基于改進(jìn)YOLOv8n的車(chē)輛及行人檢測(cè)

打開(kāi)文本圖片集

DOI：10.16652/j.issn.1004-373x.2025.17.006引用格式：，，.基于改進(jìn)YOLOv8n的車(chē)輛及行人檢測(cè)[J].現(xiàn)代電子技術(shù)，2025，48（17）：35-40.

關(guān)鍵詞：YOLOV8n；車(chē)輛行人檢測(cè)；小目標(biāo)檢測(cè)；損失函數(shù)；SPPF；C2f中圖分類(lèi)號(hào)：TN911.73-34；TP391.41 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號(hào)：1004-373X（2025）17-0035-06

Vehicleand pedestrian detectionbased on improved YOLOv8nalgorithm

ZHOU Jianxin，HAO Yingjie，HOU Zichuan （CollegeofElectricalEngineering，NorthChinaUniversityofScienceandTechnologyTangshanO63210，China）

Abstract：Inview of theobject oclusion，overlap，and small object missing in roadscene detection，avehicleand pedestrianobjectdetectionmodelbasedontheimprovedYOLOv8nalgorithmhasbeenproposed.Firstly，anew pyramidpooling layerstructurecaled SPPF-CREPisdesigned toreplace theoriginal network'spyramidpoling layerstructure SPF，nhancing themodel'seficiencyand performance during trainingand inference.Secondly，asmall object detection head （P2） is added to improvethenetwork'sdetectioncapabilityforsmallobjects.Then，theonlineconvolutionalre-parameterization（OREPA）is integratedintotheC2f moduletoenhancetheauracyrateandeficiencyofvehicleandpedestriandetectioninhigh-density environments.Finally，the WIoU- v2 isadopted asan alternative loss function to achieve higher localization accuracy. ExperimentalresultsontheKITTIvehicledetectiondatasetdemonstratesthat，incomparisonwiththeoriginalalgorithm，the accuracyrateof thedetectionoftheimprovedalgorithmisincreasedby3.6%，andthemeanaverageprecision（mAP）is enhanced by 4.2% .This experiment has proven its eficiency and superiority in the detection of vehicles and pedestrians.

Keywords：YOLOV8n;vehicleand pedestrian detection;small object detection;loss function;SPPF; C2f

0 引言

在道路目標(biāo)檢測(cè)任務(wù)中，傳統(tǒng)的目標(biāo)檢測(cè)特征泛化能力弱，對(duì)復(fù)雜場(chǎng)景的性能表現(xiàn)較差1，并且存在準(zhǔn)確率低和檢測(cè)速度緩慢等問(wèn)題。（剩余8220字）