AI-Driven Traffic Lights: Reinforcement Learning and YOLOv5 for Smarter Roads | by Keith Chadberc Niven Villanueva | Nov, 2024

1. Video Feed for Object Detection

The system begins by capturing video from web site guests cameras monitoring intersections. These video feeds are processed by YOLOv5 to detect autos in real-time. Each physique is analyzed, and the number of autos on the intersection is counted.

# Course of video frame-by-frame
import cv2
cap = cv2.VideoCapture('traffic_video.mp4')
whereas True:
ret, physique = cap.study()
if not ret:
break
outcomes = model(physique)  # Perform YOLOv5 inference
vehicle_count = len(outcomes.xyxy[0])  # Rely detected autos

The best way it really works:

• The video is break up into frames.
• YOLOv5 processes each physique and detects autos.
• The auto rely is handed to the RL agent, which determines the optimum movement (maintain the sunshine inexperienced, flip it purple, and so forth.).

2. RL Environment for Guests Administration

The RL setting simulates an intersection with web site guests lights, the place the RL agent takes actions (altering lights) based totally on vehicle counts provided by YOLOv5. The RL agent receives rewards based totally on how correctly it manages web site guests, encouraging it to attenuate prepared events and avoid pointless gentle changes.

RL setting particulars:

• State: The number of autos prepared in any respect sides of the intersection.
• Movement: Change the location guests gentle from purple to inexperienced or vice versa.
• Reward: Optimistic reward for decreasing wait events, unfavourable reward for inflicting web site guests jams.

3. Motion of Information and Willpower-Making

• The RL agent receives the number of detected autos from YOLOv5.
• Based mostly totally on the location guests density, it determines whether or not or not the sunshine ought to remain inexperienced or change to purple.
• A timer is included to ensure no gentle stays inexperienced or purple for too prolonged.
• The selection is rendered in precise time, and the location guests lights are managed based totally on this strategies.

# Present the current gentle standing on the physique
cv2.putText(frame_with_detections, f"Delicate: {current_light}", (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, 
(0, 255, 0) if current_light == "GREEN" else (0, 0, 255), 2)
cv2.imshow('Guests Delicate Administration', frame_with_detections)

Summary
Integrating YOLOv5 and RL in web site guests administration applications provides a sturdy decision for dynamic web site guests gentle adjustment, decreasing congestion, and enhancing metropolis transport effectivity.

Future Directions

• Enhanced Teaching: Growing web site guests conditions to extra improve RL model teaching.
• Multi-Intersection Administration: Extending the system to deal with various intersections.
• Prolonged-Time interval Information Analysis: Leveraging information over time to spice up web site guests predictions and administration strategies.