E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.

1051/e3sconf/202343001055
ICMPC 2023

Automated Weapon Detection System in CCTV’s

Through Image Processing
Dharmapuri Siri1*, Palle Bhanu Prasad Reddy1,KVSL Harika1 , S.Ritwika1, Shivani Sisodia2
, Karanam Madhavi1
1Department of Computer Science and Engineering , Gokaraju Rangaraju Institute of Engineering and
Technology, Nizampet Road, Bachupally, Kukatpally, Hyderabad – 500090, Telangana State, India.
2Uttaranchal School of Computing Sciences, Uttaranchal University, Dehradun, 248007, India

Abstract. In our study, we aim to detect various weapons through image

processing using a combination of traditional machine learning algorithms
and deep learning techniques. This would prove beneficial in anticipating
unusual occurrences. The utilization of Closed-Circuit Televisions (CCTVs)
has become widespread in the security and surveillance industry; however,
it requires constant human monitoring to extract useful information, which
is both tedious and prone to errors. The base paper deals with identifying
only firearms. We intend to utilize machine learning and deep learning
methods to recognize different weapons automatically, not only firearms
and alert the CCTV monitor to that specific area.

1 . Introduction
Modern technology uses computer vision algorithms and approaches to identify potential
risks or dangerous situations in visual data. This is known as danger detection using image
processing. This method has drawn a lot of interest in a number of fields, including public
safety, business settings, and transportation, thanks to the growing accessibility of digital
cameras and surveillance systems.
Automatically analysing photos or video streams and spotting visual indicators linked to
potential risks or threats is the main goal of danger detection using image processing[2-5].
This technology may recognise particular patterns or abnormalities that denote harmful
conditions by using sophisticated image analysis techniques, such as object detection, motion
tracking, and anomaly detection.
Real-time monitoring and alarms are one of the major benefits of risk detection through
image processing[8-13]. This technology can quickly spot possible threats by continuously
analysing visual data in real-time, and it can send out rapid alerts to the appropriate people
or systems. When there is a chance for accidents, dangers to be reduced, or lives to be saved,
this competence is essential. Furthermore, image processing-based risk detection can be

* Corresponding author: [email protected]

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative
Commons Attribution License 4.0 (https://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.1051/e3sconf/202343001055
ICMPC 2023

seamlessly integrated with already-installed security networks or surveillance systems. It is

feasible to increase the effectiveness and efficiency of safety standards by integrating this
technology with video surveillance cameras, drones, or even mobile devices[3]. The
proactive monitoring, automatic alerts, and enhanced decision-making processes that can
result from this integration give security staff the ability to react quickly and effectively to
possible threats.
Overall, the ability to detect danger using image processing is a revolutionary technique with
broad applications across many industries. It provides a potent tool for improving situational
awareness and minimising dangers in a variety of settings, including public safety, industrial
settings, transportation, and more. This technology has the ability to greatly enhance safety
precautions and produce safer settings for individuals and communities by utilising the
capabilities of computer vision and sophisticated picture analysis algorithms[4].
2 . Literature Review
2.1 Survey Paper 1
S. Ahmed, H. Wang, and Y. Tian, “Adaptive high-order terminal sliding mode control based
on time delay estimation for the robotic manipulators with backlash hysteresis,” IEEE
Transactions on Systems, Man, and Cybernetics: Systems, vol. 51, no. 2, pp. 1128–1137,
(2021).
Approach: In this paper, unknown dynamics of robotic manipulators in the presence of
uncertainties, external disturbances and backlash hysteresis are considered. Therefore,
further research work suggests that the TDE-based controller can be proposed to design a
robust controller under other nonsmooth nonlinearities, such as saturation and dead-zone.
Furthermore, the suppression of noise effect in terms of chattering in the control inputs can
be considered as future work.
2.2 Survey Paper 2
Warsi, M. Abdullah, M. N. Husen, M. Yahya, S. Khan, and N. Jawaid, “Gun detection system
using YOLOv3,” in Proceedings of the 2019 IEEE International Conference on Smart
Instrumentation, Measurement and Application (ICSIMA), pp. 1–4, IEEE, Kuala Lumpur,
Malaysia, August (2019).
Approach : Every year, a large amount of population reconciles gun-related violence all over
the world. In this paper, we develop a computer-based fully automated system to identify
basic armaments, particularly handguns and rifles. Recent work in the field of deep learning
and transfer learning has demonstrated significant progress in the areas of object detection
and recognition. We have implemented YOLO V3 “You Only Look Once” object detection
model by training it on our customized dataset. The training results confirm that YOLO V3
outperforms YOLO V2 and traditional convolutional neural network (CNN).
3 . Implementation
Similar procedures to the preceding response must be followed, with certain modifications
unique to weapon detection, when implementing weapon detection using image processing
utilising the YOLOv4 algorithm for CCTV surveillance[5]. Here is a detailed manual for
implementing weapon detection:

2
E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.1051/e3sconf/202343001055
ICMPC 2023

3.1 Data gathering and annotation:

Assemble a collection of photographs with different settings involving guns. Mark the areas
of the photographs where there are weapons by annotating them. Make sure the dataset
includes a range of lighting situations, angles, and weapon kinds[1].

3.2 Data Preprocessing:

Resize the photos to the correct input scale for YOLOv4. To increase the diversity of the
dataset, use data augmentation methods such random cropping, flipping, and rotation.

3.3 Model Selection and Setup:

Get YOLOv4 pre-trained weights for your model selection and setup. The environment,
libraries, and dependencies needed for training and inference should be setup. Both
TensorFlow and PyTorch offer YOLOv4 implementations that are readily available.

