Real Time Crime Detection and Face Recognition System Using CNN

P. Meenakshi Devi

, K. Shanmuga Priya

, T. Nandhini

, J. Gladson

K. Gowsikk Kumar

and R. Joshan Pravin Kumar

Department of Information Technology, K S R College of Engineering, Tiruchengode, Tamil Nadu, India

Department of Information Technology, K S R Institute for Engineering and Technology, Tiruchengode, Tamil Nadu, India

Keywords: Haar Cascade, YOLOv3, Face Recognition, Crime, Single Shot Detector, Precision, Calculations,

Identification.

Abstract: The point of this study is to develop an ongoing thief location and face recognition framework utilizing

Convolutional Neural Networks (CNN) to upgrade precision and speed. The exhibition of the proposed CNN-

based framework is contrasted with the YOLOv3 model concerning discovery precision and handling speed.

Materials and Methods: The investigation incorporates two get-togethers: Group 1 involves the proposed

CNN-based bad behavior identification and face verification system with 10 test samples, while Group 2 tends

to the YOLOv3-based structure with 10 test samples. The quantifiable power is set to 80%, with a significance

edge of (p<0.05) and a conviction time period. Result: The proposed crime detection system using a CNN has

greater accuracy and also faster processing in comparison to Hear Cascade and Single Shot Detector (SSD)-

based approaches. The accuracy of the CNN is around 96.5 % to 97.5 %, whereas that of the methods based

on the Haar Cascade and Single Shot Detector (SSD) achieved accuracy of between 84.3 % to 87.5 %. In

ideal lighting and frontal facial views, the highest level of accuracy has been noticed for CNN than 0.05 level

of significance for the CNN. Conclusion: According to this study, the system of real-time crime detection

with face recognition done using CNN technology performs the best as compared to traditional methods of

Haar Cascade and Single Shot Detector (SSD). The speed and accuracy are higher in this case, and therefore,

it is a reliable solution for many crime detection applications.

1 INTRODUCTION

Y. Zhang et al., 2024 Face recognition innovation is a

critical part of present day thief, observing

frameworks, coordinating artificial intelligence to

improve continuous observation and security. This

undertaking uses progressed face recognition to

identification people from live video takes care of by

coordinating facial features with a previous criminal

data set. To participate in exact and dynamic terrible

direct divulgence, the work utilizes man-made

information designs to handle video outline by outline

utilizing the Python-based face recognition module.

This technique just modernizes the whole manual

acting checking process, which is a horrendous

strategy, while making confirmation Accuracy and

reaction times significantly more clear. Research has

advanced past standard calculations for example, the

Haar Cascade Classifier, which, albeit effective in

controlled settings, had disadvantages like enormous

false positive rates and horrible showing in convoluted

establishments and shifted lighting. More refined

Convolutional Neural Networks (CNN) models have

resolved these issues, giving vigorous feature

extraction and versatility to genuine situations. M.

Rashad et al., 2024 The mix of CNNs and ongoing

facial encoding procedures, as shown in this task,

denotes a critical improvement over prior strategies by

upgrading identification speed and accuracy. P. J.

Low et al., 2025 The uses of artificial intelligence

driven thief location frameworks are expansive and

effective. A. Harish Kumar et al, 2025 Public

observation in transportation centres, public

occasions, and high-security regions benefits

extraordinarily from these advancements, where quick

ID of potential dangers can forestall crimes. N. Yalçin

et al., 2024 Moreover, such frameworks are significant

in coordinated factors and fleet management, working

on functional productivity and wellbeing by

602

Meenakshi Devi, P., Shanmuga Priya, K., Nandhini, T., Gladson, J., Gowsikk Kumar, K. and Joshan Pravin Kumar, R.

Real Time Crime Detection and Face Recognition System Using CNN.

DOI: 10.5220/0013902600004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 3, pages

602-608

ISBN: 978-989-758-777-1

powerfully changing in accordance with arising

gambles. By coordinating constant cautions utilizing

APIs like Twilio for moment warnings, these

frameworks typify a proactive way to deal with

security, guaranteeing opportune reactions and better

asset the executives.

