High Performance Yolov4 with Different Optimizers and Backbones to Determine Distance
DOI:
https://doi.org/10.25134/ilkom.v20i1.533Keywords:
object detection, Yolov4, CSPResNeXt50, deep learningAbstract
Monitoring the distance between individuals in public spaces remains relevant for crowd management, operational safety, and risk mitigation in public service. Manual inspections are difficult to carry out consistently and in real time, thus requiring an automated machine vision-based system. This research proposes a safe distance monitoring pipeline that detects humans, calculates the proximity between individuals. Human detection is performed with YOLOv4 and the CSPResNeXt50 backbone, then the center of the bounding box is used as a position representation. The distance between centroids is calculated using Euclidean Distance and calibrated based on camera parameters, thus that it can be mapped to distance units. To obtain a stable training configuration, two optimizers (Adam and SGD) are compared at two learning rates (1E-2 and 1E-3). Test results show that Adam with a learning rate of 1E-3 provides the best performance, achieving 96% object detection accuracy. The distance calculation module achieves 90% accuracy in determining the actual and predicted distances in optimal camera installation scenarios, with an average processing speed of 60 fps. These findings demonstrate a fast and practical approach for camera-based public space safety. Additionally, the backbone comparison shows that CSPResNeXt50 is slightly superior to DarkNet50, making it suitable for lightweight implementation on edge devices in real-time.
Downloads
References
[1] I. Ahmed, M. Ahmad, J. J. P. C. Rodrigues, G. Jeon, and S. Din, “A deep learning-based social distance monitoring framework for COVID-19,” Sustain. Cities Soc., vol. 65, no. 102571, pp. 1–12, 2021, doi: 10.1016/j.scs.2020.102571. Available: https://doi.org/10.1016/j.scs.2020.102571
[2] L. Gogulamudi, S. Mudigonda, Y. P. Renduchintala, B. Vemula, and G. H. Palnati, “A Social Distance Estimation and Crowd Monitoring System for Surveillance Cameras,” Res. Adv. Innov. Comput. Commun. Inf. Technol. ICRAIC2IT, vol. 2796, no. 1, pp. 1–21, 2022, doi: 10.1063/5.0148967
[3] M. Seker, A. Männistö, A. Iosifidis, and J. Raitoharju, “Automatic social distance estimation for photographic studies : Performance evaluation , test benchmark , and algorithm ✩,” Mach. Learn. with Appl., vol. 10, no. October, pp. 1–14, 2022, doi: 10.1016/j.mlwa.2022.100427. Available: https://doi.org/10.1016/j.mlwa.2022.100427
[4] L. Deng, Q. Zhou, S. Wang, J. M. Górriz, and Y. Zhang, “Deep learning in crowd counting: A survey,” CAAI Trans. Intell. Technol., vol. 9, no. 5, pp. 1043–1077, 2024, doi: 10.1049/cit2.12241
[5] Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “YOLOX: Exceeding YOLO Series in 2021,” pp. 1–7, 2021, Available: http://arxiv.org/abs/2107.08430
[6] C.-Y. Wang, A. Bochkovskiy, and H.-Y. M. Liao, “YOLOv7,” Cvpr, pp. 7464–7475, 2023.
[7] Y. Zhao et al., “DETRs Beat YOLOs on Real-time Object Detection,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., pp. 16965–16974, 2024, doi: 10.1109/CVPR52733.2024.01605
[8] M. Aghaei, M. Bustreo, Y. Wang, G. Bailo, P. Morerio, and A. Del Bue, “Single image human proxemics estimation for visual social distancing,” Proc. - 2021 IEEE Winter Conf. Appl. Comput. Vision, WACV 2021, pp. 2784–2794, 2021, doi: 10.1109/WACV48630.2021.00283
[9] V. Sharma and R. N. Mir, “A comprehensive and systematic look up into deep learning based object detection techniques: A review,” Comput. Sci. Rev., vol. 38, p. 100301, 2020, doi: 10.1016/j.cosrev.2020.100301
[10] Y. Jamtsho, P. Riyamongkol, and R. Waranusast, “Real-time license plate detection for non-helmeted motorcyclist using YOLO,” ICT Express, no. xxxx, 2020, doi: 10.1016/j.icte.2020.07.008
[11] A. Bochkovskiy, C. Y. Wang, and H. Y. M. Liao, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv, 2020.
[12] D. Wu, S. Lv, M. Jiang, and H. Song, “Using channel pruning-based YOLO v4 deep learning algorithm for the real-time and accurate detection of apple flowers in natural environments,” Comput. Electron. Agric., vol. 178, no. July, p. 105742, 2020, doi: 10.1016/j.compag.2020.105742
[13] C. Y. Wang, H. Y. Mark Liao, Y. H. Wu, P. Y. Chen, J. W. Hsieh, and I. H. Yeh, “CSPNet: A new backbone that can enhance learning capability of CNN,” IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. Work., vol. 2020-June, pp. 1571–1580, 2020, doi: 10.1109/CVPRW50498.2020.00203
[14] M. Mahasin and I. A. Dewi, “Comparison of CSPDarkNet53 , CSPResNeXt-50 , and EfficientNet-B0 Backbones on YOLO V4 as Object Detector,” Int. J. Eng. Sci. Inf. Technol., vol. 2, no. 3, pp. 64–72, 2022, doi: 10.52088/ijesty.v2i3.291
[15] R. Rizaldi, A. Kurniawati, and C. V. Angkoso, “Implementasi Metode Euclidean Distance untuk Rekomendasi Ukuran Pakaian pada Aplikasi Ruang Ganti Virtual,” J. Teknol. Inf. dan Ilmu Komput., vol. 5, no. 2, p. 129, 2018, doi: 10.25126/jtiik.201852592
[16] M. Nieto and N. Aginako, “Multi-camera BEV video-surveillance system,” Multimed. Tools Appl., vol. 82, pp. 34995–35019, 2023, doi: 10.1007/s11042-023-14416-y
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Nuansa Informatika
This work is licensed under a Creative Commons Attribution 4.0 International License.
