基于改进YOLOv8交通情景下检测算法研究
Research on Improved YOLOv8 Detection Algorithm for Traffic Scenarios
苏圣强1
作者信息
- 1. 西安石油大学计算机学院,陕西 西安 710000
- 折叠
摘要
近年来,随着新能源汽车的发展趋势日益火热,对辅助驾驶、自动驾驶等技术要求也越加严格,但是在现实的驾驶情景中,道路环境复杂程度较高,特别针对遮挡的现象,现有的算法检测精度不高,且车载电脑的算力有限.针对这些问题,文章设计了三点改进方案.改进后的模型在BDD100k数据集下计算量只有6.1GFLOPs,为原模型的75％,同时Map50提升了 8.4％.研究结果为后续辅助驾驶的目标识别和移动端的部署提供了参考和依据.
Abstract
In recent years,with the increasing popularity of new energy vehicles,the require-ments for technologies such as assisted driving and autonomous driving have become more strin-gent.However,in real driving scenarios,the complexity of road environments is high,especially in the case of occlusion.The detection accuracy of existing algorithms is not high,and the com-puting power of on-board computers is limited.In response to these issues,this paper proposes three improvement schemes.The improved model has a computational cost of only 6.1 GFLOPs under the BDD100k dataset,which is 75％of the original model,while Map50 has increased by 8.4％.The research results provide reference and basis for subsequent target recognition of as-sisted driving and deployment on mobile devices.
关键词
目标检测/YOLOv8/深度学习/Soft-NMSKey words
Object Detection/YOLOv8/Deep Learning/Soft-NMS引用本文复制引用
出版年
2024
NETL
NSTL
万方数据