重庆理工大学学报2025，Vol.39Issue(11)：125-131,7.DOI:10.3969/j.issn.1674-8425(z).2025.06.015
基于YOLOv8复杂环境下车辆行人检测算法研究
Research on vehicle and pedestrian detection algorithms in complex environments based on YOLOv8
摘要
Abstract
To address the insufficient feature fusion and low detection accuracy of the traditional YOLOv8 algorithm in detecting vehicles and pedestrians in complex traffic environment,a vehicle and pedestrian detection algorithm based on the improved YOLOv8 model is proposed.First,the improved CCFF cross scale fusion module is introduced in the neck part,and the small target detection branch is added simultaneously to enhance the fusion ability of the model for the multi-scale features of the target and improve the detection accuracy of the small target.Then,the upsample module is replaced by the dysampling upsampling module.Through dynamic upsampling,the feature information is better preserved,the loss of details reduced,and the adaptability of the model in complex environments improved.Finally,dyhead detection head is used to realize the unification of scale perception,space perception and task perception,further improving the ability of the model to adapt to complex environments and enhancing the accuracy of target detection.Results show the improved model increases the mAP@50 by 6％and reduces the number of parameters by 8.4％respectively,enhancing both detection accuracy and flexibility.关键词
车辆行人检测/多尺度特征融合/动态上采样/Dyhead检测头Key words
vehicle and pedestrian detection/multi-scale feature fusion/dynamic upsampling/dyhead detection head分类
信息技术与安全科学引用本文复制引用
任金霞,王金荣,吴吉林,高东华,蔡联广..基于YOLOv8复杂环境下车辆行人检测算法研究[J].重庆理工大学学报,2025,39(11):125-131,7.基金项目
国家自然科学基金项目(51665018) （51665018）
江西省重点实验室基金项目(2024SSY03161) （2024SSY03161）
