改进YOLOv8的环视车位检测算法研究
Research on the improved YOLOv8 algorithm for the detection of surrounding parking spaces
杨飞帆 1李军 1王耀弘2
作者信息
- 1. 重庆交通大学 机电与车辆工程学院,重庆 400074
- 2. 重庆市计量质量检测研究院,重庆 401121
- 折叠
摘要
为在泊车过程中更快速、准确、全面地检测停车位,提出一种针对全景环视系统AVM(around view monitor)图像的改进YOLOv8环视车位检测算法.采用部分卷积思想设计出Faster Block结构,分别替换C3模块和C2f模块中的Bottleneck结构,选取改进后更轻量化的C3-Faster模块替换YOLOv8中原有的C2f模块;在YOLOv8主干网络末端添加CoordAtt(coordinate attention)注意力层,提升模型特征提取能力;引入EIoU作为损失函数提升模型检测精度.该算法在自建的AVM环视车位检测数据集上模型精度、召回率、mAP@0.5与mAP@0.5:0.95分别达到87.6％、87.6％、93.1％、62.0％;相较于原YOLOv8n模型提升1.5％、1.4％、1.6％、1％;参数量、浮点运算量及模型大小分别下降约27％、24％、25％.
Abstract
To locate parking spaces more quickly and accurately,this paper proposes an improved YOLOv8 algorithm for around view monitor (AVM) image.First,a Faster Block structure is designed based on partial convolution,replacing the Bottleneck structure of C3 module and C2F module respectively.A lighter C3-Faster module is selected to replace the YOLOv8 C2F module in Central Plain.Then,Coordinate Attention layer is added at the end of YOLOv8 backbone network to enhance the model feature extraction capability.Finally,Eiou is introduced as a loss function to improve the detection accuracy of the model.The model accuracy,recall rate,mAP@0.5 and mAP@0.5:0.95 reach 87.6％,87.6％,93.1％ and 62.0％ respectively,up by 1.5％,1.4％,1.6％ and 1％ compared with the data of the original YOLOv8n model.The parameters,floating-point computation and model size are down by 27％,24％ and 25％ respectively.
关键词
车位检测/YOLOv8/C3-Faster/CoordAtt注意力机制/EIoU损失Key words
parking detection/YOLOv8/C3-Faster/coord attattention mechanism/EIoU loss引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据