Page 92 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

76

YOLOv7-w6, YOLOv8-x but are good results for the problem of detecting hot
weapons.
When compared with YOLO v3 with other models such as SVM, RCNN, CNN
VGG-16 [11], it shows outstanding results. However, through the Pistol data dataset
collected and processed by the author, it can be seen that YOLO v5, 7, 8 model is better
than YOLO v3 in many cases, in which YOLO V7 surpasses all Object Detec models-
tion in both speed and accuracy.

3 Conclusion

This paper has built a Pistol data set for the hot weapon identification problem with
6420 images with a variety of designs and conditions. Experimental results using
YOLOv5-n, YOLOv5-m, YOLOv5l, YOLOv7-X, YOLOv7-W6, YOLOv7-E6,
YOLOv8-l, YOLOv8-x models showed very good results with the problem of hot
weapon detection, the best accuracy of the YOLOv8-x model with 95.6%. Through the
process of comparing the results, the YOLOv7-E6 is the best model recommended. The
results achieved after the test deployment allow the YOLO model to be approached
with the best versions available today to deploy the application for surveillance camera
systems of the security field so that early warning in some cases is positive and
enforceable.

Acknowledgements

We would also like to extend my deepest gratitude to Phenikaa University for their
support.

References

1. Olmos, R., Tabik, S., Lamas, A., Pérez-Hernández, F., Herrera, F.: A binocular image fusion
approach for minimizing false positives in handpistol detection with deep learning. Infor-
mation Fusion, 49, 271-280. 2019.
2. Lamas, A., Tabik, S., Montes, A. C., Pérez-Hernández, F., García, J., Olmos, R., & Herrera,
F.: Human pose estimation for mitigating false negatives in weapon detection in video-sur-
veillance. Neurocomputing. 2022.
3.
ISBN: 978-
604-84-5517-0, trang152-157, Nxb. , 2020.
4. -V3,
.
5.
YOLO- The Information and Communication Technology
Conference (ICT), 2021.
6.
.

CITA 2023 ISBN: 978-604-80-8083-9