SYNOPSIS1

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 10

JNANAVIKAS INSTITUTE OF TECHNOLOGY

Vidya Soudha, Padmashree G.D. Goyalgi Campus, Bangalore-Mysore Highway


Bidadi Bangalore-562109

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

PROJECT SYNOPSIS [21CSP76]

PROJECT TITLE: HELMET DETECTION & FINE


ALERTING SYSTEM

UNDER THE GUIDANCE OF:


HEMANTH KUMAR
Assistant Professor
Department of CSE

SUBMITTED BY:
ASHWINI N – 1JV21CS007
ISHWARYA H N – 1JV21CS022
SAGAR A N- 1JV22CS403
SHALINI B K– 1JV22CS404
Helmet detection and Fine alerting System Using Deep learning

ABSTRACT

In recent years, road accidents involving motorcyclists have significantly increased, making helmet
usage a critical safety measure. Despite existing laws mandating helmet use, many riders neglect to
follow these regulations. This project aims to develop an automated helmet detection and fine alert
system using facial recognition and MobileNet, a deep learning model optimized for real-time
mobile and embedded vision applications. The system is designed to detect whether a rider is
wearing a helmet, identify the rider using facial recognition, and issue a fine if helmet usage is
violated. The combination of MobileNet’s efficient architecture and a facial recognition system
offers a lightweight, accurate, and real-time solution to promote road safety.
LITERATURE REVIEW

Sl.no Method Pros Cons

1 Convolutional Good at extracting Requires large


Neural Networks spatial features annotated datasets
(CNNs) for training
Efficient for 2D High computational
image analysis resources
2 YOLO (You Only Extremely fast Lower accuracy for
Look Once) and suitable for detecting small
real-time objects
applications
Single pass Struggles with
prediction overlapping objects

3 Faster R-CNN Improves robustness Slower compared to


and accuracy single-shot detectors
Good at Increased
handling small computational cost
or occluded
objects
4 SSD (Single Shot Faster than R-CNN Less accurate than
Multibox Detector) Faster R-CNN in
detecting small
objects
Suitable for Sensitive to object
embedded systems scaling

5 Mask R-CNN Provides instance Computationally


segmentation in expensive
addition to object
detection
Potential for Slower than
improved accuracy simpler models
and efficiency like YOLO

Table 2.1: Deep Learning Methods for Helmet Detection


SCOPE OF THE PROJECT

Traffic Regulation Enforcement:

 Helmet Violation Detection: Automatically detect motorcyclists without helmets.


 Fine Automation: Generate alerts or fines for violators by identifying and linking them to a
database.
 Real-time Monitoring: Use live CCTV feeds for continuous detection and enforcement.

Enhanced Road Safety:

 Promote helmet usage by deterring violators through automated systems.


 Reduce road fatalities due to lack of helmet use.

Scalable Urban Implementation:

 Deploy in smart cities as part of intelligent traffic management systems.


 Integrate with surveillance networks at traffic signals, highways, and toll booths.

Face Recognition for Accountability:

 Use LBPH for identifying offenders by comparing detected faces with a centralized
database.
 Link violations to an individual’s vehicle registration and personal details.
System analysis:
3.1 Existing System
A number of approaches have been taken to predict driver attention state for providing traffic safety
and to reduce the number of accidents. However, most of the available systems for driver’s
attention monitoring purpose usually are either expensive or limited to special high-end car models.
These systems cannot be affordable for drivers of low income or developing countries.

DRAWBACKS OF EXISTING SYSTEM


 Most of the methods which were used was only suitable for number plate recognition.
 Analyzed only in the low-level and lesser information was provided to the system which did
not produce more promising result.
 More time was taken to analyze large datasets.

3.2 Proposed System

In this project we are detecting whether two wheeler rider wearing helmet or not, if he is not wearing
helmet then we are going to extract the face images of person not wearing helmet. To recognize the
face LBPH algorithm is used. The main principle involved is object detection using deep learning at 3
levels. To implement above techniques we are following below

steps: 1) First image will be captured to the application and using YOLOV2 we will check whether
image contains person or not, if YOLO model detect both person and motorbike then we will proceed
to step2

2) In this module we will use YOLOV3 model to detect whether object wear helmet or not, if he wears
helmet then application will stop here itself. If rider does not wear helmet then application proceed to
step 3.

3) In this module we will extract recognize person face using LBPH and send fine email using Gmail
API.
Advantages:

 Our system exactly detects the person based on our dataset and put fine on person who
is not wearing helmet.

 It consumes less training time.

 It provides accuracy up to 95%.

SYSTEM REQUIREMENT SPECIFICATION

Hardware REQUIREMENTS:

 System: Pentium i5 Processor or more

 Hard Disk: 500 GB or more

 Ram: 8 GB or more

 Input Devices: Keyboard, Mouse and Camera

SOFTWARE REQUIREMENTS:

 Operating System: Windows 10 or higher version

 Languages: Python

 IDE: Spyder (Anaconda)/PyCharm

 Packages and Libraries: OpenCV, TensorFlow, Keras.


Feasibility Analysis:

1. Technical Feasibility
Advantages

 Real-Time Detection: YOLO's speed and accuracy make it suitable for real-time helmet
detection on live video streams.
 Face Recognition Robustness: LBPH works well with low-resolution images, making it
compatible with standard CCTV cameras.
 Hardware Requirements:
o YOLO can be implemented on GPUs or edge devices like NVIDIA Jetson.
o LBPH has low computational overhead, making it lightweight for face recognition
tasks.
 Scalability: Models can be trained and fine-tuned for regional helmet types or unique traffic
conditions.

Challenges

 Complex Environments:
o YOLO's performance may drop in low light, heavy traffic, or bad weather.
o LBPH may fail for partially covered or obscured faces.
 Integration:
o Merging detection (YOLO) and recognition (LBPH) requires optimized pipelines.

2. Operational Feasibility
Advantages

 Ease of Deployment:
o Can utilize existing CCTV infrastructure, reducing setup costs.
o Software updates can refine the detection and recognition over time.
 Automation:
o Automated fine generation reduces manual intervention by law enforcement.
 Integration with Databases:
o Can link to centralized vehicle registration and law enforcement systems.

Challenges

 Maintenance:
o Periodic retraining of models to account for new types of helmets or changing traffic
patterns.
 Infrastructure Dependence:
o Requires stable internet connectivity for cloud-based operations or high-speed
hardware for edge processing.
 Response Time:
o Real-time detection and recognition systems need low latency to be effective.

3. Economic Feasibility
Advantages

 Cost Savings:
o Reduces the need for on-ground personnel to monitor helmet violations.
o Generates revenue through fines.
 Long-Term Benefits:
o Improved compliance leads to fewer road accidents, reducing public health costs.
 Open Source:
o Both YOLO and LBPH are open-source technologies, reducing initial software
costs.

Challenges

 Initial Investment:
o Hardware (GPUs, edge devices) and software development costs.
o Costs for integrating with existing traffic monitoring systems.

 Scalability Costs:
o Expanding the system to cover an entire city or country increases hardware and
maintenance costs.

You might also like