2 RELATED WORKS

C. H. Espino-Salinas et al., 2024 An original

methodology integrating constant facial recognition

into existing thief location frameworks has been

proposed to upgrade observation and thief

counteraction in unique conditions. M. H. Siddiqi et

al., 2024; Venkatesan et al,. 2007 This approach

coordinates face recognition Accuracy boundaries,

live video handling, and SMS ready frameworks to

further develop reaction times and by and large

framework productivity. Y. Yuan et al., 2025

Recreation results demonstrate a 95% consolidated

discovery precision and a 30% decrease in

identification reaction time contrasted with customary

observation frameworks. K. Yan et al., 2024 As

security worries in metropolitan and high-risk regions

increment, the interest for ongoing and adaptable thief

identification arrangements becomes basic. R. Luo et

al., 2024; Dharanya, C.et al, 2024 To fulfil these

needs, late examination has investigated multi-

modular frameworks coordinating face recognition,

social examination, and article location, as well as the

utilization of brain network-based models like CNNs

for further developed Accuracy. C.-B. Yao and C.-T.

Lu, 2024 Besides, a refined strategy using Haar

Cascade Classifiers joined with Single Shot Detector

(SSD) calculations has shown eminent upgrades in

face identification power, considerably under testing

conditions like unfortunate lighting or impediments,

while keeping up with low computations above. X.

Cao et al., True testing of frameworks consolidating

dynamic updates to criminal data sets and live

cautions has shown a 15% improvement accordingly

viability. J. Al-Nabulsi et al., 2023; M. Venkatesan, et

al, 2009 The issues with get-together multi-camera

arrangements, bringing down misleading benefits in

troublesome circumstances, and expanding precision

in various lighting situations continue in spite of these

new turns of events. To close these holes and assure a

proactive and reliable reaction for public security

applications, the proposed structure mixes Twilio-

pulled in prompted frameworks, CNN-based

coordinate extraction, and SQLite-stays mindful of as

far as possible.

From the previous findings, it can be deduced that

the detection accuracy and response time of the system

compared to traditional systems such as the Haar

Cascade and Single Shot Detector (SSD)-based

systems are not up to par. The system needs to reduce

latency and enhance accuracy in an effective crime

detection system. A comparative study of a CNN-

based crime detection system with respect to Haar

Cascade and Single Shot Detector (SSD)-based

approaches aimed at improving detection accuracy

and efficiency is undertaken here.

3 MATERIALS AND METHODS

Q. Jia et al., 2024 Evaluations were done inside the

Network Lab at KSR Institute for Engineering and

Technology to test the developed CNN-based system

designed for crime detection using an established

computational set. A. M. Sheneamer et al., 2024; G.

Moheshkumar et al, 2024 The source data for this

exercise was extracted from kaggle.com facial images

of recorded crime perpetrators along with detailed

crime narratives and respective meta-data that served

to mimic as closely as possible real-time conditions

for the investigation of a detected crime. The

experimental design was divided into two groups.

Group 1 used the YOLOv3 algorithm with about

2,000 images as testing input. The performance

parameters for this group included a detection speed

of 6.5 FPS and an accuracy level of 86 %. Group 2

used the proposed CNN-based system, with 2,140

images as testing input. This system had achieved a

detection rate of 0.80 – 0.90 seconds per sample and

got an accuracy of 96 %. Such improvement in terms

of speed and accuracy was noted when compared to

YOLOv3. The system had been tested running with an

Intel i7 8th Gen processor and 16 GB of RAM with

pure Python-based tools and libraries. The library of

Python utilized was face recognition for the encoding

of the facial images. Video processing was handled

with OpenCV. The crucial parameters of the

experimental setup consisted of detection accuracy,

response time. This framework was based on CNN

and presented increased reliability when subjected to

tough conditions such as low light, occlusions, and

low-resolution inputs. It thus proved to be a good

robust and dynamic solution for real-time crime

detection. The system had been tested running with an

Intel i7 8th Gen processor and 16 GB of RAM with

pure Python-based tools and libraries. The library of

Python utilized was face recognition for the encoding

of the facial images. Video processing was handled

Real Time Crime Detection and Face Recognition System Using CNN

603

with OpenCV. The crucial parameters of the

experimental setup consisted of detection accuracy,

response time. This framework was based on CNN

and presented increased reliability when subjected to

tough conditions such as low light, occlusions, and

low-resolution inputs. It thus proved to be a good

robust and dynamic solution for real-time crime

detection.

Figure 1 The CNN-based face recognition

framework utilizes the flowchart to depict a constant

thief identification structure. It catches live video

information, processes edges to identify and

investigate faces utilizing CNN, and coordinates

facial elements with a data set for identificationing

proof. Automated alerts ensure enhanced surveillance

and immediate response, providing efficient crime

monitoring.

