Professional Documents
Culture Documents
Gen 1
Gen 1
Gen 1
PROJECT REPORT
OF BE PROJECT
BACHELOR OF ENGINEERING
Artificial Intelligence & Data Science
SUBMITTED BY
AI & DS Engineering
1
DEPARTMENT OF AI&DS ENGINEERING
CERTIFICATE
This is to certify that the seminar report entitled " UNVEILING ANOMALY” submitted by
is the bonafide record of work done by him under my supervision, in partial fulfillment of the
requirements for the Graduate degree course in AI&DS Engineering from Zeal College of
Engineering & Research, Narhe, Pune, for the academic year 2023-24.
Place: Pune
Date:
2
AI & DS Engineering
ACKNOWLEDGMENT
I would like to express my deepest sense of gratitude to my guide Prof. Dikshendra Sarpate,
who offered his continuous advice and encouragement throughout the course of this seminar
work. I thank him for the systematic guidance and providing all the assistance needed to
complete the work. He inspired me greatly to work in this area. His willingness to motivate me
contributed tremendously to my seminar. His guidance and discussions with me are invaluable
in realization of this report.
I also thank our H.O.D. Prof. Dikshendra Sarpate for their helpful suggestions which led to
improvement in my seminar. It is my pleasure to express great thanks to our Founder Director
Hon. Shri S.M. Katkar sir, and Principal Dr. Ajit M. Kate for their support and
encouragement.
AI&DS Engineering
Z.E.S. ZCOER, Pune
3
AI & DS Engineering
DEPARTMENT OF AI&DS ENGINEERING
This is to certify that the seminar presentation of with seminar title “Unveiling
Anomaly” has been held on at Department of AI&DS Engineering, Zeal College of
Engineering & Research, Narhe, Pune-411041.
(Guide) Prof.
Project Guide External Examiner
4
AI & DS Engineering
Table Of Content
Content Page-No
Abstract 6
Problem Statement 7
Introduction 8
Literature Review 9
Architecture 10
Requirements of 15
design/Algorithms/Techniques
Modules Split Up 17
Proposed System 18
Application 22
Advantages 24
Disadvantages 26
References 27
AI & DS Engineering
5
Abstract :
The Smoking Detection Project addresses the critical need for proactive surveillance in
crowded public spaces to ensure public safety and regulatory compliance. The project focuses
on leveraging advanced computer vision techniques, specifically Convolutional Neural
Networks (CNNs), to enable real-time detection of smoking activities. Smoking in public
spaces poses health risks and violates established regulations, yet existing surveillance
systems often lack the capability to promptly identify and address such incidents.
The project's core objective is the development of an intelligent video surveillance system
that employs CNNs for accurate and efficient smoking detection. By analyzing video streams
in real-time, the system aims to automatically distinguish instances of smoking from diverse
visual data encountered in public environments. The hierarchical feature learning capabilities
of CNNs contribute to the system's adaptability across varying scenarios, including crowded
streets, transportation hubs, and entertainment venues.
Through the implementation of this system, the project endeavors to enhance public safety,
streamline regulatory compliance, and contribute to the creation of secure, smoke-free public
spaces. The automated identification of smoking behaviors enables swift intervention by
authorities, mitigating health risks associated with smoking and reinforcing no-smoking
regulations. The Smoking Detection Project represents a crucial step towards utilizing
technology for the proactive enhancement of public well-being and safety, aligning with the
broader goal of fostering healthier and more compliant communities.
AI & DS Engineering
6
Problem Statement :
The pervasive challenge of maintaining public safety and adherence to regulatory standards in
crowded public spaces prompts the need for advanced video surveillance systems. In this
context, the Smoking Detection Project addresses the specific concern of identifying smoking
activities in real-time within diverse public environments. Smoking in public spaces not only
poses health and safety risks but also violates established regulations. Traditional surveillance
methods are often insufficient for timely detection, necessitating the development of a
sophisticated solution that harnesses the power of computer vision and machine learning.
