Human Detection in Disaster Scenarios for Enhanced Emergency

Response Using YOLO11

Md Sadiq Z Pattankudi, Samarth Uppin, Abdul Rafay Attar, Kunal Bhoomaraddi, Rohan Kolhar

and Sneha Varur

Department of SOCSE, KLE Technological University Hubballi, Karnataka, India

Keywords:

Yolo11, Convolution Neural Network, Disaster Events, UAV and Drones.

Abstract:

In disaster scenarios, the ability to rapidly and accurately detect key elements such as rescuers, victims, ve-

hicles, and dangerous objects is crucial for effective and timely rescue operations. This research proposes

applying You Only Look Once(YOLO11), a real time object detection model to detect aerial images captured

by drone. To train the model, a custom dataset was created, containing four classes—rescuer, victim, vehicles,

and dangerous objects—representing critical components in disaster environments. This dataset provided a

comprehensive and controlled environment to evaluate the system’s performance. The results demonstrated

that the proposed system signiﬁcantly enhances decision making processes, particularly in locating and rescu-

ing human survivors during emergency situations. The model achieved an overall precision of 82.4%, recall of

30.5%, a mean Average Precision at IoU 50 (mAP50) of 36.1%, and a mean Average Precision (mAP) at IoU

50-95 (mAP50- 95) of 16.4%. These performance metrics highlight the reliability of the model in identifying

critical objects in real time, with opportunities for further reﬁnement to improve recall and precision balance,

making it a valuable tool for disaster response teams.

1 INTRODUCTION

Natural disasters(Li et al., 2023) have a severe im-

pact on human life(Baez et al., 2010), and with the

rise in global warming, the frequency of events such

as ﬂoods has signiﬁcantly increased(Banholzer et al.,

2014). Recent extreme weather events in Kerala have

highlighted the ongoing challenges posed by climate

change and vulnerabilities in infrastructure (Chaud-

hary and Piracha, 2021). Rapid response and efﬁ-

cient resource allocation are critical to minimizing ca-

sualties and reducing damage. However, the scale

and unpredictability of such events make traditional

methods of disaster management challenging. In re-

cent years, Unmanned Aerial Vehicles (UAVs)(Das

and Roy, 2023) have emerged as a signiﬁcant technol-

ogy in disaster management. Their ability to navigate

through inaccessible terrains, capture high-resolution

imagery, and deliver real-time data has revolution-

ized emergency response strategies(He et al., 2016;

Sandino et al., 2021). UAVs have been successfully

deployed to assess disaster-hit regions, monitor the

extent of damage(Rahman et al., 2024b), and identify

people in need of assistance. Research demonstrates

that UAVs signiﬁcantly reduce emergency response

times and enhance situational awareness(van Tilburg

and Kaizer, 2021), making them invaluable assets

in the aftermath of natural disasters (Goodchild and

Glennon, 2019). Parallel to advancements in UAV

technology, object detection models have seen rapid

evolution, with You Only Look Once (YOLO)(Terven

et al., 2023) becoming a cornerstone in real-time im-

age processing, It is a state-of-the-art object detec-

tion model that has shown remarkable performance

in terms of both speed and accuracy(Sharma and Ya-

dav, 2023). The recent development of these mod-

els has further enhanced its capabilities, making it

highly suitable for real-time applications where both

detection speed and precision are paramount. In

drone-based disaster response and ability to detect ob-

jects accurately in aerial images becomes essential for

identifying survivors, vehicles in complex and clut-

tered environments. Our work contributes to enhanc-

ing UAV-based disaster response by training a cus-

tom YOLO 11 model on a disaster-speciﬁc dataset

containing classes such as rescuer, victim, vehicles,

and dangerous objects(Patel and Sharma, 2023). The

model is trained to identify critical elements in aerial

Pattankudi, M. S. Z., Uppin, S., Attar, A. R., Bhoomaraddi, K., Kolhar, R. and Varur, S.

Human Detection in Disaster Scenarios for Enhanced Emergency Response Using YOLO11.