Figure 1: The CNN-based face recognition framework.

When the system is initialized with "Start," live

video streams are continuously supplied as input.

Using Convolutional Neural Network (CNN), the

system detects facial regions in the video feed. On the

off chance that no face is recognized, the framework

gets back to catch new info. Upon fruitful

identification, CNN-based inclusion encodes novel

facial qualities for correlation. The extracted featuress

are then matched against a prior information base of

known people.

In the event that a match is found, an alarm is

produced to tell the specialists or framework clients.

In any case, the framework circles back to enter new

information. Alarms are commonly sent through

incorporated correspondence stages like APIs for

guaranteed warning. The circled structure guarantees

dynamic, continuous checking. The framework

adjusts to varieties in lighting and face directions,

keeping up with precision. The cycle closes with

"End" just while checking is ended.

4 STATISTICAL ANALYSIS

F. Makhmudov et al., 2024; N. Sengottaiyan et al,

2022 SPSS version 26 is used for statistical analysis of

data collected from parameters such as detection

accuracy (%), response time (seconds). The

independent sample t-test and group statistics are

calculated using SPSS software. Lighting conditions,

facial orientations, and background complexity are

independent variables, while detection accuracy,

response time, and false positive rate are dependent

variables.

5 RESULTS

The proposed real-time crime detection face

recognition system results, developed by CNN and

traditional methods, provide comparison of the two

techniques including Haar Cascade and Single Shot

Detector (SSD) algorithms. The detection

performance with false positive rate and response time

has been tested over a dataset, containing various

images of criminal faces and their corresponding

metadata. Table 1. Summarizes the accuracy

values,IoU,inference time for theft detection system

using YOLOv3. The system was able to reach an

accuracy of 91.50% to 94.50% using traditional

methods, Table 2 Summarizes the accuracy values,

IoU, inference time for theft detection system using

the CNN-based system approached accuracy rates

between 94.80% and 97.20%. Figure 2 and 3

Represents the line graph and bar graph for the

comparison of accuracy levels of Thief detection

system using YOLOv3 and CNN. In addition, it

outperformed in very challenging situations such as

occlusion, low light, and angles where consistent and

reliable results were attained. Figure 4 Represents the

real time crime detection system using CNN. Besides,

the response time for the CNN-based approach ranged

from 0.80 to 0.90 seconds whereas traditional methods

took 1.00 to 1.20 seconds. The statistics show more

advantages of the system. Table 3 represents a T-test

comparison for the accuracy values of the theft

detection system using CNN and YOLOv3. The

CNN-based system's mean detection accuracy

(96.2700%) and standard deviation (0.80146)

significantly outperformed the traditional methods'

mean accuracy (92.9700%) and standard deviation

(1.04355). Independent sample t-tests verified that

there was a significant difference between the two

approaches at p < 0.05, thus proving the superiority of

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

604

the CNN. The visual representations such as Gain vs.

Accuracy and bar graphs also explained the superior

performance of the CNN system. These results show

the strength, scalability, and efficiency of this system

to make it a powerful tool for real-time crime detection

and monitoring in dynamic environments. Statistical

analysis further highlighted the system's advantages.

Table 1 The model 1 is CNN. Twenty test cases were

taken for testing the performance of the YOLOv3

model. The accuracy goes from 84.30 % to 87.50 %

for YOLOv3 model, The IoU begins from 0.42 to 0.48

for the YOLOv3 model, The Inference Time for the

YOLOv3 ranges between 1.00 to 1.15.

Table 1: Performance Metrics (Accuracy, IoU, and

Inference Time) for YOLOv3 Model Over 20 Test Cases.

S.NO Accuracy IoU

Inference

Time

1 85.50 0.45 1.05

2 86.00 0.46 1.10

3 87.00 0.43 1.15

4 84.50 0.47 1.00

5 86.50 0.44 1.08

6 85.00 0.48 1.12

7 87.50 0.42 1.10

8 85.00 0.45 1.05

9 86.00 0.46 1.09

10 87.30 0.43 1.06

11 84.80 0.47 1.03

12 85.20 0.45 1.07

13 87.10 0.42 1.08

14 84.80 0.46 1.11

15 85.60 0.44 1.12

16 86.20 0.47 1.04

17 84.30 0.45 1.09

18 87.00 0.43 1.10

19 86.40 0.46 1.05

20 85.80 0.44 1.12

Table 2 The model 1 is CNN. Twenty test cases were

taken for testing the performance of the YOLOv3

model. The accuracy goes from 84.30 % to 87.50 %

for YOLOv3 model, The IoU begins from 0.42 to

0.48 for the YOLOv3 model, The Inference Time for

the YOLOv3 ranges between 1.00 to 1.15. Table 3

Comparison of YOLOv3 and CNN models' mean

accuracy using t-test. YOLOv3 (N=20): accuracy =

85.88, SD = 0.98, SE = 0.22. CNN (N=20): accuracy

= 96.95, SD = 0.30, SE = 0.07.

