Teledyne FLIR has released an expanded free thermal dataset for ADAS and self-driving vehicle researchers and developers. The expanded starter dataset almost doubles the original, industry-first free dataset and provides more than 26,000 annotated images from the US, England, and France in day and night-time conditions. It also triples the categories and now comprises bus, train, truck, person, bike, car, motorcycle, skateboard, stroller scooter, traffic light, fire hydrant, street sign, dog, and other vehicle labels. The extended free starter thermal imaging dataset enables the automotive and academic community to quickly evaluate the vehicle safety algorithm performance, thermal sensors, and neural network testing, such as the FLIR ADK.
“The industry-first free thermal dataset has been used by developers and as a tool in academic research to explore safety of automated driving technology with thermal imaging,” said Chris Posch, director of automotive engineering at Teledyne FLIR. “The expanded dataset, with more images and categories, will further enhance detection and classifications models, especially focused on automatic braking capabilities as part of a multispectral, sensor-fused system.
”When combined with visible light cameras, LIDAR, and radar, thermal sensors help create a comprehensive and redundant system to identify and classify roadway objects using sensor fusion data. Testing has demonstrated that thermal sensors are uniquely capable of seeing pedestrians, large animals, and other vulnerable road users in conditions where current automatic emergency braking systems are challenged, including win total darkness, most fog, smoke, shadows, inclement weather, and sun or headlight glare.