DOI: 10.5220/0013601000004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 739-746

ISBN: 978-989-758-763-4

739

images captured during emergencies, enabling rapid

assessment and resource allocation in complex envi-

ronments.

The primary objectives of our study are:

• To create a diverse dataset that captures com-

mon disaster scenarios, including ﬂood zones and

post-earthquake environments, with an emphasis

on detecting humans, vehicles, and hazardous ob-

jects.

• Develop a YOLO 11-based object detection

framework speciﬁcally optimized for drone im-

agery in disaster scenarios, building on this

dataset.

• Evaluate the framework comprehensively based

on critical metrics such as detection accuracy, in-

ference speed, and real-time processing capabili-

ties to ensure its reliability.

The paper contains the sections as follows:

Section II reviews literature survey, highlighting re-

cent advancements in disaster response and object de-

tection models. Section III of the paper provides a

background study, offering an overview of key con-

cepts used in UAV-based disaster management. Sec-

tion IV outlines the proposed methodology, detail-

ing data collection, model training, and deployment

strategies. Section V represents the results and dis-

cussions, analyzing the performance of the system in

various disaster scenarios. Finally, Section VI con-

cludes with a summary of ﬁndings and explores po-

tential directions for future research and implementa-

tion.

2 LITERATURE SURVEY

The integration of AI and drones for disaster man-

agement, particularly in search-rescue operations, has

garnered attention in recent years. Drones equipped

with AI algorithms are used for multiple tasks that in-

clude damage assessment, victim localization, and re-

source allocation. AI enables drones to autonomously

detect critical objects, such as humans, vehicles, and

infrastructure damage, which is crucial in situations

where human intervention is limited or unsafe (Pa-

pyan et al., 2024). Object detection models, particu-

larly those using deep learning techniques,(Deng and

Yu, 2014; Alom et al., 2018) have shown promise

in improving the efﬁciency of disaster response by

enabling real-time identiﬁcation and classiﬁcation

of objects in complex environments (Nehete et al.,

2024).

Existing AI-based solutions for disaster manage-

ment often rely on signal-based detection, such as

mobile phone triangulation, which can be unreliable

in areas where infrastructure is damaged or where sur-

vivors do not have access to mobile phones (Pan et al.,

2023). This has led to an increasing interest in in-

tegrating visual-based detection systems, which can

operate independently of infrastructure, providing a

more robust and versatile approach to search and res-

cue missions (Lygouras et al., 2019).

While signiﬁcant progress has been made in de-

veloping AI-driven drone systems for disaster man-

agement, several gaps remain in the current research.

Many models still struggle with accurate detection in

obstructed environments and often lack the real-time

processing capabilities required for effective deploy-

ment in time-critical scenarios.

Our work involves building a YOLO 11-based ob-

ject detection model on a custom dataset, which di-

rectly addresses the limitations found in existing sys-

tems. YOLO 11 is known for its fast inference time

and high accuracy in detecting objects, even in clut-

tered and partially obstructed environments. By train-

ing the model on a custom dataset that simulates dis-

aster scenarios, we can improve detection accuracy

in environments typical of natural disasters. Addi-

tionally, YOLO 11’s ability to process images in real-

time ensures that the system can be deployed in search

and rescue missions, where every second counts. In

summary, our work builds on existing research by

overcoming key limitations such as detection accu-

racy in complex environments and the need for real-

time, visual-based object detection. By addressing

these gaps, the YOLO 11-based model has the po-

tential to signiﬁcantly enhance the effectiveness of

drone-assisted disaster response systems, ultimately

improving the efﬁciency of search and rescue opera-

tions in natural disasters.

3 BACKGROUND STUDY

3.1 Unmanned Aerial Vehicles

UAV’s(Kumar et al., 2023), usually known as drones,

have proven to be invaluable tools for disaster man-

agement(Xie and Zhao, 2023), providing aerial per-

spectives that allow responders to assess and moni-

tor large areas quickly and efﬁciently(Madnur et al.,

2024). Drones can reach places that are inaccessi-

ble due to hazardous conditions(Wang and Lee, 2024;

Rahman et al., 2024a), making them essential for

locating survivors after natural disasters. These AI

systems often rely on models such as CNNs(O’Shea

and Nash, 2015) and more specialized architecture

like YOLO, which can detect objects in real time.

