THERMOGRAPHIC INFRARED SYSTEM AND ASSOCIATED METHOD

Description

technical field

[0001] The present invention relates to an infrared thermographic system configured for temperature detection (i.e., preferably temperature measurement) of an object. It also relates to a method implemented by said system.

[0002] The field of the invention is the field of thermography, and more particularly the field of real-time infrared thermography.

[0003] Such a device makes it possible to improve the accuracy of a measurement of the apparent temperature of an object in an environment. Prior art

[0004] Currently, there are infrared cameras implementing infrared thermography, with a fixed field of view that is more or less extensive depending on the choice of lens and operating at a rate of a few Hz to a few tens of Hz.

[0005] These infrared cameras include image correction mechanisms because infrared detectors are generally neither uniform nor stable. In fact, infrared cameras are generally stable in terms of gain but not in terms of offset.

[0006] Furthermore, since cameras are not stable, they are often equipped with an internal shutter that allows a uniform scene to be presented to the detector at intervals. The internal shutter therefore does not allow the scene to be presented to the detector with the same characteristics as those seen by the lens.

[0007] It follows that images corrected with an internal shutter exhibit at least spatial non-uniformities at low spatial frequencies due to a failure to compensate for the lens's parasitic flux.

[0008] The aim of the present invention is to resolve at least one of the aforementioned drawbacks by improving stability and / or uniformity and / or field of view.

[0009] Documents US 2013 / 188010 A1, WO 2016 / 146704 A1, US 2013 / 147966 A1 and WO 2004 / 027459 A2 disclose examples of prior art. Description of the invention

[0010] This objective is achieved with an infrared thermographic system configured for the detection or measurement of the temperature of an object according to claim 1.

[0011] The processing module can be arranged and / or programmed to correct spatial non-uniformity by subtracting, pixel by pixel, on each of the acquired images, the spatial non-uniformity of the reference image imaging the non-uniformity correction source and preferably by adding the average value of this non-uniformity.

[0012] The infrared thermographic system according to the invention may include an infrared reference source, called a calibration source, thermostatically controlled or equipped with means for measuring its temperature, at least one of the distinct images of each panorama imaging this calibration source, the processing module being arranged and / or programmed to calibrate in temperature the images acquired from the same panorama according to this calibration source.

[0013] The calibration source can be thermostatically controlled at a temperature above 33 degrees and / or below 40 degrees.

[0014] The processing module can be configured and / or programmed to: determine a distance between the object and the image sensor as a function of a dimension of the object on one of the acquired images and / or determine a position of the object in a field of an image acquired by the image sensor, correct the temperature of the object as a function of the distance determined between the object and the image sensor and / or as a function of the position of the object in a field of an image acquired by the image sensor.

[0015] The thermographic system according to the invention may include at least one visible light image sensor (color or black and white), arranged to capture the object simultaneously with the infrared image sensor.

[0016] The processing module can be arranged and / or programmed to determine surface temperature data (i.e. preferably, measure the surface temperature) of said object, the system further comprising means for measuring ambient temperature around the object, the processing module being arranged and / or programmed to determine an internal temperature of the object as a function of the surface temperature data (i.e. preferably, as a function of the surface temperature) of said object and the measured ambient temperature.

[0017] The processing module can be arranged and / or programmed to determine different temperature values ​​of the object over time and to determine the maximum value among all the previously determined values.

[0018] The infrared thermographic system according to the invention may include means for generating a visual or audible alert if the temperature of the object: is greater than a threshold temperature stored by the processing module, or a temperature difference greater than a threshold difference relative to the temperatures of other objects imaged by the thermographic system.

[0019] The means to generate a visual or audible alert can be arranged and / or programmed to filter out a false alarm by comparing different measured temperatures of the object.

[0020] The processing module can be configured and / or programmed to eliminate the following from images acquired by the image sensor: hot spots on the object, and / or a background surrounding the object before determining the temperature data (i.e., typically, measuring the temperature) of said object.

[0021] The infrared thermographic system includes an optical system, the optical system optically linking the near-field infrared reference source to the image sensor.

