SlideShare a Scribd company logo
1 of 15
“Engineering Solutions for FIFA World Cup Qatar 2022”
Workshop
Intelligent Crowd Management and Control Systems for FIFA
World Cup Qatar 2022
Presented by: Dr. Yassine Himeur
Professor in Computer Engineering, College of Engineering, Qatar University
Crowd counting
People identification
and tracking
Abnormal event
detection
Presentation Plan
II. Introduction
Social distancing
monitoring
III. Principal crowd management tasks
IV. Current challenges
Computation
complexity
Lack of annotated
datasets
Security and privacy
concerns
V. Perspectives and conclusions
I. Team
Introduction
Love Parade disaster
(Duisburg, 2010)
Ellis Park Stadium disaster
(Johannesburg, 2001)
The PhilSports Stadium stampede
(Manila, 2006)
21 deaths 21 deaths 73 deaths
Why crowd management in sports venues is important
Abnormal event detection
Crowd counting
Facemask detection
Crowd Management Challenges
Social distancing monitoring Security and safety
Crowd Management Challenges
• Ensure the safety and smoothness of the
World Cup events due to the inherent
occlusions and density of the crowd
• Rely on deploying cutting-edge technologies,
such as AI, surveillance drones and ICT to
optimize crowd management.
Crowd counting
Characteristics
• Capture visitor traffic flows
• Understand occupancy level
• Reduce queuing, at entry points and vending areas
• Executed on AI-based smart cameras or common cameras and
dedicated servers as an edge/cloud solution
Application Environments
• Stadiums.
• Sports events.
• Musical events.
• Massive events.
• Etc.
Dilated and scaled neural networks
People identification and tracking
Characteristics
• Consider faces under pose variations using a multi-task CNN
• Manage access control and capacity control
• Identify spectators banned from the facilities
Application Environments
• Stadiums.
• Sports events.
• Musical events.
• Massive events.
• Etc.
People identification and tracking
Set
parameters
ResNet-50
Network
training
Masked face dataset
Images
Labels
Pre-processing
+
SVM
Decision tree
EBT
Performance
measurement
Feature extraction Classification
YOLO-v4
YOLO-v4 Tiny
Mask-RCNN
YOLO-v5
Youtube Press
conferences
TV channels ….
Dataset
collection
• Due to Covid-19 pandemic, masks have become an everyday part of our society
• Face recognition has come under scrutiny regarding the accuracy when a
mask is worn
• Maintain a 96% accuracy with individuals wearing masks
and can set up alerts for any persons not wearing a mask
Abnormal event detection
Characteristics
• Understand big data
• Providing behavioural analytics
• Avoid annotation of anomalous events in training videos
to reduce the computational cost
Application Environments
• Stadiums.
• Sports events.
• Musical events.
• Massive events.
• Etc.
Goals
• Detect violent spectators
• Prevent hooliganism
Social distancing monitoring
Data
collection
Surveillance and monitoring
center
Deep Learning based
analysis
Pedestrian detection
Distance
measurement
Internet
Data center
Characteristics
• Apply AI to detect social distancing
• Create a better fan experience
• Ensure safety is maintained at the same time
Application Environments
• Stadiums.
• Sports events.
• Musical events.
• Massive events.
• Etc.
Social distancing monitoring
Computational complexity
Knowledge distillation
Teacher model
Large neural network
Training data
Student model 1 Student model 2
Loss
Gradient flow
Data flow
Crowd counting using knowledge distillation
Lack of annotated datasets
Batch sample
𝜃𝑖
𝜏
𝜃𝑖
𝑆
𝜃𝑖
𝑓
𝜃𝑖
𝑏
Neuron linear transformation
Frozen
Update
Target model Predict density map
Target domain
Few shot data
Source domain
Scene 400
Scene 0
𝜃𝑆
𝜃𝜏
1 2
3 Update
4
6
5
Load
Football Supporters Crowd (FSC-Set) Dataset Transfer learning
Security and privacy concerns
Benchmarks on diverse CV tasks
Research-oriented
Distributed computing Flexible APIs
FedCV: federated learning frameworks for CV tasks
Personalization
Security/privacy
System efficiency
…
Computer vision applications Federated learning algorithms
CIFAR 10
93%
99%
CIFAR 100
70%
94.5%
ImageNet
80%
88%
Current accuracy in FL
SOTA in centralized ML
How to bridge the research gap for federated learning on Computer vision (CV)?
Can be overcome using:
• Blockchain
• Federated learning
• Edge computing
Conclusion
• Fewer Cameras, Higher Definition
• Technology and the Layered Approach
to Stadium Security
• AI is Changing the Way We Secure Stadiums