INCOFT 2025 - International Conference on Futuristic Technology

740

Several notable studies have been pivotal in laying

the foundation for human detection in disaster sce-

narios. These studies have focused on improving

the accuracy and speed of AI-based detection mod-

els in dynamic and complex environments(Norvig,

2022; Chowdhury and Bose, 2024). Through these

advances, our goal is to deploy an edge device capa-

ble of saving lives by helping to carry out effective

search-and-rescue operations.

Figure 1: Schematic diagram of uav (Kontogiannis and

Ekaterinaris, 2013).

3.2 You Only Look Once (YOLO 11)

YOLO 11, an advanced version of the You Only Look

Once (YOLO) series, enhances aerial human detec-

tion in emergency response scenarios. By leveraging

a single deep learning pass over an image, YOLO 11

can quickly detect and classify human ﬁgures in real-

time from aerial footage, such as drones, even in en-

vironments like natural disasters. The network oper-

ates by applying convolutional layers to extract visual

features such as shapes, movements, and patterns at

varying levels of abstraction. This allows it to ac-

curately recognize human ﬁgures, even in cluttered

or obstructed areas. YOLO 11 has been trained in

datasets with various aerial images, allowing it to gen-

eralize across hazardous conditions, providing ﬁrst re-

sponders with immediate information on human loca-

tion during emergencies. It’s high speed and accuracy

make it an invaluable tool in enhancing emergency re-

sponse efforts by enabling faster, more effective res-

cue operations(Khanam and Hussain, 2024).

4 PROPOSED METHOD

4.1 Dataset Collection and Preparation

During the dataset creation process, we encountered

a signiﬁcant challenge: the lack of human presence

in many publicly available disaster scenario image

datasets. This absence of representation for victims

Figure 2: Architecture of YOLO 11 (Rao, 2024).

and rescuers created obstacles in effectively training

the model for real-world applications. To overcome

this, we developed a comprehensive dataset that sim-

ulates various disaster scenarios, such as ﬂoods, wild-

ﬁres, and post-earthquake environments. Drone im-

ages of disaster-stricken areas were initially sourced

from publicly available platforms like Google and dis-

aster management resources.

Using Roboﬂow’s advance annotation tool,

bounding boxes were manually drawn around each

object of interest within the images(Aqeel et al.,

2024). This step was crucial for accurately identify-

ing and localizing objects in disaster scenarios. Each

object within the bounding boxes was classiﬁed into

one of the predeﬁned categories: victim, rescuer,

vehicle, or obstacle. This classiﬁcation ensured that

the dataset was comprehensive and well-structured,

reﬂecting the speciﬁc objects essential for search-

and-rescue missions. By annotating each image in

detail, the process maintained high consistency and

precision, allowing the model to learn the necessary

features for detecting these objects in real-world

disaster situations.

4.2 Model Conﬁguration and Training

Upon completing the annotation process, the dataset

consisted of 150 images, carefully designed to repre-

sent a variety of disaster environments. This ensured

that the model could generalize across different dis-

aster conditions. The annotated dataset was then ex-

ported and prepared for training the YOLOv11 model.

The model was trained on this dataset for 175 epochs,

with a learning rate of 0.01, a value selected based

on previous research, which strikes a balance between

fast convergence and model stability for object detec-

tion tasks.

The model architecture utilizes a standard loss

Human Detection in Disaster Scenarios for Enhanced Emergency Response Using YOLO11

741

Figure 3: Methodology used to train the model.

function for object detection, which combines both

classiﬁcation and localization losses. The total loss

(L) used in the YOLO model is the sum of the classi-

ﬁcation loss (L

cls

), localization loss (L

loc

), and conﬁ-

dence loss (L

con f

L = L

cls

+ L

loc

+ L

con f

(1)

The classiﬁcation loss is typically calculated using the

cross-entropy loss:

cls

= −

∑

i=1

log( ˆy

) (2)

where y

represents the ground truth label, ˆy

is the

predicted class probability, and N is the number of

object classes.