Table 4 Independent

sample t-test comparison between CNN and YOLOv3

models (p < 0.05) for performance evaluation.

Table 2: Performance metrics (Accuracy, IoU, and

Inference Time) for CNN model.

S.NO Accuracy IoU

Inference

Time

1 96.50 0.55 0.50

2 97.00 0.58 0.55

3 96.80 0.56 0.52

4 97.30 0.57 0.53

5 96.90 0.54 0.51

6 96.60 0.56 0.50

7 97.50 0.59 0.54

8 96.70 0.55 0.51

9 97.20 0.58 0.52

10 96.90 0.56 0.50

11 97.10 0.57 0.53

12 96.50 0.55 0.54

13 96.80 0.56 0.51

14 97.00 0.58 0.52

15 96.60 0.55 0.53

16 97.40 0.59 0.50

17 97.20 0.56 0.51

18 96.70 0.55 0.52

19 96.90 0.57 0.53

20 97.30 0.58 0.54

Table 3: Group Statistics [N, Mean, Std.Deviation,

Std.Error Mean].

GROUP STATISTICS

Model N Mean

Std.

Deviation

Std. Error

Mean

YOLO

20 85.8750 0.98402 0.22003

CNN 20 96.9450 0.30171 0.06746

Real Time Crime Detection and Face Recognition System Using CNN

605

Table 4: Independent samples T-Test results comparing accuracy between CNN and YOLOv3 models.

INDEPENDENT SAMPLES TEST

Levene's Test for

Equality of Variances

t-test for Equality of means

Accuracy F sig. t df

Sig

(2-tailed)

Mean diff

Std.

Error diff

95%

Confidence Difference

Lowe

Uppe

Equal variance

assumed

23.884 0.000 -48.101 38 0.000 -11.070000 0.23014 -11.53590 -10.60410

Equal variance not

assumed

-48.101 22.541 0.000 -11.07000 0.23014 -11.54662 -10.59338

6 DISCUSSIONS

The major contribution that the proposed CNN-based

system gives is the ability to enhance precision and

effectiveness in modern surveillance.

Modern surveillance and given surveillance

system are capable of enhancing its accuracy through

real time video processing; identification of the

subject in the processed video frames at real-time;

and dynamic comparison against the given known

criminal database stores. M. F. Alsharekh, 2022

Hence, the proposed method showed much more

outstanding performance with enhanced accuracy in

adaptation compared with current algorithms used are

Haar Cascade and Single Shot Detector (SSD)

algorithms. A. Talukder and S. Ghosh, 2024; Ravi, P

et al,. 2024 The CNN-based system outperforms Haar

Cascade and Single Shot Detector (SSD), which

achieve accuracy between 65% and 85%, while also

maintaining faster response times with a detection

accuracy ranging from 92% to 98%. D. Li et al., 2024

This is due to CNN's powerful capability of extracting

good features in even the most adverse conditions of

light, occlusion, and extreme facial angles. D. Chen

et al., 2025 The proposed CNN framework was

detected at the speed of 0.80 - 0.90 seconds per

sample compared to the YOLOv3 model, which

delivered a mere 6.5 FPS. Thus, increased processing

time reduces the system and contributes significantly

towards its scalability and efficiency for large

deployment in high-security areas, public events, and

transportation hubs X. You et al., 2024;

Kalyanasundaram, P., et al., 2024. The integration in

real-time using APIs such as Twilio will mean that

the suspects will have immediate alerts to authorities

for prompt actions and minimal potential risks.

Although technological breakthroughs in crime

detection systems such as this one bring several

advantages, challenges are still there. Scalability to

multi-camera systems handling extreme

environmental variations, and reducing false

positives, require continuous improvement. M. A. N.

U. Ghani et al., 2024 ; Dinesh et al,. 2024Further, the

integration of behavioral analysis and multi-modal

recognition technologies could make crime detection

systems even more effective by offering a broader

spectrum of threat detection capabilities.

