FlashLightNet: An End-to-End Deep Learning Framework for Real-Time Detection and Classification of Static and...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Accurate traffic light detection is crucial for autonomous vehicles (AVs) and traffic management.
• Current systems struggle with real-time classification of all traffic light states, including flashing signals.

Purpose of the Study:

• To develop a novel deep learning framework, FlashLightNet, for robust traffic light detection and classification.
• To improve real-time recognition of conventional (red, green, yellow) and flashing (flash red, flash yellow) traffic light states.

Main Methods:

• FlashLightNet integrates YOLOv10n for detection, ResNet-18 for feature extraction, and LSTM for temporal classification.
• A custom dataset of real-world and simulated traffic light videos was created, capturing spatiotemporal dynamics.

Main Results:

• YOLOv10n achieved 99.2% mean average precision (mAP) for traffic light detection.
• The ResNet-18 and LSTM model achieved a 96% F1-score for classifying all traffic light states, including flashing signals.

Conclusions:

• FlashLightNet demonstrates high accuracy and robustness in detecting and classifying traffic light states under challenging conditions.
• The framework offers a significant advancement for AV navigation and intelligent traffic management systems.