More Related Content

Similar to Crwod_Management.pptx

Ubiquitous computing presentation 2
Ubiquitous computing presentation 2Ubiquitous computing presentation 2
Ubiquitous computing presentation 2
Arpan Patel
 
Detecting Anomalous Behavior with Surveillance​ Analytics​
Detecting Anomalous Behavior with Surveillance​ Analytics​Detecting Anomalous Behavior with Surveillance​ Analytics​
Detecting Anomalous Behavior with Surveillance​ Analytics​
Databricks
 

Similar to Crwod_Management.pptx (20)

CN presentation
CN presentationCN presentation
CN presentation
 
Regional Cyber Security Summit 2016 May 11th-13th Weston Hotel Nairobi Kenya
Regional Cyber Security Summit 2016 May 11th-13th Weston Hotel Nairobi KenyaRegional Cyber Security Summit 2016 May 11th-13th Weston Hotel Nairobi Kenya
Regional Cyber Security Summit 2016 May 11th-13th Weston Hotel Nairobi Kenya
 
Cloud security From Infrastructure to People-ware
Cloud security From Infrastructure to People-wareCloud security From Infrastructure to People-ware
Cloud security From Infrastructure to People-ware
 
STRATEGIES FOR DEVELOPING & IMPLEMENTING INFORMATION SECURITY POLICIES BASED ...
STRATEGIES FOR DEVELOPING & IMPLEMENTING INFORMATION SECURITY POLICIES BASED ...STRATEGIES FOR DEVELOPING & IMPLEMENTING INFORMATION SECURITY POLICIES BASED ...
STRATEGIES FOR DEVELOPING & IMPLEMENTING INFORMATION SECURITY POLICIES BASED ...
 
Webinar - Reducing the Risk of a Cyber Attack on Utilities
Webinar - Reducing the Risk of a Cyber Attack on UtilitiesWebinar - Reducing the Risk of a Cyber Attack on Utilities
Webinar - Reducing the Risk of a Cyber Attack on Utilities
 
CCTV in the CLOUD
CCTV in the CLOUDCCTV in the CLOUD
CCTV in the CLOUD
 
2 partners ed_kickoff_sirris
2 partners ed_kickoff_sirris2 partners ed_kickoff_sirris
2 partners ed_kickoff_sirris
 
Privacy and technology
Privacy and technologyPrivacy and technology
Privacy and technology
 
weyai cybersecurity.pptx
weyai cybersecurity.pptxweyai cybersecurity.pptx
weyai cybersecurity.pptx
 
EvIM: a real time complex event discovery platform for CPSS
EvIM: a real time complex event discovery platform for CPSSEvIM: a real time complex event discovery platform for CPSS
EvIM: a real time complex event discovery platform for CPSS
 
AI-MiguelGonzalez.pdf
AI-MiguelGonzalez.pdfAI-MiguelGonzalez.pdf
AI-MiguelGonzalez.pdf
 
Data-Driven Monitoring in Safety Critical Infrastructure
Data-Driven Monitoring in Safety Critical InfrastructureData-Driven Monitoring in Safety Critical Infrastructure
Data-Driven Monitoring in Safety Critical Infrastructure
 
SDI @ISCWest 2017: A Systems Integrator Perspective
SDI @ISCWest 2017: A Systems Integrator PerspectiveSDI @ISCWest 2017: A Systems Integrator Perspective
SDI @ISCWest 2017: A Systems Integrator Perspective
 
Ubiquitous computing presentation 2
Ubiquitous computing presentation 2Ubiquitous computing presentation 2
Ubiquitous computing presentation 2
 
Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018
 
Detecting Anomalous Behavior with Surveillance​ Analytics​
Detecting Anomalous Behavior with Surveillance​ Analytics​Detecting Anomalous Behavior with Surveillance​ Analytics​
Detecting Anomalous Behavior with Surveillance​ Analytics​
 
ATAGTR2017 Security Testing / IoT Testing in Real World
ATAGTR2017 Security Testing / IoT Testing in Real WorldATAGTR2017 Security Testing / IoT Testing in Real World
ATAGTR2017 Security Testing / IoT Testing in Real World
 
Digital Forensics Triage and Cyber Security
Digital Forensics Triage and Cyber SecurityDigital Forensics Triage and Cyber Security
Digital Forensics Triage and Cyber Security
 
Deep learning Data Analysis.pptx
Deep learning Data Analysis.pptxDeep learning Data Analysis.pptx
Deep learning Data Analysis.pptx
 