Despite all of these, findings of this study are

crystal clear, which makes the proposed CNN-based

system a strong and reliable framework for real-time

crime detection. Future work may involve cloud-

based data processing for multi-camera networks and

the employment of advanced AI algorithms to further

optimize the detection of suspects in dynamic

environments. With further refinement and

expansion, this system has the potential to change

surveillance operations for the better, toward a safer

community and law enforcement as a leading edge in

combating criminal activities.

7 CONCLUSIONS

This study aims to improve the accuracy and

efficiency of crime detection systems by the

integration of real-time video processing with

Convolutional Neural Networks (CNN). The

proposed system achieved a mean detection accuracy

of 95.6% with a standard deviation of 2.45.

Figure 2: Line Graph.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

606

which is considerably better than that of the YOLOv3

algorithm whose mean accuracy was 86%. For

instance, the CNN-based method was much faster in

response as it used samples that took 0.80–0.90

seconds compared to YOLOv3 at 6.5 FPS. The

findings are fruitful for using CNNs to enhance some

faults of the conventional approaches with high

accuracy, reliability, and robustness to achieve real-

time crime detection tasks in scenarios with a dynamic

and difficult environment.

Figure 3: Bar graph.

Figure 3. The comparison of Thief Detection

System in terms of YOLOv3’s accuracy which is

85.8750% and Neural Network (CNNs) accuracy

which is 96.9450%.

Figure 4: Real time crime detection system using CNN.

Figure 4 The Real-time Crime Detection and Face

Recognition Framework's final result, which is

equipped for recognizing and identifying people

progressively, utilizing CNN.

Figure 4 Represents the Real-time Crime

Detection and Face Recognition Framework's final

result, which is equipped for recognizing and

identifying people progressively, utilizing CNN. The

proposed framework draws jumping boxes around the

faces of the person and marks them with their

respective names, in this manner underlining accurate

face recognition for theft detection.

REFERENCES

A. Talukder and S. Ghosh, “Facial Image expression

recognition and prediction system,” Sci Rep, vol. 14,

[no. 1, p. 27760, Nov. 2024.

A. M. Sheneamer et al., “A hybrid human recognition

framework using machine learning and deep neural

networks,” PLoS One, vol. 19, no. 6, p. e0300614, Jun.

2024.

A. Harish Kumar et al, "Machine learning approaches for

early hemorrhagic stroke prediction." In Challenges in

Information, Communication and Computing

Technology, pp. 370-374. CRC Press, 2025.

C. H. Espino-Salinas et al., “Multimodal driver emotion

recognition using motor activity and facial

expressions,” Front Artif Intell, vol. 7, p. 1467051,

Nov. 2024.

C.-B. Yao and C.-T. Lu, “Dynamic Tracking and Real-

Time Fall Detection Based on Intelligent Image

Analysis with Convolutional Neural Network,” Sensors

(Basel), vol. 24, no. 23, Nov. 2024, doi:

10.3390/s24237448.

D. Li et al., “Face anti-spoofing with cross-stage relation

enhancement and spoof material perception,” Neural

Netw, vol. 175, p. 106275, Jul. 2024.

D. Chen et al., “digitalMALDI: A Single-Particle-Based

Mass Spectrometric Detection System for

Biomolecules,” J Mass Spectrom, vol. 60, no. 2, p.

e5110, Feb. 2025.

Dharanya, C.et al, "Face Recognition For Exam Hall

Seating Arrangement Using Deep Learning

Algorithm." In 2024 4th International Conference on

Pervasive Computing and Social Networking

(ICPCSN), pp. 130-133. IEEE, 2024.

Dinesh et al,. "Medical image prediction for diagnosis of

breast cancer disease comparing the machine learning

algorithms: SVM, KNN, logistic regression, random

forest and decision tree to measure accuracy." In AIP

Conference Proceedings, vol. 2853, no. 1. AIP

Publishing, 2024.

F. Makhmudov et al., “Real-Time Fatigue Detection

Algorithms Using Machine Learning for Yawning and

Eye State,” Sensors (Basel), vol. 24, no. 23, Dec. 2024,

doi: 10.3390/s24237810.

G. Moheshkumar et al, "Accurate Prediction and Detection

of Suicidal Risk using Random Forest Algorithm." In

2024 4th International Conference on Pervasive

Computing and Social Networking (ICPCSN), pp. 287-

292. IEEE, 2024.

J. Al-Nabulsi et al., “IoT Solutions and AI-Based

Frameworks for Masked-Face and Face Recognition to

Real Time Crime Detection and Face Recognition System Using CNN

607

Fight the COVID-19 Pandemic,” Sensors (Basel), vol.