The localization loss is calculated using the mean

squared error between the predicted and true bound-

ing box coordinates:

loc

∑

i=1



− ˆx

)

+ (y

− ˆy

)

+ (w

− ˆw

)

+ (h

−

)



where x

, y

are the center coordinates, and w

, h

are the width and height of the bounding box for the

i-th object.

The conﬁdence loss quantiﬁes how well the model

predicts the presence of an object in a particular cell

of the grid:

con f

∑

i=1



1 −



where C

is the conﬁdence score (whether an ob-

ject is present in the cell), and

is the predicted con-

ﬁdence score.

The learning rate was set at 0.01, following a

learning rate decay scheme to gradually reduce the

rate as training progresses to improve model stability

and performance. The model architecture, compris-

ing 238 layers and 2,582,932 parameters, was specif-

ically designed to handle the complexities of aerial

imagery in disaster scenarios, ensuring that it could

effectively detect and classify objects, even in clut-

tered and obstructed environments typically found in

disaster zones.

The model was trained with a validation set to

monitor performance and prevent overﬁtting. In or-

der to study the generalization capability and practical

performance of the model, the testing was performed

on the test set.

4.3 Area Calculation

Victim detection and positioning depend on the pose

of the UAV camera and its projection of its environ-

mental footprint. A victim is detected if their 2D local

position s

(x, y) lies within the projected footprint of

the camera, determined by summing the angles be-

tween s

and the corners of the boundaries of the

footprint.

Geometrically, based on the design shown in Fig-

ure 6, the expected extent l of a vision-based sensor

2D projected footprint can then be computed using

the following Equations (14) and (15):

top, bottom

= s

(z) · tan



α ± tan

−1



2 f



(3)

left, right

= s

(z) · tan



± tan

−1



2 f



(4)

where s

is the UAV altitude, α and β are the

camera’s pointing angles from the vertical z-axis and

the horizontal x-axis of the world coordinate frame,

w is the lens width, h is the lens height, and f is the

focal length(Nandan Date, 2024).

Figure 4: UAV camera coverage(Sandino, 2022).

INCOFT 2025 - International Conference on Futuristic Technology

742

As illustrated in 4, in Field Of View (FOV) Pro-

jection and Footprint Extent of a Vision-Based Sensor

you have the setup of a camera mounted on the frame

of a UAV deﬁning the variable α as the angle going

up/down from the vertical (or pitch), laying down the

coordinates of the footprint corners c Using the fol-

lowing transformation, the footprint corners c in the

camera center local coordinate frame I are translated

to the world’s coordinate frame W :



(X)

(Y )





(X)

(Y )





cos(ϕ

) − sin(ϕ

)

sin(ϕ

) cos(ϕ

)



c(X)

c(Y )



, (5)

where s

represents the next position of the UAV,

and ϕ

is Euler yaw angle of the UAV, Since no ac-

tions change the heading of the UAV mid-ﬂight, and

yaw estimation errors are negligible, we can rewrite

Equation(4) in the following form:



(X)

(Y )





(X) + c(X)

(Y ) + c(Y )



, (6)

The detection conﬁdence o

that comes as part of

the output data from the CNN object detector is mod-

eled using the following equation:

(1 − ζ

min

)(d

− z

min

+ ζ

min

)

max

− z

min

, (7)

where ζ

min

is the minimally accepted conﬁdence

threshold, z

max

and z

min

are the maximum and mini-

mum UAV ﬂying altitudes, respectively, and d

is the

Manhattan distance between the UAV and the victim.

5 RESULTS AND DISCUSSION

The model has been optimized during training to

alleviate the output of the loss function, and its output

was evaluated based on: precision(P), where P is the

proportion of accurately anticipated positive exam-

ples were among all positive predictions. recall(R),

where R is the proportion of the genuinely antici-

pated positive instances among all positive examples.

mAP50 is a mean Average Precision at a threshold of

50% IoU, which reveals the model’s precision in ob-

jects’ prediction. mAP50-95 – a mean Average Preci-

sion on IoU thresholds.