The primary problem is the lack of an efficient and proactive system to detect smoking
incidents within crowded and dynamic public spaces. Existing surveillance systems are
generally limited to passive monitoring and lack the capability to identify smoking behaviors
promptly. This gap in surveillance capabilities not only compromises public safety but also
hinders the enforcement of no-smoking regulations. The Smoking Detection Project aims to
bridge this gap by deploying an intelligent system that leverages Convolutional Neural
Networks (CNNs) and advanced object detection techniques to identify and alert authorities to
instances of smoking in real-time. The challenge lies in developing a robust and adaptable
system that can accurately distinguish smoking activities from the complex and diverse visual
data inherent in public spaces, while also considering factors such as varying lighting
conditions, behavioral diversity, and real-time processing requirements. By addressing this
problem, the project seeks to enhance public safety, contribute to regulatory compliance, and
provide a scalable solution applicable to a wide range of public environments.
AI & DS Engineering
7
Introduction :
Public safety and adherence to regulatory norms are paramount concerns in crowded
public spaces. The prevalence of smoking in such environments not only poses health
risks but also raises challenges for maintaining a safe and compliant atmosphere.
Traditional surveillance systems often fall short in promptly identifying smoking
incidents, creating a need for a proactive and intelligent solution that harnesses the
capabilities of computer vision and machine learning. The Smoking Detection Project
addresses this imperative by deploying advanced technologies to enable real-time
detection of smoking activities in diverse public settings.
The project's central objective is to design and implement a robust video surveillance
system that employs state-of-the-art Convolutional Neural Networks (CNNs) and
object detection techniques. This intelligent system aims to analyze video streams in
real-time, distinguishing instances of smoking from the complex visual data inherent in
public spaces. By leveraging the hierarchical feature learning capabilities of CNNs, the
system seeks to automatically recognize smoking behaviors across diverse scenarios,
including crowded streets, transportation hubs, entertainment venues, and other public
environments.
The significance of this project lies in its potential to significantly enhance public
safety, streamline regulatory compliance, and contribute to the creation of secure and
smoke-free public spaces. By providing authorities with a tool to promptly identify and
address smoking incidents, the system not only mitigates health risks but also aids in
enforcing no-smoking regulations. The introduction of intelligent video surveillance for
smoking detection represents a proactive step towards creating safer and more
compliant public spaces, aligning with the broader goal of utilizing technology to
enhance the well-being of communities.
AI & DS Engineering
8
Literature Review:
AI & DS Engineering
9
5. Architecture:
Description: This module captures live video feeds from surveillance cameras deployed in
public spaces. It interfaces with the video sources, continuously streaming frames for
analysis.
Functionality: Acquiring real-time video inputs is the initial step in the system's operation,
ensuring a constant flow of visual data for smoking detection.
Description: The preprocessing module is responsible for cleaning and enhancing the
incoming video frames. It may include tasks such as resizing, normalization, and noise
reduction to optimize data quality for subsequent analysis.
Functionality: Preprocessing ensures that the input frames are standardized and ready for
feature extraction, enhancing the overall efficiency of the system.
Description: The heart of the system, this module incorporates Convolutional Neural
Networks (CNNs) to perform hierarchical feature extraction from the preprocessed video
frames. The CNN architecture is designed to recognize patterns and spatial hierarchies
associated with smoking behaviors.
Functionality: CNNs analyze video frames to automatically identify relevant features
indicative of smoking activities, enabling the system to learn and adapt to diverse scenarios.
Description: This module handles the real-time processing of video frames, ensuring prompt
analysis and detection of smoking incidents. It integrates with the feature extraction module
to receive and process continuous video streams.
Functionality: Real-time processing enables the system to make instantaneous decisions,
contributing to the timely identification of smoking behaviors within public spaces.
AI & DS Engineering
10
5.5. Decision and Alerting Module:
Description: The decision and alerting module interprets the output of the feature extraction
module, determining whether smoking activities are present in the video frames. In the case
of a positive detection, it triggers an alert mechanism.
Functionality: This module initiates alerts to relevant authorities, security personnel, or
surveillance centers, ensuring swift intervention in response to identified smoking incidents.
Description: The logging and reporting module records information about detected smoking
incidents, maintaining a comprehensive log for post-analysis, reporting, and system
evaluation purposes.
Functionality: Logging and reporting contribute to system transparency, accountability, and
the ability to analyze trends or patterns in smoking activities over time.
AI & DS Engineering
11
The architecture of the Smoking Detection Project is designed for scalability, adaptability,
and real-time responsiveness, ensuring that it can effectively operate in diverse public spaces
with varying surveillance requirements. The seamless integration of components enables the
system to provide enhanced public safety and regulatory compliance through automated
smoking detection and timely alerting mechanisms.
AI & DS Engineering
12
6. Software Requirement Specification :
The system shall capture live video feeds from surveillance cameras.
