Analysis of Thermal Imaging Performance under Extreme Foggy Conditions: Applications to Autonomous Driving

Area of Science:

• Automotive perception systems
• Sensor technology for autonomous vehicles
• Computer vision in adverse weather

Background:

• Object detection is crucial for self-driving cars, but current sensors (visible, LIDAR, RADAR) struggle in harsh weather.
• Thermal imaging offers a complementary solution, enabling perception in extreme conditions and compatibility with artificial neural networks.

Purpose of the Study:

• To analyze the resilience of thermal sensors in severe fog conditions.
• To determine the operational limits of thermal cameras in degraded visibility.
• To identify key parameters influencing thermal camera performance in fog.

Main Methods:

• Analysis of thermal sensor resilience using mean pixel intensity and contrast as indicators.
• Evaluation of thermal camera performance under various fog densities.
• Testing object detection software with thermal images from cameras with different angles of view (AOV).

Main Results:

• The angle of view (AOV) significantly impacts object detection performance in foggy conditions.
• Thermal cameras with 18° and 30° AOVs remain effective for object detection in thick fog (down to 13 m meteorological optical range).
• Object detection software achieved a pedestrian detection rate of ≥90% using images from 18° and 30° AOV thermal cameras.

Conclusions:

• Thermal imaging is a viable and robust sensor technology for enhancing self-driving car perception in adverse weather.
• Specific thermal camera AOVs (18° and 30°) are suitable for reliable object detection in challenging fog.
• Further research can leverage these findings to improve the safety and reliability of autonomous driving systems.