[0022] The near-field infrared reference source is close enough to the image sensor to be located outside the depth of field of the optical system optically linking the source to the sensor, so that the sensor is arranged to intercept only light rays from the source in the focal plane of the optical system optically linking the source to the sensor.

[0023] The optical system can be located between the near-field reference source and the image sensor.

[0024] According to yet another aspect of the invention, an infrared thermographic method is proposed for the detection or measurement of the temperature of an object according to claim 13.

[0025] The processing module can correct spatial non-uniformity by subtracting, pixel by pixel, on each of the acquired images, the spatial non-uniformity of the reference image imaging the source of non-uniformity correction and preferably by adding the average value of this non-uniformity.

[0026] The infrared thermographic process according to the invention may include: thermostating and / or temperature measurement of an infrared reference source, known as a calibration source, and at least one of the distinct images of each panorama imaging this calibration source, the processing module calibrating in temperature the images acquired from the same panorama according to this calibration source.

[0027] The calibration source can be thermostatically controlled at a temperature above 33 degrees and / or below 40 degrees.

[0028] The processing module can: determine a distance between the object and the image sensor based on a dimension of the object on one of the acquired images and / or determine a position of the object in a field of an image acquired by the image sensor, correct the temperature of the object based on the distance determined between the object and the image sensor and / or based on the position of the object in a field of an image acquired by the image sensor.

[0029] The infrared thermographic process according to the invention may include capturing, by at least one visible light image sensor (color or black and white), the object simultaneously with the infrared image sensor.

[0030] The processing module can determine surface temperature data (i.e. preferably, measure the surface temperature) of said object, the process further comprising an ambient temperature measurement around the object, the processing module determining an internal temperature of the object as a function of the surface temperature data (i.e. preferably, as a function of the surface temperature) of said object and the measured ambient temperature.

[0031] The processing module can determine different temperature values ​​of the object over time and determines the maximum value among all the previously determined values.

[0032] The infrared thermographic process according to the invention may include generating a visual or audible alert if the object's temperature: is greater than a threshold temperature stored by the processing module, or a temperature difference greater than a threshold difference relative to the temperatures of other objects imaged by the thermographic system.

[0033] The generation of the visual or audible alert can filter out a false alarm by comparing different measured temperatures of the object.

[0034] The processing module can eliminate the following from images acquired by the image sensor: hot spots on the object, and / or a background surrounding the object before determining the temperature data (i.e., typically, measuring the temperature) of said object. Description of the figures and methods of realization

[0035] Other advantages and features of the invention will become apparent upon reading the detailed description of implementations and embodiments, which are by no means limiting, and the following attached drawings: [ Fig. 1 ] there figure 1is a schematic cross-sectional profile view of an infrared thermographic system according to the invention, which is the preferred embodiment of the invention. Fig. 2 ] there figure 2 is a schematic top view of the system according to the invention of the figure 1 .

[0036] We will first describe, with reference to figures 1 and 2 , a first embodiment of an infrared thermographic system (100) according to the invention configured for the temperature detection of an object (11).

[0037] The system (100) includes a thermal infrared image sensor (1) arranged to collect infrared radiation and construct at least one image from this radiation.

[0038] The system (100) is arranged and / or programmed for the detection, tracking of objects (11) and association with a corrected apparent temperature measurement of an object (11).

[0039] The applications are diverse, for example: detection of people with fever in a wide area, early fire detection, monitoring of industrial processes.

[0040] The thermography function is also advantageously associated with the function of detecting and / or tracking intrusions.

[0041] The image sensor (1) is typically a thermal infrared type, for example with microbolometers. The image sensor (1) perceives the surrounding scene and creates at least one high-resolution image of a limited portion of the panorama.

[0042] The term “infrared” refers to radiation related to the natural emission of objects (11) at room temperature or long wave infrared (“Long Wave InfraRed” (LWIR)), the wavelength of which is between 8000 nm and 14000 nm for example.

[0043] According to the figure 1The system (100) also includes a drive support (2) arranged to rotate the sensor (1) around a rotation axis (10). The image sensor (1) is fixed to the drive support (2) such that, as the drive support (2) rotates, the sensor (1) images different distinct areas surrounding the thermographic system. The image sensor (1) is arranged to acquire several distinct images so that the combination of these different images forms a continuous panorama of at least 180 degrees (preferably 360°) around the rotation axis (10) of the drive support (2).

