FIR Image Action Dataset

Haoyu Zhang
[email protected]
Visiongo Co., Ltd. 2020.06.23

Sensors

• Melexis MLX90640-BAA (short for mlx90640)
• Heimann HTPA 32x32d L2.1 (short for htpa32)
• Raspberry Pi Camera Module v2 (short for camerav2)

Condition

Description/Sensors	mlx90640	htpa32-01	htpa32-02	camerav2
Sensor model	Melexis MLX90640-BAA	Heimann HTPA 32x32d L2.1	Heimann HTPA 32x32d L2.1	Raspberry Pi Camera Module v2
Resolution of the sensor	32 × 24 pixels	32 × 32 pixels	32 × 32 pixels	320 × 240 pixels
Sensor position	Mid-center of the ceiling	Mid-center of the ceiling	Corner of the ceiling	Mid-center of the ceiling
Sensor height	2355 mm
Sensor FOV	110° × 75°	90° × 90°	90° × 90°	62.2° × 48.8°
Viewing Direction	Vertically downward	Vertically downward		Vertically downward
Frame rate	8 fps			24fps

Dataset

Environment

• At most one person in the room
• To get closer to the real environment, there is at least one heat source in the room.
• Two or three persons are acting in a sequence (No frames contain more than two persons at the same time.)

Indoor scenes

Actions

• "sit"
• "stand"
• "walk"
• "fall"
• "lie"
• "standup"
• "sitdown"
• "lying"

Dataset file format

The data file name format as following:

<date>_<sensor model>_<indoor scene>_<lighting>_<heat source>.csv

E.g., 20200623_mlx90640_01_natural_none.csv

Sensor model

It's correspond to the sensors in the "Condition" part above, such as "mlx90640", "htpa32-01", "camerav2".

Lighting

• "light" means that data is collected when fluorescent lamp is turned on.
• "dark" means that data is collected at night without any light source.
• "natural" means that data is collected with natural lighting.

Heat source

There are five type of heat source.

• "none": no any heat source.

Citations

@misc{zhang2020fir,
  title={FIR-Image-Action-Dataset},
  author={Zhang, H},
  year={2020},
  publisher={Rep{\'e}r{\'e} {\`a} https://github. com/visiongo-kr/FIR-Image-Action-Dataset\# fir-image~…}
}