Advanced Analytics and Data Science Expertise
Advanced Analytics and Data Science ExpertiseAdvanced Analytics and Data Science Expertise
Advanced Analytics and Data Science Expertise
 

Recently uploaded

一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
pyhepag
 
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
Amil baba
 
一比一原版纽卡斯尔大学毕业证成绩单如何办理
一比一原版纽卡斯尔大学毕业证成绩单如何办理一比一原版纽卡斯尔大学毕业证成绩单如何办理
一比一原版纽卡斯尔大学毕业证成绩单如何办理
cyebo
 
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
zifhagzkk
 
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptxAudience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Stephen266013
 
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotecAbortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
Abortion pills in Riyadh +966572737505 get cytotec
 
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
acoha1
 

Recently uploaded (20)

社内勉強会資料  Mamba - A new era or ephemeral
社内勉強会資料   Mamba - A new era or ephemeral社内勉強会資料   Mamba - A new era or ephemeral
社内勉強会資料  Mamba - A new era or ephemeral
 
Sensing the Future: Anomaly Detection and Event Prediction in Sensor Networks
Sensing the Future: Anomaly Detection and Event Prediction in Sensor NetworksSensing the Future: Anomaly Detection and Event Prediction in Sensor Networks
Sensing the Future: Anomaly Detection and Event Prediction in Sensor Networks
 
一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
一比一原版(Monash毕业证书)莫纳什大学毕业证成绩单如何办理
 
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
 
一比一原版纽卡斯尔大学毕业证成绩单如何办理
一比一原版纽卡斯尔大学毕业证成绩单如何办理一比一原版纽卡斯尔大学毕业证成绩单如何办理
一比一原版纽卡斯尔大学毕业证成绩单如何办理
 
Digital Marketing Demystified: Expert Tips from Samantha Rae Coolbeth
Digital Marketing Demystified: Expert Tips from Samantha Rae CoolbethDigital Marketing Demystified: Expert Tips from Samantha Rae Coolbeth
Digital Marketing Demystified: Expert Tips from Samantha Rae Coolbeth
 
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
 
Genuine love spell caster )! ,+27834335081) Ex lover back permanently in At...
Genuine love spell caster )! ,+27834335081)   Ex lover back permanently in At...Genuine love spell caster )! ,+27834335081)   Ex lover back permanently in At...
Genuine love spell caster )! ,+27834335081) Ex lover back permanently in At...
 
Seven tools of quality control.slideshare
Seven tools of quality control.slideshareSeven tools of quality control.slideshare
Seven tools of quality control.slideshare
 
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptxAudience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptx
 
MATERI MANAJEMEN OF PENYAKIT TETANUS.ppt
MATERI  MANAJEMEN OF PENYAKIT TETANUS.pptMATERI  MANAJEMEN OF PENYAKIT TETANUS.ppt
MATERI MANAJEMEN OF PENYAKIT TETANUS.ppt
 
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotecAbortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
Abortion pills in Riyadh Saudi Arabia (+966572737505 buy cytotec
 
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
如何办理(UPenn毕业证书)宾夕法尼亚大学毕业证成绩单本科硕士学位证留信学历认证
 
2024 Q2 Orange County (CA) Tableau User Group Meeting
2024 Q2 Orange County (CA) Tableau User Group Meeting2024 Q2 Orange County (CA) Tableau User Group Meeting
2024 Q2 Orange County (CA) Tableau User Group Meeting
 
Predictive Precipitation: Advanced Rain Forecasting Techniques
Predictive Precipitation: Advanced Rain Forecasting TechniquesPredictive Precipitation: Advanced Rain Forecasting Techniques
Predictive Precipitation: Advanced Rain Forecasting Techniques
 
123.docx. .
123.docx.                                 .123.docx.                                 .
123.docx. .
 
