Author(s): Consuelo Latorre Sánchez, Joaquín Sanchiz Navarro, Jose Manuel Rojas Artuñedo, Ricardo Bayona Salvador, Elisa Signes Pérez, Jose Laparra Hernández, Carlos Atienza Vicente, Fermin Basso Della Vedova
Institute of Biomechanics (IBV). Universitat Politècnica de València. Building 9C. Camino de Vera, s/n. (46022) Valencia. Spain
Temperature control optimization has become increasingly important in today's climate scenario. Thermal comfort is a crucial human factor during the performance of tasks, often insofar as user safety and the prevention of errors and accidents are concerned. It is a variable that depends on many physical parameters, including the temperature of the air and the regulation of body temperature, as well as sex, age, clothing and other general and local body characteristics.
In order to detect differences in heat maps due to pathologies, skin conditions or joint injuries, the Instituto de Biomecánica (IBV) has created a methodology that analyses the body's thermoregulatory response and measures surface temperatures and their evolution. Using a huge infrared database (>300,000 images), models have been trained to estimate such parameters as the user’s sex, age, thermotype and identification all of which facilitate the prediction of thermal discomfort by means of a thermal image. Using contactless technology and artificial intelligence, applications have been implemented in the field of Health and Well-being to help professionals (clinicians, physicians, etc.) to diagnose certain diseases, such as circulatory and vascular problems, the effect of therapies or cosmetic products, or to prevent such risks as heat stroke.
The aim of this study is to better understand human thermal behavior by harnessing the potential of thermography. By applying highly promising artificial vision and 3D reconstruction techniques, we have been able to obtain information about a user from thermal images that allows professionals to propose treatment monitoring and helps them to make diagnoses in the field of Health and Well-being and to predict thermal sensation and thermal comfort.
Although considerable progress has been made in recent years in Artificial Intelligence (AI) techniques applied to images, especially in visible or RGB datasets, there is however a gap in the application of thermal images. This project, which merges visible and thermal images and retrains neural networks using the IBV’s experimental databases (>300.000 images), has made it possible to fill that gap. To do so, we have implemented what has been termed “Deep Thermovision”, a thermal model based on the use of thermal images and Deep Learning. (Figure 1).
Figure 1. Study context: approach, scope and applications.
In order to implement our “Deep Thermovision”, the Instituto de Biomecánica (IBV) has integrated thermal information with applications based on anthropometric scanners, thereby making it possible to combine variations in shape, posture, movement and temperature.
Thermography is a non-invasive and radiation-free technique that is particularly useful for the exploration of the topography or temperature pattern of the skin over the entire body (1). One specific application of this branch of knowledge is medical infrared thermography, which is used to analyse functions related to skin temperature. Thanks to the technological advances that have been made, this is nowadays a tried-and-trusted medical measurement tool (2).
When the human body is not in a thermally neutral state, it regulates its body temperature by vasodilatation and sweating. However, in the event of pathologies or injuries to skin, joints or muscles, the body’s thermoregulatory response presents localized and/or generalized changes. Acute and chronic pain, which can be a challenging medical problem in terms of diagnosis and treatment, may be caused by tissue injury, inflammation, a surgical procedure or disease. Pain often goes hand-in-hand with a variation in local temperature.
Skin pathologies also present variations in a subject’s thermal pattern. In such cases, Infrared Imaging (IR) can provide additional and complementary information. Numerous studies carried out with users (Figure 2) under controlled extreme conditions, have revealed differentiated regulation patterns that have motivated the study of thermal information as a complement to traditional techniques.
Figure 2. Tests with users under controlled conditions.
Over the course of many years, the IBV has compiled an extensive and highly varied database of adult users (age, sex, physical condition, BMI, etc.), with different degrees of insulation in clothing, extreme scenarios and different facial and postural orientations. By applying Artificial Intelligence (IA) techniques we have been able to detect and correlate small changes in temperature to the said user information.
In collaboration with companies from the Community of Valencia, we have explored several artificial vision networks, models and libraries and applied Machine Learning and Deep Learning AI techniques to extract information from these images. The work we have undertaken has led us to conclude that open solutions and networks do not work accurately on thermographies, which is why we have used our thermal database to re-train these models. The results have been considerably better and we have been able to detect body shapes and faces and recognize and predict user characteristics (such as age, sex or thermotype), and obtain thermal comfort predictions that are useful when it comes to personalized design or diagnosis.
Within the framework of the TERMO 4D project, funded by the Valencian Institute for Business Competitiveness of the Regional and co-financed by the European Union, we have implemented innovative methodologies based on IR applying AI techniques on a database of a wide range of cameras, ranging from highly accurate and high-cost laboratory systems (< 0.02K), to more low-cost solutions.
This IBV database contains thermograms and thermal sequences, in a temperature range from -5ºC to +40ºC, with different garments and postures. The dataset used in this study contains IR and visible RGB (Red-Green-Blue) images from FLIR T650sc, FLIR A35, OPTRIX thermal cameras and the Intel RealSense depth sensor. For the sake of completeness, open repositories have also been used (3). (Table 1).
Table 1. Collection of images: cameras, resolution, number of images and sample.
We applied a dual methodology: on the one hand, we applied AI on thermal images and, on the other hand, we extrapolated the results on images combined with visible information. By combining these two techniques, we have been able to use a large number of models to optimize both the objective and the subjective information provided by the tests. Table 1 shows the training database used and figure 3 a diagram of the methodology implemented.
Figure 3. Methodology, combination of thermal images with deep learning techniques.
Numerous studies have demonstrated how thermal imaging can be used to locate inflammation, acute or chronic injuries, or pain in the human body. The algorithms put forward by the IBV identify the subject and automatically measure his or her temperature at predefined key points, making it possible to track changes in the temperature of parts of the body and generating a myriad of applications in the field of Healthcare.
In its efforts to improve people's health and quality of life, the IBV is developing different applications related to the measurement of body temperature patterns and is using vision and artificial intelligence techniques to apply those applications in the field of high-resolution thermal imaging. Figure 4 shows a number of examples of evaluated applications, such as thermal asymmetries between extremities, peripheral circulatory pathologies, acute or chronic injuries, or the effect of ambient temperature on human regulatory patterns.
In the field of Healthcare, being able to predict a person's thermal response and identifying local changes in the body associated with temperature variation facilitates the diagnosis of varicose veins, facial infections, inflammation due to hidden lesions or the concentration of fat in the body. In the field of Occupational Healthcare, and in order to protect company personnel exposed to these high temperatures, we are studying how to optimize our analysis processes so that we can personalize them according to each individual's characteristics, such as age, sex or the physiological response of his or her body under certain working conditions, which may leave the worker open to health risks such as heatstroke
Figure 4. Thermal response and its applications: Well-being, health monitoring and thermal comfort.
The IBV has conducted this research within the framework of the TERMO 4D project, in collaboration with companies in the aesthetic, clinical or well-being sectors, among others, such as BIOINICIA, BIONOS BIOTECH, STADLER, VENUE NETWORK, ANALOG DEVICES and VANESA VENDRELL ESTÉTICA.
The TERMO 4D project (Ref. IMDEEA/2022/48) is funded by the aid programme of the Valencian Institute for Business Competitiveness (IVACE) aimed at technology centers in the Valencian Community for the development of non-economic R&D projects carried out in collaboration with companies and co-financed by the European Union.
REFERENCES AND SOURCES
(1) (Fournet, D., 2013; James C et al., n.d.).
(2) (Hildebrandt et al., 2010).
(3) (Nikisins et al., 2014).