Preprocessing Module:
The system shall resize and normalize video frames for optimal analysis.
The system shall utilize Convolutional Neural Networks (CNNs) for hierarchical
feature extraction.
The system shall perform real-time analysis of video frames for smoking
detection.
The system shall determine smoking activities and trigger alerts based on
predefined thresholds.
Alert Mechanism:
The system shall notify relevant authorities or systems upon detecting smoking
incidents.
The system shall maintain a comprehensive log of smoking detection events for
analysis and reporting.
The system may include a user interface for monitoring and managing the
system, displaying alerts, and offering a dashboard for administrators.
AI & DS Engineering
13
6.2. Non-functional Requirements:
Performance:
The system shall process video frames in real-time with minimal latency.
Scalability:
Reliability:
The system shall be reliable, with a low rate of false positives and false negatives
in smoking detection.
Availability:
Security:
Usability:
Maintainability:
AI & DS Engineering
14
7. Requirements of design/Algorithms/Techniques :
7.1. Design Requirements:
1. Modular Architecture:
Requirement: The system should have a modular architecture to allow for easy
scalability, maintenance, and future updates.
2. Scalability:
Requirement: The system must be designed to handle a variable number of
surveillance cameras, ensuring scalability without compromising performance.
3. Real-Time Processing:
Requirement: The design must prioritize real-time processing to ensure prompt
smoking detection in live video streams.
4. User Interface (Optional):
Requirement: If a user interface is implemented, it should be intuitive, providing
administrators with clear insights into smoking detection events.
7.2. Algorithmic Requirements:
1. Convolutional Neural Network (CNN):
Requirement: Implement a CNN-based algorithm for hierarchical feature
extraction, enabling effective smoking detection in diverse scenarios.
2. Transfer Learning:
Requirement: Utilize transfer learning techniques to leverage pre-trained models
on large datasets, enhancing the system's ability to detect smoking behaviors.
3. Confidence Thresholding:
Requirement: Apply confidence thresholding to filter out low-confidence
predictions and improve the accuracy of smoking detection.
4. Non-Maximum Suppression:
Requirement: Implement non-maximum suppression to eliminate redundant
bounding box predictions and refine smoking detection results.
7.3. Techniques Requirements:
1. Data Augmentation:
Requirement: Apply data augmentation techniques during the training phase to
enhance the system's ability to generalize to various environmental conditions.
2. Optical Flow Analysis (Optional):
Requirement: Optionally, incorporate optical flow analysis techniques to detect
motion patterns associated with smoking activities.
AI & DS Engineering
15
3. ONNX Model Export (Optional):
Requirement: If applicable, export the trained model to the ONNX format for
seamless integration into the video surveillance system.
7.4. Integration Requirements:
1. API Integration:
Requirement: The system should provide APIs for easy integration with existing
surveillance systems, allowing for widespread adoption.
2. Communication Protocols:
Requirement: Define clear communication protocols between system modules,
ensuring smooth information flow for timely smoking detection and alerts.
AI & DS Engineering
16
8.0 Modules Split Up :
The Unveiling Anomalies project can be split up into the following modules:
1. Data collection and preprocessing module: This module is responsible for collecting
and preprocessing video data for training and evaluating the deep learning model. The
preprocessing steps may include resizing the videos, extracting features from the
videos, and normalizing the features.
2. Model training module: This module is responsible for training the deep learning
model on the preprocessed video data. The training process involves adjusting the
parameters of the model to minimize the loss function.
3. Model inference module: This module is responsible for using the trained model to
detect anomalies in new video footage. The inference process involves feeding the new
video footage to the model and predicting whether or not the footage contains an
anomaly.
4. System management module: This module is responsible for managing the overall
system. This includes tasks such as starting and stopping the system, configuring the
system parameters, and monitoring the system performance.
AI & DS Engineering
17
9.0 Proposed System :
9.1. Deep Learning Model:
Training Data:
The deep learning model is trained on a diverse dataset of video footage that
includes both normal and anomalous events. This dataset is carefully
curated to cover a wide range of scenarios and anomalies relevant to the
specific application domain.
Model Architecture:
Specify the architecture of the deep learning model, including the types of
layers, activation functions, and any specialized components tailored for
anomaly detection. Convolutional Neural Networks (CNNs) or Recurrent
Neural Networks (RNNs) may be suitable architectures.
