Human faces encode a vast amount of information form a subject. Facial processing from visual content allows for user re-identification in videosurveillance scenarios, remote estimation of soft biometric traits such as height and weight and moreover a contactless monitoring of the health status of a subject.

Facial soft biometric and eHealth models traditionally based their estimations on images acquired in the visible spectrum. Despite those networks have reached a practical success, they are highly affected by compromising factors such as occlusion and illumination changes.

Thermal imagery has proven itself as superior to visible imaging in hard conditions such as the presence of smoke, dust and absence of light sources

The visible and thermal visuals were acquired with a FLIR Duo R camera. The visible and thermal dual sensor of this camera are a CCD sensor and an uncooled VOx microbolometer with a pixel resolution of 1920x1080 and 640x512 respectively.

Image and video acquisition were performed in an indoor environment where the ambient temperature was set to 25 degrees Celsius. The acquisition setup included a white wall acting as background, a chair at a fixed distance of 0.25 m from the camera which is placed at a height of 1 meter from the ground, and a two-point lighting kit placed to limit shadows.

Various devices were used for gathering annotation from the subjects. A contactless infrared thermometer with a precision of 0.2 degrees Celsius (C) between 34 and 42.0 degrees C was used for computing the user’s body temperature. For calculating the BP, an OMRON HEM-7155-E tensiometer was employed together with a LED finger oximeter for SpO2 measurement with a precision of 2%. For HR tracking, the subjects were asked to wear a Garmin Vivoactive 4 smartwatch. For quantifying bodyweight related measures, we rely on the RENPHO Body Fat Smart scale which returns 13 metrics including weight and BMI.

Each volunteer participated in two separate acquisition sessions, with an average time interval of 6 weeks.

The visual data includes 6 images per person (3 visible and their associated thermal pair) in each session with 3 different conditions, Neutral (N), Ambient light (A) and an occlusion in the form of eyeglasses (O) resulting in a total of 612 images. In addition, four 60-second videos are recorded per subject in each session with N conditions. The first pair of videos (one in visible spectrum and its paired thermal) are taken after the subject has been resting for at least 5 minutes and the second pair follows moderate exercise in the form of climbing up stairs to increase their HR values making a total of 408 60s videos.

LVT face database contains a .csv file with a compendium of 22 health metrics and soft biometric traits per session per subject. In addition to subject id, gender, age and height were collected to describe the subject. Furthermore, other parameters were quantified to assess health status: body temperature, HR, BP, SpO2, weight and BMI. Likewise to weight and BMI, the smart scale used during the data collection, provided other 11 variables: body fat and body water percentages, skeletal muscle, fat-free weight, muscle mass and bone mass, protein, subcutaneous and visceral fat, Basal Metabolic Rate (BMR) and metabolic age.

Any publication using this database must cite the following paper:

Mirabet-Herranz, N., and Dugelay, J. L. (2023, September). LVT Face Database: A benchmark database for visible and hidden face biometrics. In BIOSIG 2023, 22nd International Conference of the Biometrics Special Interest Group.

@inproceedings{mirabet2023lvt,
title={LVT Face Database: A benchmark database for visible and hidden face biometrics},
author={Mirabet-Herranz, Nelida and Dugelay, Jean-Luc},
booktitle={BIOSIG 2023, 22nd International Conference of the Biometrics Special Interest Group},
year={2023} }