Atlantic Grupa Case Study (Mintec Data AI)
Atlantic Grupa Case Study (Mintec Data AI)Atlantic Grupa Case Study (Mintec Data AI)
Atlantic Grupa Case Study (Mintec Data AI)
 
NOAM AAUG Adobe Summit 2024: Summit Slam Dunks
NOAM AAUG Adobe Summit 2024: Summit Slam DunksNOAM AAUG Adobe Summit 2024: Summit Slam Dunks
NOAM AAUG Adobe Summit 2024: Summit Slam Dunks
 
Formulas dax para power bI de microsoft.pdf
Formulas dax para power bI de microsoft.pdfFormulas dax para power bI de microsoft.pdf
Formulas dax para power bI de microsoft.pdf
 
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
 

Crwod_Management.pptx

  • 1. “Engineering Solutions for FIFA World Cup Qatar 2022” Workshop Intelligent Crowd Management and Control Systems for FIFA World Cup Qatar 2022 Presented by: Dr. Yassine Himeur Professor in Computer Engineering, College of Engineering, Qatar University
  • 2. Crowd counting People identification and tracking Abnormal event detection Presentation Plan II. Introduction Social distancing monitoring III. Principal crowd management tasks IV. Current challenges Computation complexity Lack of annotated datasets Security and privacy concerns V. Perspectives and conclusions I. Team
  • 3. Introduction Love Parade disaster (Duisburg, 2010) Ellis Park Stadium disaster (Johannesburg, 2001) The PhilSports Stadium stampede (Manila, 2006) 21 deaths 21 deaths 73 deaths Why crowd management in sports venues is important
  • 4. Abnormal event detection Crowd counting Facemask detection Crowd Management Challenges Social distancing monitoring Security and safety
  • 5. Crowd Management Challenges • Ensure the safety and smoothness of the World Cup events due to the inherent occlusions and density of the crowd • Rely on deploying cutting-edge technologies, such as AI, surveillance drones and ICT to optimize crowd management.
  • 6. Crowd counting Characteristics • Capture visitor traffic flows • Understand occupancy level • Reduce queuing, at entry points and vending areas • Executed on AI-based smart cameras or common cameras and dedicated servers as an edge/cloud solution Application Environments • Stadiums. • Sports events. • Musical events. • Massive events. • Etc. Dilated and scaled neural networks
  • 7. People identification and tracking Characteristics • Consider faces under pose variations using a multi-task CNN • Manage access control and capacity control • Identify spectators banned from the facilities Application Environments • Stadiums. • Sports events. • Musical events. • Massive events. • Etc.
  • 8. People identification and tracking Set parameters ResNet-50 Network training Masked face dataset Images Labels Pre-processing + SVM Decision tree EBT Performance measurement Feature extraction Classification YOLO-v4 YOLO-v4 Tiny Mask-RCNN YOLO-v5 Youtube Press conferences TV channels …. Dataset collection • Due to Covid-19 pandemic, masks have become an everyday part of our society • Face recognition has come under scrutiny regarding the accuracy when a mask is worn • Maintain a 96% accuracy with individuals wearing masks and can set up alerts for any persons not wearing a mask
  • 9. Abnormal event detection Characteristics • Understand big data • Providing behavioural analytics • Avoid annotation of anomalous events in training videos to reduce the computational cost Application Environments • Stadiums. • Sports events. • Musical events. • Massive events. • Etc. Goals • Detect violent spectators • Prevent hooliganism
  • 10. Social distancing monitoring Data collection Surveillance and monitoring center Deep Learning based analysis Pedestrian detection Distance measurement Internet Data center Characteristics • Apply AI to detect social distancing • Create a better fan experience • Ensure safety is maintained at the same time Application Environments • Stadiums. • Sports events. • Musical events. • Massive events. • Etc.
  • 12. Computational complexity Knowledge distillation Teacher model Large neural network Training data Student model 1 Student model 2 Loss Gradient flow Data flow Crowd counting using knowledge distillation
  • 13. Lack of annotated datasets Batch sample 𝜃𝑖 𝜏 𝜃𝑖 𝑆 𝜃𝑖 𝑓 𝜃𝑖 𝑏 Neuron linear transformation Frozen Update Target model Predict density map Target domain Few shot data Source domain Scene 400 Scene 0 𝜃𝑆 𝜃𝜏 1 2 3 Update 4 6 5 Load Football Supporters Crowd (FSC-Set) Dataset Transfer learning
  • 14. Security and privacy concerns Benchmarks on diverse CV tasks Research-oriented Distributed computing Flexible APIs FedCV: federated learning frameworks for CV tasks Personalization Security/privacy System efficiency … Computer vision applications Federated learning algorithms CIFAR 10 93% 99% CIFAR 100 70% 94.5% ImageNet 80% 88% Current accuracy in FL SOTA in centralized ML How to bridge the research gap for federated learning on Computer vision (CV)? Can be overcome using: • Blockchain • Federated learning • Edge computing
  • 15. Conclusion • Fewer Cameras, Higher Definition • Technology and the Layered Approach to Stadium Security • AI is Changing the Way We Secure Stadiums