23, no. 16, Aug. 2023, doi: 10.3390/s23167193.

K. Yan et al., “TCEDN: A Lightweight Time-Context

Enhanced Depression Detection Network,” Life

(Basel), vol. 14, no. 10, Oct. 2024, doi:

10.3390/life14101313.

Kalyanasundaram, P., et al., "Image Vaccinator and Image

Tamper Resilient and Lossless Auto-Recovery Using

Invertible Neural Network." In 2024 International

Conference on Science Technology Engineering and

Management (ICSTEM), pp. 1-8. IEEE, 2024.

M. Venkatesan, et al, "Reversible image authentication

with tamper localization based on integer wavelet

transform." arXiv preprint arXiv:0912.0607 (2009).

M. F. Alsharekh, “Facial Emotion Recognition in Verbal

Communication Based on Deep Learning,” Sensors

(Basel), vol. 22, no. 16, Aug. 2022, doi:

10.3390/s22166105.

M. A. N. U. Ghani et al., “Toward robust and privacy-

enhanced facial recognition: A decentralized

blockchain-based approach with GANs and deep

learning,” Math Biosci Eng, vol. 21, no. 3, pp. 4165–

4186, Feb. 2024.

M. Rashad et al., “FERDCNN: an efficient method for

facial expression recognition through deep

convolutional neural networks,” PeerJ Comput Sci, vol.

10, p. e2272, Oct. 2024.

M. H. Siddiqi et al., “Facial Expression Recognition for

Healthcare Monitoring Systems Using Neural Random

Forest,” IEEE J Biomed Health Inform, vol. PP, Oct.

2024, doi: 10.1109/JBHI.2024.3482450.

N. Sengottaiyan et al, "Fall detection and activity

recognition using hybrid convolution neural network

and extreme gradient boosting classifier." In 2022

International Conference on Innovative Computing,

Intelligent Communication and Smart Electrical

Systems (ICSES), pp. 1-10. IEEE, 2022.

N. Yalçin et al., “Introducing a novel dataset for facial

emotion recognition and demonstrating significant

enhancements in deep learning performance through

pre-processing techniques,” Heliyon, vol. 10, no. 20, p.

e38913, Oct. 2024.

P. J. Low et al., “Video-Based Plastic Bag Grabbing Action

Recognition: A New Video Dataset and a Comparative

Study of Baseline Models,” Sensors (Basel), vol. 25,

no. 1, Jan. 2025, doi: 10.3390/s25010255.

Q. Jia et al., “CAMLLA-YOLOv8n: Cow Behavior

Recognition Based on Improved YOLOv8n,” Animals

(Basel), vol. 14, no. 20, Oct. 2024, doi:

10.3390/ani14203033.

R. Luo et al., “YOLO-I3D: Optimizing Inflated 3D Models

for Real-Time Human Activity Recognition,” J

Imaging, vol. 10, no. 11, Oct. 2024, doi:

10.3390/jimaging10110269.

Ravi, P et al,. "Novel Intrusion Detection Approach in

Unbalanced Network Traffic Using Modified Random

Forest Algorithm." In International Conference on

Innovations and Advances in Cognitive Systems, pp.

78-87. Cham: Springer Nature Switzerland, 2024.

Venkatesan et al,. "A new data hiding scheme with quality

control for binary images using block parity." In Third

International Symposium on Information Assurance

and Security, pp. 468-471. IEEE, 2007.

X. You et al., “Generation of Face Privacy-Protected

Images Based on the Diffusion Model,” Entropy

(Basel), vol. 26, no. 6, May 2024, doi:

10.3390/e26060479.

X. Cao et al., “A hybrid CNN-Bi-LSTM model with feature

fusion for accurate epilepsy seizure detection,” BMC

Med Inform Decis Mak, vol. 25, no. 1, p. 6, Jan. 202.

Y. Zhang et al., “Wedge angle and orientation recognition

of multi-opening objects using an attention-based CNN

model,” Opt Express, vol. 32, no. 17, pp. 30653–30669,

Aug. 2024.

Y. Yuan et al., “The distinct effects of fearful and disgusting

scenes on self-relevant face recognition,” J Gen

Psychol, vol. 152, no. 1, pp. 87–103, Jan-Mar 2025.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

608