基于改进YOLOv8的自动驾驶场景目标检测算法  

Object Detection Algorithm for Autonomous Driving Scenes Based on Improved YOLOv8

作　　者：杨磊 陈艳菲[1,2] 李海鸣 石教兴 安培 YANG Lei;CHEN Yanfei;LI Haiming;SHI Jiaoxing;AN Pei(Key Laboratory of Optical Information and Pattern Recognition in Hubei Province,Wuhan Institute of Technology,Wuhan 430205,China;School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205,China;School of Electronic Information and Communications,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]武汉工程大学光学信息与模式识别湖北省重点实验室,武汉430205 [2]武汉工程大学电气信息学院,武汉430205 [3]华中科技大学电子信息与通信学院,武汉430074

出　　处：《计算机工程与应用》2025年第1期131-141,共11页Computer Engineering and Applications

基　　金：武汉工程大学研究生教育创新基金(CX2023547)。

摘　　要：针对自动驾驶场景遮挡目标和小目标检测困难问题,提出了FAN-YOLOv8n自动驾驶检测算法。设计了特征感受野融合模块(EFFVM),增强模型主干部分对局部特征的提取,提高模型对遮挡目标的检测能力;在模型头部增加了更浅特征层P2的检测头,提高模型对于小目标的检测效果;在模型颈部设计了特征指导模块(FGM)来融合浅层和深层的特征信息,使得两层之间能够更好地进行特征交互,让模型更关注细粒特征。提出了特征层融合模块(FLFM),融合多尺度特征层并进行特征增强,使模型能够自适应不同尺度目标的检测。实验结果表明,在SODA10M数据集和部分BDD100K数据集上,改进模型的mAP0.5对比原始YOLOv8n模型提升了7个百分点和6.5个百分点,适用于实际自动驾驶检测任务。A FAN-YOLOv8n autonomous driving detection algorithm is proposed to address the difficulties in detecting occluded and small targets in autonomous driving scenarios.Firstly,an enhanced feature field of view module(EFFVM)is designed to enhance the extraction of local features in the backbone of the model and improve its ability to detect occluded targets.Secondly,a shallower feature layer P2 detection head has been added to the model head to improve the detection performance of the model for small targets.Then,a feature guidance module(FGM)is adopted at the neck of the model to fuse shallow and deep feature information,enabling better feature interaction between the two layers and making the model more focused on fine-grained features.Finally,a feature layer fusion module(FLFM)is proposed to fuse multiscale feature layers and perform feature enhancement,enabling the model to adaptively detect targets of different scales.The experimental results show that on the SODA10M dataset and some BDD100K dataset,the improved model’s mAP0.5 improves by 7 percentage points and 6.5 percentage points compared to the original YOLOv8n model,making it suitable for practical autonomous driving detection tasks.

关 键 词：自动驾驶 YOLOv8n 小目标 遮挡目标 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]