Machine Learning Based Enhanced Road Safety
Data Trio: Midhun Jisha Manoj, Risheek Sood, Shaik Sohail Hasan
Project Impact Pipeline
Problem Statement
Drowsiness Detection:
● Currently, there's a noticeable absence of automated tools that factor in driver characteristics. This raises the question:
How can we leverage machine learning algorithms to detect driver drowsiness in real-time, thereby preventing accidents
caused by drowsy driving?
License Plate Recognition:
● Accountability is key to ensuring safety, especially concerning vehicles. Thus, our focus is on identifying number plates to
hold individuals accountable. How can we develop and implement a machine learning model to accurately recognize and
extract license plate numbers from images for the automated enforcement of traffic rules and regulations?
Ticketing Analysis:
● Understanding the significant factors contributing to parking infractions in Toronto is crucial. Moreover, analyzing these
factors to uncover trends and patterns in parking ticket issuance can inform strategic deployment and optimization of
human resources, reducing costs while enhancing performance.
Datasets Results
Parking Tickets: https://open.toronto.ca/dataset/parking-tickets/
Ticketing Analysis License Plate Recognition: Driver Drowsiness Detection
Toronto Neighbourhoods: https://open.toronto.ca/dataset/neighbourhoods/
Driver Population Statistics: https://data.ontario.ca/en/dataset/driver-population-statistics Efficiently pinpointing high-violation areas enables targeted resource allocation, promoting traffic safety and enhancing Using YOLOv5, precision is perfect (1), recall is high (0.958), and mean The accuracy and loss graphs demonstrate strong convergence,
overall safety for drivers, passengers, and pedestrians. Average Precision (mAP) at IoU 0.5 is impressively high (0.993), yet the indicating balanced generalization and absence of overfitting, with both
License Plate Recognition: Open Images Website
overall mAP score (0.734) suggests challenges with overlapping or training and validation metrics closely aligned throughout training.
Drowsiness Detection: https://www.kaggle.com/datasets/dheerajperumandla/drowsiness-dataset
closely packed objects.
Methodology and Tools
Project Learnings Future Scope Data Product
❖ Data Gathering and Integration ❖ Model Development ❖ Website Link: https://youtu.be/wa_gxY8OSuE
❖ Incorporating pedestrian information to analyze interactions between drivers ❖ Integrating weather data to provide insights into ❖ Integrate with emergency response services.
❖ Data Cleaning ❖ Web Application Development ❖ GitHUB Link: https://github.sfu.ca/mja125/ML-Based-Enhanced-Road-Safety
and pedestrians for comprehensive road safety assessment. road conditions affecting safety. ❖ Dataset: https://data.sfgov.org/dataset/EMSA-Emergency-Medical-
❖ Processing, and Visualization ❖ Problem-Solving and Troubleshooting
❖ Dataset: https://universe.roboflow.com/vincent-huard-axo4r/dataset_0610 ❖ Dataset: https://climatedata.ca/download/ Services-Response-Times-Dat/faug-73ss/data_preview
❖ Project Management ❖ Skills Acquisition and Transferability
Use-Case Scenarios
PROJECT PIPELINE
TICKETING ANALYSIS
TICKETING ANALYSIS
TICKETING ANALYSIS
DRIVER POPULATION STATISTICS
LICENSE PLATE RECOGNITION - RESULTS
Model Input Model Output
DRIVER DROWSINESS - RESULTS
Different Classes for our
Classification Model
Data Product / Website
Youtube Link: https://youtu.be/wa_gxY8OSuE
CHALLENGES
CONCLUSION