Fig. 5 shows the F1-score versus conﬁdence of

39% with conﬁdence of 0.51. Basically, the conﬁ-

dence value gives an idea about how conﬁdent the

model is with the defect detection; values close to 1

are completely conﬁdent in the positive detection of

the correct defect.

Figure 5: Plotting of F1-score vs. conﬁdence

Figure 6: Plotting of precision vs. conﬁdence

Fig. 6 illustrates the precision versus conﬁdence

for all classes. The model achieves a precision of

100% at a conﬁdence threshold of 0.98. Higher conﬁ-

dence values indicate more reliable predictions, with

vehicle performing consistently well, while rescuer

and victim exhibit variability.

Figure 7: Plotting of precision vs. recall

Fig. 7 illustrates the precision-recall relationship

of the proposed system, providing an overview of its

detection performance across various classes.

Fig. 8 depicts the relationship between recall and

conﬁdence, illustrating how the model’s ability to

Human Detection in Disaster Scenarios for Enhanced Emergency Response Using YOLO11

743

Figure 8: Plotting of recall vs. conﬁdence.

identify relevant objects varies across different con-

ﬁdence thresholds.

Table 1: Detection performance for different classes.

Class Box(P) R mAP50 mAP50-95

All 0.824 0.305 0.361 0.164

Danger 1.000 0.000 0.0222 0.00731

Rescuer 0.753 0.241 0.371 0.192

Vehicle 0.975 0.500 0.559 0.259

Victim 0.567 0.478 0.493 0.199

From table1, the class-wise performance of the

model is as follows:

• Danger Class: Precision was 1.0, but recall was

0.0, indicating no detections. The mAP50 was

0.0222, and mAP50-95 was 0.00731, showing

poor performance.

• Rescuer Class: Precision was 0.753, and recall

was 0.241. The mAP50 was 0.371, and mAP50-

95 was 0.192, highlighting the need to improve

recall.

• Victim Class: Precision was 0.567, and recall was

0.478. The mAP50 was 0.493, and mAP50-95

was 0.199, indicating moderate performance with

room for improvement.

In terms of inference speed, the model demon-

strated signiﬁcant efﬁciency, which is crucial for real-

time applications in disaster response. The prepro-

cessing time was 0.2 ms, meaning that the data was

prepared quickly for analysis. The inference time,

or the time taken by the model to process each im-

age, was 2.2 ms, which is sufﬁciently fast for rapid

decision-making. Overall, the model’s efﬁciency in

processing and analyzing images makes it suitable for

deployment in emergency response situations, where

every millisecond counts.

As shown in the ﬁgure9, we developed an appli-

cation on the Roboﬂow platform, designed and tested

on an edge device (a mobile phone) within a con-

trolled experimental setting to assess its performance.

Extending this approach, the edge device can be em-

bedded in drones for real-time deployment, enabling

implementation in critical scenarios, such as emer-

gency rescue operations during natural disasters, such

as ﬂoods. This deployment can facilitate rapid re-

sponse by delivering automated insights, enhancing

the efﬁciency and accuracy of search-and-rescue mis-

sions in challenging environments.

The ﬁndings demonstrate YOLO11’s effective-

ness in detecting crucial objects during disaster sce-

narios, signiﬁcantly aiding rescue operations. The

model’s high speed allows it to perform well on

drones with limited processing power, meeting the

real-time constraints necessary for urgent decision-

making. While YOLO11 provides substantial im-

provements over earlier models, challenges such as

detection accuracy in low-light or highly obstructed

conditions remain. Future work could focus on in-

tegrating additional sensors, such as infrared or Li

DAR, to enhance detection capabilities in complex

environments.

