Traffic Violation Detection

Overview

Real-time computer vision system detecting traffic violations including red-light running, speeding, and helmet violations using YOLOv8 and custom tracking algorithms.

The Problem

Manual traffic monitoring is inefficient, costly, and prone to human error. Cities need automated systems to enforce traffic rules, reduce accidents, and improve road safety at scale.

The Solution

Developed an end-to-end traffic violation detection system using YOLOv8 for vehicle detection, DeepSORT for tracking, and custom rule engines for violation classification. The system processes multiple camera feeds simultaneously and generates automated alerts with evidence (video clips, timestamps, license plates).

Project Gallery

Technical Architecture

Multi-stream video processing pipeline with real-time violation detection

Methodology

1. Dataset: Custom dataset of 10K+ annotated traffic images from local intersections
2. Fine-tuned YOLOv8-l on traffic-specific classes (cars, bikes, trucks, signals)
3. Implemented virtual tripwires for red-light detection
4. Speed estimation using perspective transformation and tracking
5. Helmet detection for motorcycle riders using separate classifier

Results & Impact

Key Impact

• Deployed at 3 major intersections in pilot program
• Detected 500+ violations in first month
• Reduced manual monitoring costs by 70%
• Improved traffic compliance by 35% at monitored intersections

Challenges & Solutions

Key Implementation

class RedLightDetector:
    def __init__(self, stop_line_coords):
        self.stop_line = stop_line_coords
        self.violations = {}
    
    def check_violation(self, track_id, bbox, signal_state):
        """
        Detect if vehicle crossed stop line during red signal
        """
        vehicle_center = self.get_bbox_center(bbox)
        
        # Check if vehicle crossed stop line
        if self.crossed_line(vehicle_center, self.stop_line):
            if signal_state == 'RED':
                if track_id not in self.violations:
                    self.violations[track_id] = {
                        'timestamp': time.time(),
                        'bbox': bbox,
                        'signal': signal_state
                    }
                    return True
        return False
    
    def crossed_line(self, point, line):
        # Point-line crossing detection
        return point[1] > line['y']  # Simplified

Technologies Used