基于改进YOLOv8的车辆与行人检测算法  

Vehicle and Pedestrian Detection Algorithm Based on Improved YOLOv8

作　　者：孙君峰 张赵良 刘云平[1] 张涛[1] 张诗云 SUN Junfeng;ZHANG Zhaoliang;LIU Yunping;ZHANG Tao;ZHANG Shiyun(School of Automation,Nanjing University of Information Science&Technology,Nanjing 210000,China;School of Transportation and Vehicle Engineering,Wuxi University,Wuxi 214105,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210000 [2]无锡学院交通与车辆工程学院,江苏无锡214105

出　　处：《测控技术》2025年第3期9-17,共9页Measurement & Control Technology

基　　金：“太湖之光”科技攻关(基础研究)基金(K20221050);无锡学院科研启动项目(550221034)。

摘　　要：针对当前主流目标检测算法存在目标对象和车辆行人误检与漏检的问题,基于YOLOv8提出了一种改进的目标检测算法。首先,采用多点距离交并比(Multi-Point Distance Intersection over Union,MPDI-oU)边界框回归损失函数替代原完全交并比(Complete Intersection over Union,CIoU)损失函数,有效解决了当预测边界框与地面实况边界框宽高比相同时传统CIoU函数失效的问题。然后,通过嵌入大可分离内核注意力(Large Separable Kernel Attention,LSKA)机制增强算法的多尺度特征提取能力。最后,融合SCConv模块,在降低模型计算复杂度的同时提升检测精度。实验结果表明:改进后的算法相比原YOLOv8算法,准确率提升了4.07%,召回率提升了约2.95%,且检测速率达到85 f/s。Aiming at the problems of misdetection and omission of target objects,vehicles,and pedestrians existing in the current mainstream target detection algorithm,an improved target detection algorithm based on YOLOv8 is proposed.Firstly,the multipoint distance intersection over union(MPDIoU)bounding box regression loss function is adopted to replace the original complete intersection over union(CIoU)loss function,effectively solving the problem that the traditional CIoU loss function will fails when the predicted bounding box has the same aspect ratio as the ground truth bounding box.Then,the multi-scale feature extraction capability of the algorithm is enhanced by embedding large separable kernel attention(LSKA)mechanism.Finally,the SCConv module is integrated to improve the target detection accuracy while reducing the computatational complexity of the model.The emperimental results show that compared with the original YOLOv8 algorithm,the improved algorithm has increased the precision by 4.07%,the recall by about 2.95%,and the detection rate reaches 85 f/s.

关 键 词：目标检测 YOLOv8 注意力机制 MPDIoU 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]