摘要
楼梯区域作为一种典型的环境目标,无论是可以自主爬楼梯的机器人系统,还是可以提醒视力障碍者注意障碍物的软件系统,都需要有检测识别楼梯的功能。为了帮助他们在周围环境中导航,设计了一种基于YOLOv5s的楼梯图像检测算法。首先采用Labelme对楼梯数据集进行图片标注,对标注文件进行格式转化;其次搭建YOLOv5s网络模型训练环境,修改预训练模型配置文件,然后启动模型权重数据迁移训练,并输出最优模型参数;最后加载测试数据集对训练最优模型算法进行效果测试,算法均能识别出上下楼梯图像,并通过与其他目标检测算法对比测试,其具有更高的识别准确率。结果表明,该算法检测平均精度达到80.3%,泛化能力强,将该算法用于上下楼梯区域检测方法是可行的,可以为机器人对楼梯的自动检测识别提供一些参考,也可以为视障者提供有效帮助,市场应用前景良好。
Stairway area is a typical environmental target.Whether it is a robot system that can climb stairs autonomously or a software system that can alert the visually impaired to obstacles,it needs to have the function of detecting and recognizing stairs.To help them navigate their surroundings,a stair case image detection algorithm based on YOLOv5s was designed.Firstly,Labelme was used to annotate the stair data set,and the format of the an-notated file was transformed.Secondly,the YOLOv5s network model training environment was built,the pre-trained model configuration file was modified,the model weight data migration training was started,and the optimal model parameters were output.Finally,the test data set was loaded to test the effect iveness of the trained optimal model algorithm.The algorithms can recognize the up and down stairs image,and through the comparison test with other target detection algorithms,it has a higher recognition accuracy.The results showed that the mean accuracy of the algorithm reached 80.3%,and the generalization ability was strong.It is feasible to apply the algorithm to the detec-tion method of stairway up and down stairs,which can provide some references for the automatic detection and recognition of stairs by robots,and can also provide effective help for the visually impaired.The market application prospect is good.
作者
高瑞
雷文礼
GAO Rui;LEI Weni(School of Physics and Electronic Information,Yan’an University,Yan’an 716000,China)
出处
《延安大学学报（自然科学版）》
2024年第1期77-81,共5页
Journal of Yan'an University：Natural Science Edition
