针对行人目标数量密集、目标尺度小和目标周围背景光照强弱不一而导致的检测精度低的问题,提出一种基于特征融合的轻量化行人检测算法。以TinyYOLOv4为基础框架,首先,搭建新的主干特征提取网络(CSPDarknet53-S),在原主干网络的基础上加入新的特征提取模块(REM)来增强网络提取行人特征的能力。其次,改进特征融合结构,在主干网络提取高低层特征图后,先是在主干网络与特征融合网络间加入特征融合模块(RM-block)来增大感受野;然后引入浅层特征信息保留更多小目标特征,形成新的特征融合网络(IFFM)。最后,通过YOLO Head对融合来的特征图进行处理获得输出结果。实验结果表明,提出的算法在行人数据集(PASCAL VOC2007和VOC2012的person数据)上取得了较高的检测精度以及较好的检测效果。
Abstract
A lightweight pedestrian detection algorithm based on feature fusion is proposed to solve the problem of low detection accuracy caused by dense pedestrian targets, small target scales, and varying background illumination around the target. Firstly, build a new backbone feature extraction network (CSPDarknet53-S), and add a new feature extraction module (REM) to the original backbone network to enhance the networks ability to extract pedestrian features. Secondly, improve the feature fusion structure. After extracting high-low feature maps from the backbone network, add a feature fusion module (RM block) between the backbone network and the feature fusion network to increase the receptive field. And then introduce shallow feature information to retain more small target features to form a new feature fusion network (IFFM). Finally, the fused feature map is processed through YOLO Head to obtain the output results. The above steps are based on the basic framework of TinyYOLOv4. Experimental results show that the proposed algorithm achieves higher detection accuracy and better detection results on pedestrian data sets (PASCAL VOC2007 and VOC2012 person data).
关键词
目标检测 /
特征融合 /
浅层特征 /
TinyYOLOv4算法 /
注意力机制
Key words
target detection /
feature fusion /
shallow characteristics /
TinyYOLOv4 algorithm /
attention mechanism
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
［1］王洪斌, 于菲, 李一骏, 等. 分块特征匹配与局部差分结合的运动目标检测［J］. 计量学报, 2015, 36(4): 352-355.
WANG H B, YU F, LI Y J, et al. Detection of moving object by combining block features matching and local differential ［J］. Acta Metrologica Sinica, 2015, 36(4): 352-355.
［2］LIENHART R, MAYDT J. An extended set of Haar-like features for rapid object detection ［C］ //IEEE International Conference on Image Processing. New York, USA, 2002.
［3］DALAL N, TRIGGS B. Histograms of oriented gradients for human detection ［C］ //IEEE Conference on Computer Vision and Pattern Recognition. New York, USA, 2005.
［4］WU B, NEVATIA R. Detection of multiple, partially occluded humans in a single image by bayesian combination of edgelet part detectors ［C］ //IEEE International Conference on Computer Vision. Beijing, China, 2005.
［5］程淑红, 高许, 周斌. 基于多特征提取和SVM参数优化的车型识别［J］. 计量学报, 2018, 39(3): 348-352.
CHEN S H, GAO X, ZHOU B. Vehicle recognition based on multi-feature extraction and SVM parameter optimization ［J］. Acta Metrologica Sinica, 2018, 39(3): 348-352.
［6］VIOLA P, JONES M. Rapid object detection using a boosted cascade of simple features ［C］ //IEEE Conference on Computer Vision and Pattern Recognition. Kauai, HI, USA, 2001.
［7］KAZEMI F M, SAMADI S, POORREZA H R, et al. Vehicle recognition using Curvelet transform and SVM ［C］ //4th International Conference on Information Technology. Las Vegas NV, USA, 2007.
［8］GIRSHICK R, DONAHUE J, DARRELLl T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation ［C］ //IEEE Conference on Computer Vision and Pattern Rec-ognition. New York, USA, 2014.
［9］GIRSHICK R. Fast R-CNN ［C］ //IEEE International Conference on Computer Vision. Santiago, Chile, 2015.
［10］REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real time object detection with region proposal networks ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
［11］张立国, 蒋轶轩, 田广军. 基于多尺度融合方法的无人机对地车辆目标检测算法研究［J］. 计量学报, 2021, 42(11): 1436-1442.
ZHANG L G, JIANG Y X, TIAN G J. Research on Unmanned Aerial Vehicle to Ground Vehicle Target Detection Algorithm Based on Multiscale Fusion Method ［J］. Acta Metrologica Sinica, 2021, 42(11): 1436-1442.
［12］LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector ［C］ //14th European Conference on Computer Vision. Amsterdam, Netherlands, 2016.
［13］REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, reatime object detection ［C］ //IEEE Conference on Computer Vision and Pattern Recognition. Seattle, WA, 2016.
［14］REDMON J, FARHADI A. Yolov3: An incremental improvement ［J/OL］. https://arxiv.org/abs/1804.02767. 2018.
［15］BOCHKOVSKIY A, WANG C Y, LIAO H M. Yolov4: Optimal speed and accuracy of object detection ［J/OL］. https://arxiv.org/abs/2004.10934. 2004.
［16］HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37 (9): 1904-1916.
［17］WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module ［C］ //15th European Conference on Computer Vision. Munich, German, 2018.
［18］WANG C Y, LIAO H M, WU Y H, et al. CSPNet: A new backbone that can enhance learning capability of CNN ［C］ //IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Seattle, WA, 2020.
［19］LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection ［C］ //IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu, USA, 2017.
［20］ZHENG Z, WANG P, REN D, et al. Enhancing geometric factors in model learning and inference for object detection and instance segmentation［J］. arXiv preprint arXiv: 2005. 03572, 2020.
［21］谭芳, 穆平安,马忠雪. 基于YOLOv3检测和特征点匹配的多目标跟踪算法［J］. 计量学报, 2021, 42(2): 157-162.
TAN F, MU P A, MA Z X. Multi-target Tracking Algorithm Based on YOLOv3 Detection and Feature Point Matching［J］. Acta Metrologica Sinica, 2021, 42(2): 157-162.
［22］齐向明, 董旭. 改进Yolov7-tiny的钢材表面缺陷检测算法［J］. 计算机工程与应用, 2023, 59(12): 176-183.
QI X M, DONG X. Improved Yolov7-tiny algorithm for steel surface defect detection［J］. Computer Engineering and Applications, 2023, 59(12): 176-183.
［23］WANG C Y, BOCHKOVSKIY A, LIAO H M. Scaled-YOLOv4: Scaling Cross Stage Partial Network ［C］ //IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Virtual, 2021.
基金
国家重点研发项目子课题(2020YFB1711001);河北省科学技术研究与发展计划科技支撑计划项目(20310302D)
PDF(924 KB)
