EU Thermal Dataset



EU Thermal Dataset


FLIR European Regional Thermal Dataset for Algorithm Training

The FLIR Enhanced European Thermal Dataset is available for sale to automotive developers. It enables developers to start training convolutional neural networks (CNN), empowering the automotive community to create the next generation of safer and more efficient ADAS and driverless vehicle systems using cost-effective thermal cameras from FLIR.

The dataset was acquired via a thermal camera mounted on a vehicle. It contains a total of 3,895 annotated thermal images with 3,472 images sampled from short videos and images from a continuous 14-second video. Videos were taken at a variety of locations, light conditions, and weather conditions (see "extra_info" in the images section of the annotations json).

The videos were captured at the following locations London (England); Paris (France);  Madrid, Toledo, Granada, Malaga (Spain)


Why Use FLIR Thermal Sensing for ADAS?

The ability to sense thermal infrared radiation, or heat, within the ADAS context provides both complementary and distinct advantages to existing sensor technologies such as visible cameras, Lidar and radar systems:

  • With over 15 years of experience working with Veoneer to make the only automotive-qualified thermal camera, FLIR’s thermal sensors are deployed in over 600,000 cars today for driver warning systems.
  • The FLIR thermal cameras can be used to detect and classify objects in challenging conditions including total darkness, fog, smoke, inclement weather and glare, providing a supplemental dataset beyond LiDAR, radar and visible cameras.
  • When combined with visible-light data and distance scanning data from LiDAR and radar, thermal data paired with machine learning creates a more comprehensive detection and classification system.


Dataset Details & Specifications


  • Synced annotated thermal imagery and annotated RGB imagery for reference. Camera centerlines approximately 2 inches apart and collimated to minimize parallax.


  • Frames were sampled from 191 different videos:
    3,472 frames
    51,939 annotations

Frame Annotation Labels: '

  • Bike: 542
    Bus: 1,047
    Car: 19,771
    Hydrant: 2
    Light: 4,695
    Motorcycle: 1,203
    Person: 14,810
    Sign: 9,125
    Truck: 744
    Total: 51,393


  • Clear
    Partly cloudy


  • City Street: 2,155
    Highway: 814
    Residential: 279
    Parking Lot: 20
    Tunnel: 94
    Other: 94
    Unknown: 16
    Total: 3,472

Time of Day:

  • Day: 57%
    Night: 25%
    Dawn/Dusk: 4%
    Unknown: 14%
    Total: 100%

Image Capture Refresh Rate: 

  • Recorded at 30Hz. Dataset sequences sampled at 2 frames/sec or 1 frame/ second. Video annotations were performed at 30 frames/sec recording.


  • The videos were captured in London ( England ); Paris ( France ); Madrid, Toledo, Granada, Malaga ( Spain )

Capture Camera Specifications: 

  • IR Tau2 640x512, 13mm f/1.0 (HFOV 45°, VFOV 37°) FLIR BlackFly (BFS-U3-51S5C-C) 1280x1024, 4-8mm f/1.4-16 megapixel lens (FOV set to match Tau2)

Capture Camera Specifications:

  • IR Tau2 640x512, 13mm f/1.0 (HFOV 45°, VFOV 37°) FLIR BlackFly (BFS-U3-51S5C-C) 1280x1024, 4-8mm f/1.4-16 megapixel lens (FOV set to match Tau2)

Dataset File Format: 

  • 1. Thermal - 14-bit TIFF (no AGC)
    2. Thermal 8-bit JPEG (AGC applied) w/o bounding boxes embedded in images
    3. Thermal 8-bit JPEG (AGC applied) with bounding boxes embedded in images for viewing purposes
    4. RGB - 8-bit JPEG
    5. Annotations: JSON (MSCOCO format)
    6. No temporal filter applied

Sample Results : 

  • Accuracy (mAP)
    Value Overall: 0.475
    Person: 0.547
    Car: 0.701
    Bus: 0.602
    Motorcycle: 0.724
    Bike: 0.415 

FLIR ADK Training and Development Settings: 

  • Use the FLIR ADK with default settings to begin data collection