当前位置>主页 > 期刊在线 > 计算机技术 >

计算机技术23年6期

基于边缘计算的行人检测算法研究
张超,王亮
(西安工程大学 电子信息学院,陕西 西安 710048)

摘  要:文章基于多个边缘计算设备,构建轻量高效的行人检测算法。首先,基于原始 YOLOv5 模型,替换为MobileNetv3 作为特征提取网络,减少算法参数量和计算量;其次采用 ReLu6 激活函数,加速网络收敛,减少模型复杂度;最后利用 K-means 聚类算法仅选取 3 个先验框分别匹配到 3 个不同尺度检测头中。实验结果表明:改进后的模型参数量和计算量于仅为原始模型的 51.6% 和 39.8%,在 Jetson AGX Xavier 设备上采用 TensorRT 进行推理达到了 66 FPS 的实时速度,在RK3399 设备上达到了 2.1 FPS。


关键词:边缘计算;行人检测;YOLOv5 算法



DOI:10.19850/j.cnki.2096-4706.2023.06.021


中图分类号:TP391.4                                       文献标识码:A                                  文章编号:2096-4706(2023)06-0081-04


Research on Pedestrian Detection Algorithm Based on Edge Computing

ZHANG Chao, WANG Liang

(School of Electronics and Information, Xi’an Polytechnic University, Xi’an 710048, China)

Abstract: This paper constructed a lightweight and effective pedestrian detection algorithm based on multiple Edge Computing devices. Firstly, based on the original YOLOv5 model, MobileNetv3 is replaced as the feature extraction network to reduce the number of algorithm parameters and computation. Secondly, ReLu6 activation function is used to accelerate network convergence and reduce model complexity. Finally, K-means clustering algorithm is used to select only three prior frames and match them to three detection heads of different scales. Experimental results show that the number of parameters and computation of the improved model are only 51.6% and 39.8% of that of the original model, and the real-time speed of 66 FPS is achieved on Jetson AGX Xavier and 2.1 FPS on RK3399.

Keywords: Edge Computing; pedestrian detection; YOLOv5 algorithm


参考文献:

[1] 郭振堂,祝永新,田犁 . 基于多模态数据的车辆行人检测[J]. 激光杂志,2021,42(12):52-58.

[2] HAN H,ZHOU M C,SHANG X W,et al. KISS + for Rapid and Accurate Pedestrian Re-Identification [J].IEEE Transactions on Intelligent Transportation Systems,2020,22(1):394-403.

[3] CHENG J,WANG P S,LI G,et al. Recent Advances in Efficient Computation of Deep Convolutional Neural Networks [J]. Frontiers of Information Technology & Electronic Engineering,2018,19(1):64-77.

[4] 曹巧慧,葛万成 . 基于俯仰角修正的 HOG 特征快速行人检测算法 [J]. 现代电子技术,2014,37(21):62-65.

[5] 陈辉东,丁小燕,刘艳霞 . 基于深度学习的目标检测算法综述 [J]. 北京联合大学学报,2021,35(3):39-46.

[6] 王小燕,谢文昊,杨艺芳,等 . 基于深度学习的垃圾分类检测方法 [J]. 现代电子技术,2021,44(21):110-113.

[7] REDMON J,FARHADI A. YOLO9000:Better,Faster, Stronger [J/OL].arXiv:1612.08242 [cs.CV].(2016-12-25).https://arxiv.org/abs/1612.08242.

[8] BOCHKOVSKIY A,WANG C-Y,LIAO H-Y M. YOLOv4: Optimal Speed and Accuracy of Object Detection [J/OL].arXiv: 2004.10934 [cs.CV].(2020-04-23).https://arxiv.org/abs/2004.10934. 

[9] REDMON J,FARHADI A. YOLOv3:An incremental improvement [J/OL].arXiv:1804.02767 [cs.CV].(2018-04-08).https://arxiv.org/abs/1804.02767.

[10] 刘俊明,孟卫华 . 基于深度学习的单阶段目标检测算法研究综述 [J]. 航空兵器,2020,27(3):44-53.

[11] HOWARD A,SANDLER M,CHU G,et al. Searching for MobileNetV3 [J/OL].arXiv:1905.02244 [cs.CV].(2019-05-06).https://arxiv.org/abs/1905.02244v2.

[12] 栾奕,刘昌华 . 基于 TPU 和 FPGA 的深度学习边缘计算平台的设计与实现 [J]. 计算机工程与科学,2021,43(6):976-983.

[13] LIU S,QI L,QIN H F,et al. Path Aggregation Network for Instance Segmentation [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Salt Lake:IEEE,2018:8759-8768.


作者简介:张超(1996—),男,汉族,陕西渭南人,硕士研究生在读,研究方向:计算机视觉与模式识别;王亮(1995—),男,汉族,陕西西安人,硕士研究生在读,研究方向:计算机视觉与模式识别。