3.4 Model Fine-tuning:

Utilising your annotated dataset, fine-tune the YOLOv4 model. In particular, if class weights
are necessary, adjust the hyperparameters. You may wish to prioritise high precision to
reduce false positives given the sensitive nature of weapon detection. Post-processing: Use
confidence thresholding and non-maximum suppression (NMS) to weed out duplicate and
low-confidence detections. Make the post-processing specific to your use case.

3.5 Data file

Fig. 1. Weapon Detection Data Files

3.6 Names file

Fig. 2. Weapon Detection of Gun and Knife

3
E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.1051/e3sconf/202343001055
ICMPC 2023

4. Testing Results

Fig. 3. Weapon Detection of Gun

Fig. 4. Weapon Detection and Identification

4
E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.1051/e3sconf/202343001055
ICMPC 2023

Fig.5. Weapon Detection in Video

5. Conclusion
To improve safety and security in varied situations, the risk detection system built through
automatic image processing employing the YOLOv4 algorithm has demonstrated promising
results. The system has demonstrated accurate and real-time identification of possible threats
by utilising deep learning and object detection, enabling proactive efforts to limit risks and
prevent incidents. The idea is technically possible, according to the feasibility analysis, with
the YOLOv4 algorithm delivering great accuracy and efficiency in threat detection.
According to the economic feasibility analysis, the advantages of increased operational
effectiveness, cost savings, and safety outweigh the initial investment needed for hardware,
software, and data collection.
6. Further Scope
Although the YOLOv4 algorithm-based risk detection system is a big accomplishment, there
are a number of opportunities for further development and improvement: Optimisation and
fine-tuning: The YOLOv4 model can be further improved to increase the speed and accuracy
of detection. The model's performance in identifying particular kinds of threats can be
improved by optimising hyperparameters and fine-tuning it using data from the relevant
domains.Integration with other sensors: Combining image processing with other sensor
technologies, such thermal cameras or lidar, can create a more thorough detection system that
can spot more dangerous situations or different kinds of threats. Real-time video analytics:
Increasing the system's capacity to analyse video feeds in real-time can help with proactive
threat identification and response, as well as continuous monitoring of dynamic situations.
Collaboration with current security systems: The risk detection system can be integrated with
current security infrastructure, including access control systems or surveillance networks, to
build a full security ecosystem with improved capabilities.Support for numerous cameras:
Increasing the system's capacity to handle many camera inputs at once can enhance coverage
and make it possible to fully monitor greater regions or challenging settings.

References

1. S. Ahmed, H. Wang, and Y. Tian, “Adaptive high-order terminal sliding mode control based
on time delay estimation for the robotic manipulators with backlash hysteresis,” IEEE
Transactions on Systems, Man, and Cybernetics: Systems, vol. 51, no. 2, pp. 1128–1137,
(2021).
View at: Publisher Site | Google Scholar
2. J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: unified, real-time
object detection,” in Proceedings of the IEEE Conference on Computer Vision and Pattern
Recognition, pp. 779–788, Las Vegas, NV, USA, June (2016).
View at: Google Scholar
3. A. Farhadi and R. Joseph, CVPR, (2018).
View at: Google Scholar
4. M. Grega, A. Matiolański, P. Guzik, and M. Leszczuk,NIH Sensors, vol. 16, no. 1, p. 47,
(2016).
View at: Publisher Site | Google Scholar
5. L. Pang, H. Liu, Y. Chen, and J. Miao, “MDPI,” Sensors, vol. 20, no. 6, p. 1678, (2020)

5
E3S Web of Conferences 430, 01055 (2023) https://doi.org/10.1051/e3sconf/202343001055
ICMPC 2023

View at: Publisher Site | Google Scholar

6. Warsi, M. Abdullah, M. N. Husen, M. Yahya, S. Khan, and N. Jawaid, “Gun detection system
using YOLOv3,” in Proceedings of the 2019 IEEE International Conference on Smart
Instrumentation, Measurement and Application (ICSIMA), pp. 1–4, IEEE, Kuala Lumpur,
Malaysia, August (2019).
View at: Google Scholar
7. G. K. Verma and A. Dhillon, “A handheld gun detection using faster r-cnn deep learning,”
in Proceedings of the 7th International Conference on Computer and Communication
Technology, pp. 84–88, Kurukshetra, Haryana, November (2017).
View at: Google Scholar.
8. P. C. S. Reddy, S. G. Rao, G. R. Sakthidharan and P. V. Rao, Age Grouping with Central
Local Binary Pattern based Structure Co-occurrence Features, In International Conference
on Smart Systems and Inventive Technology (ICSSIT), (2018)
9. Gundavarapu, M.R., Saginala, R., Varma, M.A., ...Bodduluri, A.S., Moparthy, L.C.
Deep Learning Framework for Liver CT Image Segmentation and Risk Prediction in
Conference,Lecture Notes in Networks and Systemsthis link is disabled, 2023,
645 LNNS, pp. 189–201.
10. Chandra Sekhar Reddy P , Sakthidharan G, Kanimozhi Suguna S, Mannar Mannan J,
Varaprasada Rao P, International Journal of Engineering and Advanced Technology. 8,
(2019)
11. P.Chandra Sekhar Reddy, B. Eswara Reddy and V. Vijaya Kumar, International Journal of
Image, Graphics and Signal Processing. 4, (2012)
12. Somasekar, J Ramesh, G “Beneficial Image Preprocessing by Contrast Enhancement
Technique for SEM Images”, IJEMS Vol.29(6) December 2022, NIScPR-CSIR,India.
13. P. C. S. Reddy, S. G. Rao, G. R. Sakthidharan and P. V. Rao, Age Grouping with Central
Local Binary Pattern based Structure Co-occurrence Features, In International Conference
on Smart Systems and Inventive Technology (ICSSIT), (2018)