合肥工业大学校徽

合肥工业大学学报自科版

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

主管：中华人民共和国教育部

主办：合肥工业大学

ISSN 1003-5060, CN 34-1083/N

面向智能变胞车的改进 YOLOv5 楼梯目标识别算法

Improved YOLOv5 staircase target recognition algorithm applied to intelligent metamorphic vehicles

期刊信息

合肥工业大学（自然科学版），2024年7月，第47卷第7期：879-886，899

DOI: 10.3969/j.issn.1003-5060.2024.07.003

作者信息

刘俊，张成，阮小栋

(合肥工业大学汽车与交通工程学院，安徽合肥230009)

摘要和关键词

摘要: 针对智能变胞车在室内楼梯环境下自主攀爬过程中的楼梯识别问题,文章提出一种基于改进YOLOv5的楼梯识别算法。为满足算法模型的实时性要求,利用轻量级网络EfficientNetV2替换YOLOv5算法的主干网络;使用GSConv模块和VoV-GSCSP模块替换原颈部网络中的Conv模块和CSP模块,在增强目标特征响应的基础上进一步减少计算成本;为弥补算法模型简化带来的精度损失,在颈部网络上添加坐标注意力机制,通过强化目标关注以提升在复杂场景下的目标识别效果;最后将改进的算法模型应用于嵌入式平台进行实验检测。实验结果表明:改进后的算法模型平均检测精度为91.99%,模型大小仅为3.1MB,相较于其他目标检测算法具有明显的优越性。文章所提算法能够有效地对楼梯进行实时、准确的检测识别,为后续变胞车自主越障奠定了一定的理论基础。

关键词: 智能变胞车；楼梯目标检测；YOLOv5算法；网络轻量化；注意力机制

Authors

LIU Jun, ZHANG Cheng, RUAN Xiaodong

(School of Automobile and Traffic Engineering, Hefei University of Technology, Hefei 230009, China)

Abstract and Keywords

Abstract: A staircase recognition algorithm based on improved YOLOv5 is proposed for staircase target recognition during autonomous climbing of intelligent metamorphic vehicles in indoor staircase environment. In order to meet the real-time requirements of the model, the lightweight network EfficientNetV2 is used to replace the backbone network of the YOLOv5 algorithm. Then, the GSConv module and the VoV-GSCSP module are used to replace the Conv module and the CSP module in the original neck network to further reduce the computational cost based on the enhanced target feature response. To compensate for the loss of accuracy due to model simplification, a coordinate attention mechanism is added to the neck network to enhance target recognition in complex scenes by strengthening target attention. Finally, the improved model is applied to the embedded platform, and the experimental results show that the average detection accuracy of the improved model is 91.99% and the model size is only 3.1 MB, which has obvious superiority compared with other target detection algorithms. This algorithm can effectively detect stairs in real time and accurately, which lays a theoretical foundation for the subsequent process of autonomous obstacle crossing of the metamorphic vehicle.

Keywords: intelligent metamorphic vehicle; staircase target detection; YOLOv5 algorithm; network lightweighting; attention mechanism

基金信息

国家自然科学基金资助项目（51875148）；安徽省重点研发计划资助项目（202104a05020040）

前一篇后一篇

返回期刊目录