[0044] The image sensor (1) is configured to acquire the panoramic image in segments, generally by column or two-dimensional sector, for example. The panoramic field of view is suitable for areas where people move (11) in multiple directions and are unpredictable, such as in train stations, airport terminals, hospitals, and shopping centers. Constructing a panoramic image from several separate images is achieved using an algorithm or a technology known as image stitching.

[0045] The system (100) is arranged to extend the field of view of the image sensor (1) to a complete or almost complete panorama by associating it with means arranged for real-time correction of the measurement in order to improve the accuracy of the measurement of the apparent temperature of objects (11).

[0046] The drive support (2) is connected to a rotating electromechanical drive device (4) which orients the axis of the image sensor (1) at least in azimuth. An orientation of the axis (10) is also useful depending on the desired vertical field of view.

[0047] The electromechanical drive device (4) in rotation is typically of the compact brushless servomotor type, for example.

[0048] The drive support (2) is arranged to fix the image sensor (1) onto the axis of the electromechanical drive device (4). The image sensor (1) has, for example, a vertical field of view of at least fifteen degrees, typically twenty or forty degrees. Advantageously, the combination of the different images acquired by the image sensor (1) forms a continuous 360° panorama.

[0049] The system 100 also includes a processing module (6) arranged and / or programmed to determine temperature data (i.e. preferably to measure a temperature) of said object (11) from images acquired by the image sensor (1).

[0050] Temperature data determination or temperature measurement preferably corresponds to an absolute temperature measurement, the absolute temperature not corresponding to a simple temperature deviation from an unknown reference, but corresponding to an actual or estimated temperature of the object typically expressed in degrees Celsius, or degrees Fahrenheit or Kelvin.

[0051] The processing module (6) is thus arranged and / or programmed to measure a temperature of said object (11) from the images acquired by the image sensor (1).

[0052] Module (6) includes at least one computer, central processing unit or computing unit, analog electronic circuit (preferably dedicated), digital electronic circuit (preferably dedicated), and / or microprocessor (preferably dedicated), and / or software means.

[0053] According to the figure 1 The system (100) includes a data transmission device (3) configured to transmit: the video signal from the image sensor (1) between the moving part (1) and the fixed part (6), the power supply of the image sensor (1) and the bidirectional communication signals between the image sensor (1) and the control and communication means (9).

[0054] This is a collector configured for the transmission of the video signal from the sensor (1) in a wired or wireless manner to the module (6) and / or to the means (9).

[0055] The control and communication means (9) are arranged to control all the functions of the image sensor (1) and to act on its parameters.

[0056] The control and communication means (9) include at least one computer, a central processing unit or computing unit, an analog electronic circuit (preferably dedicated), a digital electronic circuit (preferably dedicated), and / or a microprocessor (preferably dedicated), and / or software means. According to the figure 1 , the system (100) is fixed on a fixed base (5).

[0057] The system (100) further comprises an electronic module (14) arranged and / or programmed to: generate each panoramic image to send to module (6), manage locally the commands for the sensor (1) and the electromechanical drive device (4), including the control of the camera elements (sensor 1 and its optical system) and communication with the camera, receive commands from unit (6) for the sensor (1) and the electromechanical drive device (4).

[0058] The module (14) includes at least one computer, central processing unit or computing unit, analog electronic circuit (preferably dedicated), digital electronic circuit (preferably dedicated), and / or microprocessor (preferably dedicated), and / or software means.

[0059] The system (100) includes a near-field infrared reference source (7), referred to as the non-uniformity correction source. At least one of the distinct images in each panorama images this non-uniformity correction source. The processing module (6) is configured and / or programmed to correct the spatial non-uniformity of the acquired images according to the non-uniformity correction source. The near-field source (7) is typically a blackbody. This source is made of a sufficiently dense material with conductivity ensuring good thermal uniformity, which guarantees good correction of the spatial non-uniformities of the sensor (1). The material can advantageously be an aluminum alloy, but if the source is passive, a dense plastic material may suffice. The material is coated with a layer having an emissivity close to 1, which can advantageously be deposited as a paint film.