6 CONCLUSION

The training results highlight the strengths and ar-

eas for improvement of the model in detecting objects

within disaster scenarios. The model demonstrates

strong performance in detecting victims, as evidenced

by the higher mAP scores (mAP50 = 0.56), suggest-

ing that it is effective at identifying survivors in dis-

tress.However, performance in detecting other criti-

cal objects,such as dangerous objects and rescuer,

showed relatively lower mAP scores, indicating areas

that need further development. This performance gap

can likely be attributed to the relatively small size of

the dataset, especially the limited instances of certain

categories, such as dangerous objects and rescuers.

The model also exhibited strong detection capabili-

ties for vehicles, suggesting that it performs reason-

ably well in identifying vehicles during disaster re-

sponse operations. In disaster response scenarios, the

model’s precision and recall values reﬂect that it is

relatively good at identifying objects of interest, but

there is notable room for improvement, particularly in

detecting distant objects, which are often encountered

in real-world disaster scenarios.

The model’s effectiveness could be enhanced

through techniques such as data augmentation and

expanding the dataset to include a broader range of

disaster environments, helping to improve its gener-

alization capabilities. These ﬁndings are consistent

with similar studies in the ﬁeld, where object detec-

INCOFT 2025 - International Conference on Futuristic Technology

744

Figure 9: Testing in Real Time

tion models, particularly those based on the YOLO

architecture, have been successfully deployed in real-

time disaster monitoring systems, achieving effective

results in terms of both speed and accuracy, to fur-

ther optimize the model’s performance, future work

should include experimenting with different learning

rates and exploring more diverse and comprehensive

datasets. These advancements will be critical for im-

proving the model’s ability to detect various objects

in dynamic and challenging disaster scenarios, ulti-

mately enhancing its deployment in real-time search-

and-rescue operations.

REFERENCES

Alom, J., Taha, T. M., and Asari, V. K. (2018). Deep learn-

ing for computer vision: A brief review. IEEE Access.

Aqeel, M., Norouzzadeh, P., Maazallahi, A., Tutun, S.,

Miab, G., Al Dehailan, L., Stoeckel, D., Snir, E., and

Rahmani, B. (2024). Dental cavity analysis, predic-

tion, localization, and quantiﬁcation using computer

vision. Artiﬁcial Intelligence in Health, 1:80.

Baez, J., Fuente, A., and Santos, I. (2010). Do natural dis-

asters affect human capital? an assessment based on

existing empirical evidence. Institute for the Study of

Labor (IZA), IZA Discussion Papers.

Banholzer, S., Kossin, J., and Donner, S. (2014). The Im-

pact of Climate Change on Natural Disasters, pages

21–49.

Chaudhary, M. T. and Piracha, A. (2021). Natural disas-

ters—origins, impacts, management. Encyclopedia,

1(4):1101–1131.

Chowdhury, N. and Bose, S. (2024). Future perspectives on

ai-assisted disaster management. IEEE Transactions

on Intelligent Systems.

Das, M. and Roy, K. (2023). Enhancing uav accuracy in

search missions with yolov11. IEEE Access.

Deng, L. and Yu, D. (2014). Overview of deep learning.

In Proceedings of the 2014 IEEE International Con-

ference on Acoustics, Speech and Signal Processing

(ICASSP).

Goodchild, M. and Glennon, J. A. (2019). Applications of

unmanned aerial vehicles (uavs) in disaster manage-

ment. Geography Compass.

He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep resid-

ual learning for image recognition. In Proceedings of

the IEEE Conference on Computer Vision and Pattern

Recognition (CVPR).

Khanam, R. and Hussain, M. (2024). Yolov11: An

overview of the key architectural enhancements.

arXiv:2410.17725v1 [cs.CV] 23 Oct 2024.

Kontogiannis, S. G. and Ekaterinaris, J. A. (2013). De-

sign, performance evaluation and optimization of a

uav. Aerospace science and technology, 29(1):339–

350.

Kumar, R., Singh, A., and Patel, V. (2023). Human detec-

tion from unmanned aerial vehicles’ images for search

and rescue missions: A state-of-the-art review. IEEE

Access.

Li, J., Wang, X., and Zhang, H. (2023). Uav-enhanced

Human Detection in Disaster Scenarios for Enhanced Emergency Response Using YOLO11

745

dataset for human detection in disaster scenarios.

arXiv Preprint.