Editor's Notes

  1. Crowd disasters have taken many human lives. The Love Parade disaster (Duisburg, 2010), the Ellis Park Stadium disaster (Johannesburg, 2001), the PhilSports Stadium stampede (Manila, 2006) are just a few recent examples. Yet, controlling crowds is still an unsolved problem. Fortunately, smart security cameras based on IoT and drone-based surveillance equipped with AI can help improve both security and operations in sports venues and stadiums. To that end, our ultimate goal, to be achieved in this project, is to develop a system that interacts with the crowd in order to prevent escalation of risk situations into actual disasters.
  2. To achieve that goal, The team directed by Dr. Somaya is actively working on developing intelligent crowd management and control systems, including multiple components for crowd counting, people identification and tracking, abnormal event detection (AED), facemask detection, and social distancing monitoring .
  3. In this respect, Crowd management at the World Cup stadiums and their perimeters is crucial to ensure the safety and smoothness of the World Cup events due to the inherent occlusions and density of the crowd inside and outside the stadiums. FIFA World Cup Qatar 2022 will rely on deploying cutting-edge technologies, such as AI, surveillance drones and ICT to optimize crowd management.
  4. In this direction, Our team has first developed a crowd counting system from drones’ data, which exploits the dilated and scaled neural networks to extract pertinent features and density crowd estimations. Crowd Counting is performed whenever crowds become too dense to detect and localize individual objects with tradition object detectors. The results can be viewed and analyzed in a dashboard. The data can be exported in many portable formats, enabling users to further analyze the data or use it in presentations and process evaluations. Our solution can be directly executed on Edge or cloud-based platforms.
  5. The research team's effort has also focused on developing a face recognition system, which considers faces under pose variations using a multitask CNN. This is a robust and agile solution when it comes to managing access control and capacity control and identifying in real time individuals previously banned from the facilities for violent behavior or facilitating the entry of registered users such as VIP members.
  6. On the other hand, because of COVID-19 pandemic, masks have become an everyday part of our society, and therefore, facemask detection has become a challenge. Also, face recognition has come under scrutiny regarding the accuracy when a mask is worn. To that end, we developed a new solution that still maintains a 96% accuracy with individuals wearing masks and can set up alerts for any persons not wearing a mask.
  7. Besides, The research team, has developed a novel Abnormal Event Detection scheme, which aims at learning abnormal actions by training both normal and abnormal segments. It enables to avoid the annotation of anomalous events in training video sequences to reduce the computational cost and hence be easily implemented on drones. Concretely, our solution is able to catch instances of fans’ hooliganism and violence.
  8. In addition, the application of artificial intelligence in stadiums and sports venues can help in monitoring social distancing to create a better fan experience, and ensure safety is maintained at the same time.  To that end, we have developed a new solution to monitor social distancing of spectators inside or outside the stadiums using CNN. This helps identify social distancing violations and report them to a surveillance and monitoring center.
  9. Specifically, our approach is based on first detecting spectators or fans using a CNN-based object detector than transforming video frames to a bird’s eye view, where Euclidian distance metric can be used to estimate the distance between detected spectators. Finally, detected distances are compared with a threshold to detect SD violations.
  10. Despite the progress made by the team, we face different challenges among them is the computational complexity of deep learning models. That is why, we are trying to develop new light-weight models using knowledge distillation. That means a new student model is developed from the initial complex model with much less complexity while maintaining almost the same performance.
  11. Another issue is the lack of annotated datasets. As you know most deep learning and CNN models are based on supervised learning, so they need large-scale datasets with labels. To overcome this issue, we’ve developed a new Football Supporters Crowd (FSC-Set) dataset which includes 6000 manually annotated images of different types of scenes that contain thousands of people gathering in or around the stadiums. Also, we are more focusing now on using transfer learning that enables training Deep learning models on synthetic datasets that are easily created with different software then transfer the learned knowledge to real-world datasets.
  12. Other significant issues with most crowd management systems are the security and privacy concerns that raise when processing spectators images. To close this gap, we plan to use blockchain as it is based on utilizing private and public keys. Blockchain systems use asymmetric cryptography to secure transactions between users, where each user has a public and private key. Also, federated learning has been considered as another option,  as it helps train algorithms across multiple decentralized edge devices or servers holding local data samples, without exchanging them. 
  13. At the end, AI in sports venues and stadiums is not thought of the future, but rather a necessity and goal in today’s post-pandemic world. A layered approach to video surveillance that combines video management technology with trained staff and access control systems is necessary to help security operators effectively detect and deter suspicious events, avoid crowd disasters and ensure healthy environments both within and outside stadium walls.