Traffic lights detection and tracking for HD map creation.

Area of Science:

• Robotics
• Computer Vision
• Geographic Information Systems

Background:

• High-definition maps (HD-maps) are essential for autonomous vehicle localization and navigation.
• Current HD-map creation methods are often costly, time-consuming, or prone to errors.
• Automated solutions are needed for efficient and accurate HD-map generation.

Purpose of the Study:

• To develop an automated pipeline for creating detailed HD-maps of traffic lights.
• To accurately georeference and describe traffic light features (shape, orientation, pictogram) for autonomous driving.
• To improve the robustness and precision of traffic light detection and mapping.

Main Methods:

• A novel pipeline for traffic light HD-map creation using camera data from a surveying vehicle.
• Utilizes consecutive detection of traffic lights and Kalman filtering for precise georeferenced positioning.
• Validation performed on the DriveU Traffic Light Dataset.

Main Results:

• The pipeline demonstrates robust performance in detecting and mapping traffic lights, even with noisy GPS data.
• Successfully identifies distant traffic light targets not present in the original dataset.
• Provides accurate georeferenced positions and descriptions of traffic lights.

Conclusions:

• The developed pipeline offers an effective automated solution for traffic light HD-map creation.
• Enhances the accuracy and reliability of mapping crucial infrastructure for autonomous vehicles.
• Shows potential for improving traffic light detection capabilities in challenging conditions.