Application of Improved Lightweight YOLOv5 Algorithm in Pedestrian Detection

doi:10.11871/jfdc.issn.2096-742X.2023.06.015

Abstract

Abstract:

[Objective] In this paper, we propose an improved YOLOv5 algorithm to address the problems of the high computational complexity of pedestrian detection algorithms, low detection accuracy, and slow detection speed, which can be better applied to pedestrian detection. [Methods] Firstly, the vanilla convolution in the YOLOv5 backbone network is replaced by the depthwise separable convolution, which reduces the number of calculations and parameters while improving detection accuracy. Then, channel attention and spatial attention are incorporated into the feature fusion part of the backbone network, which can force our network to focus on the location and channel information of pedestrians in the image. Finally, the EIOU loss function is used to optimize the proposed model, and the K-means++ clustering algorithm is used to generate priori boxes. [Results] The results show our proposed model can achieve a detection accuracy of 89%, which is 7.6% higher than the original backbone, and the detection speed reaches 106 frames per second when using the INRIA pedestrian detection dataset. [Conclusions] Our proposed method significantly improves the speed and accuracy of pedestrian detection, has also small parameters and is easier to detect and deploy in real-time.

Key words: pedestrian detection, deep learning, YOLOv5, deep separable convolution, attention mechanism

WANG Ziyuan, WANG Guozhong. Application of Improved Lightweight YOLOv5 Algorithm in Pedestrian Detection[J]. Frontiers of Data and Computing, 2023, 5(6): 161-172, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2023.06.015.

Figures/Tables 21

Table 1

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Fig.9

Fig.10

Fig.11

Fig.12

Table 2

Table 3

Fig.13

Table 4

Table 5

Fig.14

Fig.15

Table 6

References 23

[1]	DALAL N, TRIGGS B. Histograms of oriented gradients for human detection[C]// 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR'05), Ieee, 2005, 1: 886-893.
[2]	DOLLÁR P, TU Z, PERONA P, et al. Integral channel features[J]. Proceedings of the British Machine Conference, 2009, 91: 1-11.
[3]	LIU W, ANGUELOV D, ERHAN D, et al. Ssd: Single shot multibox detector[C]// European conference on computer vision, Springer, Cham, 2016: 21-37.
[4]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 779-788.
[5]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2014: 580-587.
[6]	ZHAO Y, SHI Y, WANG Z. The Improved YOLOV5 Algorithm and Its Application in Small Target Detection[C]// International Conference on Intelligent Robotics and Applications, Springer, Cham, 2022: 679-688.
[7]	WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C]// Proceedings of the European conference on computer vision (ECCV), 2018: 3-19.
[8]	ZHANG Y F, REN W, ZHANG Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157. doi: 10.1016/j.neucom.2022.07.042
[9]	VASSILVITSKII S, ARTHUR D. k-means++: The advantages of careful seeding[C]// Proceedings of the eighteenth annual ACM-SIAM symposium on Discrete algorithms, 2006: 1027-1035.
[10]	KIRILLOV A, GIRSHICK R, HE K, et al. Panoptic feature pyramid networks[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019: 6399-6408.
[11]	REZATOFIGHI H, TSOI N, GWAK J Y, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019: 658-666.
[12]	CHOLLET F. Xception: Deep learning with depthwise separable convolutions[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2017: 1251-1258.
[13]	ZHU X, CHENG D, ZHANG Z, et al. An empirical study of spatial attention mechanisms in deep networks[C]// Proceedings of the IEEE/CVF international conference on computer vision, 2019: 6688-6697.
[14]	SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-cam: Visual explanations from deep networks via gradient-based localization[C]// Proceedings of the IEEE international conference on computer vision, 2017: 618-626.
[15]	YU J, JIANG Y, WANG Z, et al. Unitbox: An advanced object detection network[C]// Proceedings of the 24th ACM international conference on Multimedia, 2016: 516-520.
[16]	AGARWAL N, GOEL S, ZHANG C. Acceleration via fractal learning rate schedules[C]// International Conference on Machine Learning, PMLR, 2021: 87-99.
[17]	TANG S, GOTO S. Histogram of template for human detection[C]// 2010 IEEE International Conference on Acoustics, Speech and Signal Processing, IEEE, 2010: 2186-2189.
[18]	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 & Machine Intelligence, 2017, 39(06): 1137-1149.
[19]	LI J, LIANG X, SHEN S M, et al. Scale-aware fast R-CNN for pedestrian detection[J]. IEEE transactions on Multimedia, 2017, 20(4): 985-996.
[20]	GIRSHICK R. Fast r-cnn[C]// Proceedings of the IEEE international conference on computer vision, 2015: 1440-1448.
[21]	WANG X, XIAO T, JIANG Y, et al. Repulsion loss: Detecting pedestrians in a crowd[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7774-7783.
[22]	LIU W, LIAO S, HU W. Efficient Single-Stage Pedestrian Detector by Asymptotic Localization Fitting and Multi-Scale Context Encoding[J]. IEEE Transactions on Image Processing, 2019, 29(99): 1413-1425. doi: 10.1109/TIP.83
[23]	LIU W, HASAN I, LIAO S. Center and Scale Prediction: Anchor-free Approach for Pedestrian and Face Detection[J]. arXiv preprint arXiv:1904.02948, 2019.

Algorithms	Input size	Test dataset	Speed/(frame s-1)	mAP(0.5)/%
Fast RCNN	600×1000	VOC2007	0.5	70.2
Faster RCNN	600×1000	VOC2007	5	73.5
SSD	512×512	VOC2007	18	76.6
YOLOv1	448×448	VOC2007	46	62.8
YOLOv2	544×544	MS COCO	40	43.6
YOLOv3	608×608	MS COCO	24	58.7
YOLOv4	608×608	MS COCO	58	66.9
YOLOv5s	608×608	MS COCO	69	57.8

环境名称	相关配置
操作系统	Ubuntu18.04
中央处理器（CPU）	AMD Ryzen 7 5800H 3.20 GHz
运行内存RAM/GB	16
GPU	NVIDIA GeForce RTX 3060
编程环境	Python 3.7
深度学习框架	PyTorch
CUDA	11.3
CUDNN	8.2.1

Type	Positive	Negative
True	TP	FN
False	FP	TN

Model	Para(M)	FLOPs(B)	MB
YOLOv5s	7.2	16.8	14.8
YOLOv5s-D	3.8	7.3	7.8
YOLOv5s-DA	3.8	7.5	8.0

Model	Precision/%	Recall/%	AP(0.5)/%	FPS
Faster RCNN	76.4	70.8	73.4	16.2
SAF-RCNN	78.2	72.8	76.2	18.3
RepLoss	82.0	73.9	79.1	10.7
SSD	86.3	76.0	77.8	60.1
ALFNet	87.5	70.6	78.3	46.0
CSP	89.0	75.8	79.5	58.4
YOLOv5s	88.8	80.2	81.6	83.2
YOLOv5s-D	87.6	76.7	80.7	90.7
YOLOv5s-DA	89.2	75.2	87.3	100.3
YOLOv5s-DAE	91.5	72.1	89.2	106.7