This project focuses on automating traffic management through an intelligent Computer Vision system. Designed to reduce human error and increase enforcement efficiency, the system processes video feeds to detect multiple types of traffic violations in real-time. It integrates vehicle detection, tracking, and character recognition to generate automated alerts for authorities.

Technical Stack:

• Core Technologies: Python, OpenCV, PyTorch/TensorFlow.
• Models Used: YOLO (You Only Look Once) for object detection; Custom CNNs for violation classification.
• OCR Integration: ANPR (Automatic Number Plate Recognition) using tools like EasyOCR or TrOCR.
• Data Handling: Utilization of datasets like the India Driving Dataset (IDD) for diverse road condition training.

Key Modules & Functionality:

• Multi-Class Violation Detection:
  • Helmet Non-Compliance: Identifies motorcyclists riding without helmets.
  • Triple Riding Detection: Detects more than two passengers on two-wheelers.
  • Signal Violation: Monitors intersections to catch vehicles jumping red lights.
• Automated Number Plate Recognition (ANPR):
  • Once a violation is detected, the system isolates the vehicle’s license plate region.
  • Applies Optical Character Recognition (OCR) to extract the registration number for challan generation.
• Real-Time Analytics: Processes live CCTV feeds with low latency, flagging incidents immediately as they occur.

Impact:

• Reduces the manual workload on traffic police by automating surveillance.
• Improves road safety compliance through consistent, unbiased monitoring.
• Scalable architecture capable of handling multiple video streams simultaneously.