 高洋 

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10期93-94,97,共3页Industrial Control Computer

摘　　要：目标识别是一个涉及到多个学科领域的综合性问题,而投篮机器人最主要的识别对象是各种球类。针对传统识别算法对于多变的环境下对目标球的识别效果较差的问题,提出了基于Faster-RCNN的目标球识别算法,采用ZF-Net网络模型作为特征提取器。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 comprehensive problem involving multiple disciplines,and the main recognition object of the shooting robot is various balls.Aiming at the problem that the traditional recognition algorithm has poor recognition effect on the target ball in the changing environment,a target ball recognition algorithm based on Faster-RCNN is proposed,and the ZF-Net network model is used as the feature extractor.Experiments verify that the recognition effect of the algorithm in different scenarios is better than the traditional contour color recognition algorithm,and the robustness and accuracy are improved.

关 键 词：目标识别 投篮机器人 Faster-RCNN 

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]