Lygouras, E., Santavas, N., Taitzoglou, A., Tarchanidis, K.,

Mitropoulos, A., and Gasteratos, A. (2019). Unsuper-

vised human detection with an embedded vision sys-

tem on a fully autonomous uav for search and rescue

operations. Sensors, 19(16):3542.

Madnur, P., Shetty, P., Parashetti, G., Varur, S., and M,

M. S. (2024). Advancing in cricket analytics: Novel

approaches for pitch and ball detection employing

opencv and yolov8. In 2024 IEEE 9th International

Conference for Convergence in Technology (I2CT),

pages 1–8.

Nandan Date, Rohan Kolhar, N. B. S. K. S. V. (2024). Au-

tonomous grid-driven unmanned aerial vehicle tech-

nology for precise quantiﬁcation in multi-cropped

ﬁelds. International Journal of Advances in Electron-

ics and Computer Science (IJAEC).

Nehete, P., Dharrao, D., Pise, P., and Bongale, A. (2024).

Object detection and classiﬁcation in human rescue

operations: Deep learning strategies for ﬂooded en-

vironments. International Journal of Safety and Secu-

rity Engineering, 14(2):599–611.

Norvig, P. (2022). Algorithms for artiﬁcial intelligence.

IEEE Signal Processing Magazine.

O’Shea, K. and Nash, R. (2015). An introduction to convo-

lutional neural networks. ArXiv e-prints.

Pan, M., Li, Y., Tan, W., and Gao, W. (2023). Drone-

assisted ﬁngerprint localization based on kernel global

locally preserving projection. Drones, 7(7):480. Sub-

mission received: 16 June 2023 / Revised: 15 July

2023 / Accepted: 18 July 2023 / Published: 20 July

2023.

Papyan, N., Kulhandjian, M., Kulhandjian, H., and

Aslanyan, L. (2024). Ai-based drone assisted human

rescue in disaster environments: Challenges and op-

portunities. Pattern Recognition and Image Analysis,

34(1):169–186.

Patel, V. and Sharma, R. (2023). Improved vehicle detection

using yolov11 in disaster zones. IEEE Transactions on

Transportation Systems.

Rahman, M. A., Islam, M. S., and Hossain, M. S. (2024a).

Human detection in drone images using yolo for

search-and-rescue operations. ACM Digital Library.

Rahman, M. A., Islam, M. S., and Hossain, M. S. (2024b).

A search operation unmanned aerial vehicle using

yolo v5 with real-time human detection and counting.

IEEE Xplore.

Rao, S. N. (2024). Yolov11 architecture explained: Next-

level object detection with enhanced speed and accu-

racy.

Sandino, J. (2022). Autonomous decision-making for UAVs

operating under environmental and object detection

uncertainty. PhD thesis.

Sandino, J., Maire, F., Caccetta, P., Sanderson, C., and Gon-

zalez, L. (2021). Drone-based autonomous motion

planning system for outdoor environments under ob-

ject detection uncertainty. Remote Sensing, 13:4481.

Published on November 8, 2021.

Sharma, K. and Yadav, R. (2023). Advanced ai techniques

in uav rescue missions: A yolov11 case study. IEEE

Transactions on Rescue AI.

Terven, J., C

ordova-Esparza, D.-M., and Romero-

Gonz

alez, J.-A. (2023). A comprehensive review of

yolo architectures in computer vision: From yolov1 to

yolov8 and yolo-nas. Machine Learning and Knowl-

edge Extraction, 5(4):1680–1716.

van Tilburg, C. and Kaizer, M. (2021). Reducing emergency

response times using uavs: A case study. Journal of

Emergency Management.

Wang, J. and Lee, C. (2024). Optimizing yolov11 for uav

disaster response. IEEE Robotics and Automation

Magazine.

Xie, H. and Zhao, Y. (2023). Yolov11 in disaster monitor-

ing: Insights and innovations. IEEE Computational

Intelligence Magazine.

INCOFT 2025 - International Conference on Futuristic Technology

746