Training Process:
Describe the training process, including data preprocessing, model
compilation, and optimization techniques. Transfer learning may be
employed to leverage pre-trained models and adapt them to the anomaly
detection task.
Evaluation Metrics:
Define the evaluation metrics used to assess the performance of the model,
such as precision, recall, F1 score, and area under the ROC curve. The goal
is to achieve minimal false positives and false negatives.
9.2. Software System:
Deployment:
Detail how the trained deep learning model is deployed within the software
system. This may involve containerization technologies like Docker for
efficient deployment across different environments.
Model Management:
Outline the mechanisms for managing the deep learning model, including
versioning, updates, and monitoring. This ensures the model remains
effective over time and can be improved as needed.
User Interface:
Describe the user interface provided by the software system, allowing users
to interact with the system. This interface may include features such as a
dashboard for real-time monitoring, configuration options for adjusting
AI & DS Engineering
18
parameters, and a log of detected anomalies.
Alerting System:
Implement an alerting system within the software to notify users or
administrators when anomalies are detected. Alerts can be sent through various
channels such as email, SMS, or integrated into existing communication
platforms.
9.3. Video Surveillance System Integration:
Continuous Monitoring:
Specify how the proposed system integrates with the existing video surveillance
system. The system should continuously monitor the video footage from
surveillance cameras in real-time.
Anomaly Detection Triggers:
the criteria that trigger the deep learning model to perform anomaly detection.
This could include predefined thresholds or dynamic adaptation based on
environmental conditions.
Alert Generation:
Describe the process of generating alerts within the video surveillance system
when anomalies are detected. Clearly outline how these alerts are communicated
to relevant stakeholders and what information is included.
Integration Protocols:
Identify the integration protocols used to communicate between the proposed
system and the video surveillance system. This may involve APIs, standardized
protocols, or direct integration with existing surveillance software.
AI & DS Engineering
20
7. Version Control:
Git:
Use Git for version control to track changes in code, collaborate with team
members, and maintain a history of the project.
8. IDE (Integrated Development Environment):
Jupyter Notebooks, PyCharm, or Visual Studio Code:
Choose an IDE that supports Python and provides a convenient environment
for developing, debugging, and testing code.
9. Database Management:
MySQL, PostgreSQL, or MongoDB:
Depending on the project requirements, use a database management system to
store metadata, configuration settings, or other relevant information.
10. Communication and Alerting:
Slack, Email, or Twilio (for SMS):
Implement communication channels for receiving alerts. Slack, email, or SMS
notifications can be integrated into the software system.
11. Documentation:
Markdown or Sphinx:
Create and maintain project documentation using Markdown or Sphinx to
ensure clear and comprehensive documentation of code, architecture, and
usage.
12. Project Management:
Jira, Trello, or Asana:
Use project management tools to plan, track progress, and collaborate on tasks
within the development team.
13. Continuous Integration and Continuous Deployment (CI/CD):
Jenkins or GitLab CI:
Set up CI/CD pipelines for automated testing, model training, and deployment
to streamline development processes.
14. Collaboration Tools:
GitHub, GitLab, or Bitbucket:
Utilize platforms for version control, collaboration, and project management to
facilitate teamwork and code sharing.
AI & DS Engineering
21
Application:
AI & DS Engineering
22
8. Event Security:
Use Case: Surveillance during large events or gatherings.
Application: Detecting unusual crowd behaviours, identifying potential
security threats, or managing crowd control.
9. Environmental Monitoring:
Use Case: Surveillance in natural environments.
Application: Detecting anomalies such as illegal logging, poaching, or
environmental hazards to protect natural resources.
10. Critical Infrastructure Protection:
Use Case: Surveillance around critical infrastructure like power plants or water
treatment facilities.
Application: Detecting unauthorized access, unusual activities, or potential
threats to ensure the security of critical facilities.
AI & DS Engineering
23
Advantages:
AI & DS Engineering
25
Disadvantages:
[2] "A Comprehensive Survey of Video Anomaly Detection Techniques"Authors: Chen, L.,
Zhu, W., Tao, D.Journal/Conference: ACM Computing Surveys, 2020.Title: "Efficient
Video Anomaly Detection using Recurrent Neural Networks.
[5] "A Comprehensive Survey of Video Anomaly Detection Techniques"Authors: Chen, L.,
Zhu, W., Tao, D.Journal/Conference: ACM Computing Surveys, 2020
AI & DS Engineering
27