摘 要:通过借鉴目标检测YOLO 系列算法中的优秀做法,提出基于YOLOv3 的行人检测方法。深入研究YOLOv3 模型,分析YOLOv3 算法的网络结构和检测流程,提出基于YOLOv3 的行人检测框架。在自制的行人数据库上实现YOLOv3 算法,根据实验结果发现并评估该方法在实际场景的行人检测中存在的问题,为改进YOLOv3 算法在实际场景中的行人检测提供一定的参考。
关键词:YOLO 算法;行人检测;目标框;激活函数
中图分类号:TP391.41 文献标识码:A 文章编号:2096-4706(2020)10-0081-04
Research on Pedestrian Detection Method Based on YOLOv3
ZHENG Xueyuan
(Ji’an College,Ji’an 343000,China)
Abstract:This paper proposes a pedestrian detection method based on YOLOv3 by referring to the excellent methods in the YOLO series of target detection algorithms. The paper deeply studies the model of YOLOv3,analyzes the network structure and detection process of YOLOv3 algorithm,and proposes a pedestrian detection framework based on YOLOv3. The YOLOv3 algorithm is implemented on the selfmade pedestrian database. According to the experimental results,we find and evaluate the existing problems of this method in the pedestrian detection in the actual scene,and provide some reference for improving the YOLOv3 algorithm in the actual scene pedestrian detection.
Keywords:YOLO algorithm;pedestrian detection;target frame;activation function
基金项目:2019 年江西省教育厅科学技术研究项目(GJJ191449)
参考文献:
[1] VIOLA P,JONES M J,SNOW D. Detecting PedestriansUsing Patterns of Motion and Appearance [C]//Proceedings Ninth IEEEInternational Conference on Computer Vision,Nice,France:IEEE,2003,63(2):734-741.
[2] DALAL N,TRIGGS B. Histograms of Oriented Gradientsfor Human Detection [C]//2005 IEEE Computer Society Conferenceon Computer Vision and Pattern Recognition (CVPR’05),SanDiego,CA,USA:IEEE,2005:886-893.
[3] FELZENSZWALB P,MCALLESTER D,RAMANAN D. ADiscriminatively Trained,Multiscale,Deformable Part Model [C]//2008IEEE Conference on Computer Vision and Pattern Recognition,Anchorage,AK,USA:IEEE,2008:1-8.
[4] 高宗,李少波,陈济楠,等. 基于YOLO 网络的行人检测方法 [J]. 计算机工程,2018,44(5):215-219+226.
[5] REDMON J,DIVVALA S,GIRSHICK R,et al. You OnlyLook Once:Unified,Real-Time Object Detection [J/OL].2016IEEEConference on Computer Vision and Pattern Recognition(CVPR),arXiv:1506.02640 [cs.CV].(2015-06-08).https://arxiv.org/abs/1506.02640.
作者简介:郑学远(1987—),男,汉族,江西吉安人,高级工程师,硕士研究生,研究方向:计算机视觉、深度学习、区块链技术等。