[0060] Source 7 is in the near field.

[0061] The term "near-field source" refers to a source that is sufficiently close to the image sensor (1): to be located outside the depth of field of the optical system optically linking the source (7) to the sensor (1), and such that, for at least one position of the drive support (2), the image obtained by the sensor (1) is entirely occupied by this source (7), which means that the source (7) covers the entire geometric extent of the camera (formed by the sensor 1 and the optical system optically linking the source (7) to the sensor (1)), i.e. the sensor (1) only intercepts light rays from the source (7) in the focal plane of the optical system optically linking the source (7) to the sensor (1).

[0062] The near-field source (7) is configured to present a uniform infrared scene to the image sensor (1). When placed in the near field, the source (7) at least temporarily obscures part of the panorama. It can be either passive or active.

[0063] We understand "active" to mean a thermostatically controlled source (7) and "passive" to mean a non-thermostatically controlled source (7).

[0064] For the purposes of non-uniformity correction, the near-field source (7) is inserted at intervals around the image sensor (1) so that it is seen at least once by the sensor (1) per panorama. If the source is fixed, it is then present in each panorama and obscures a portion of it.

[0065] The source (7) is advantageously retractable so that the image sensor (1) regains a useful horizontal field of view of 360° when rotating.

[0066] The image sensor 1 is connected to the optical system (not shown in the figure). The optical system optically connects the near-field infrared reference source (7) to the image sensor (1). Advantageously, the optical system is located between the near-field reference source (7) and the image sensor (1).

[0067] Ideally, the optical system forms the set of solid optical elements optically linking the near-field infrared reference source (7) to the image sensor (1).

[0068] The remaining elements optically linking the near-field infrared reference source (7) to the image sensor (1) consist only of gas, typically air.

[0069] According to the figures 1 and 2The system (100) also includes an infrared reference source, called the calibration source (8). The calibration source (8) is thermostatically controlled or equipped with means for measuring its temperature. At least one of the separate images of each panorama images this source (8), at least partially. The processing module (6) is arranged and / or programmed to calibrate the acquired images of the same panorama with respect to this calibration source (8). The calibration source (8) is typically a blackbody type and its dimensions are such that its image on the sensor (1) is large, meaning it represents a large number of pixels, for example, at least 10x10.

[0070] Source (8) is in the distant field.

[0071] The term "far-field source" refers to a source sufficiently distant so that it is within the depth of field of system (100). System (100) then sees the source (8) clearly.

[0072] The calibration source (8) is arranged to present the image sensor (1) with an apparent temperature of interest and is advantageously placed in a location that does not obstruct the panorama. This source (8) is active or passive, and separate from or combined with the source (7), or even optional.

[0073] We understand "active" to mean a thermostatically controlled source (8) and "passive" to mean a non-thermostatically controlled source (8).

[0074] The calibration source (8) is thermostatically controlled at a temperature above 33 degrees and / or below 40 degrees. Advantageously, the calibration source (8) is thermostatically controlled at a temperature close to that of the object (11) of interest. In the case of fever detection, the calibration source (8) is thermostatically controlled at approximately 37°, for example.

[0075] The system (100) (more precisely module 6) is arranged and / or programmed to correct the spatial non-uniformity of each distinct image composing a panorama by subtracting pixel by pixel, on each of the distinct images acquired for this panorama, the spatial non-uniformity of the reference image of this panorama imaging the non-uniformity correction source 7 and adding to it the average value of this non-uniformity.

[0076] In the panoramic image portions acquired by the sensor (1), there is a first reference area corresponding to the near reference source (7) and, optionally, the image of the scene in which the image of the calibration source (8) is located.

[0077] The near-field source (7) is positioned for recording by the sensor (1) and the module (6) of spatial non-uniformities in order to correct them. This periodic recording of the source (7) by the sensor (1) is used by the module (6) to eliminate residual non-uniformity in all portions of the panoramic image, each portion exhibiting the same residual non-uniformity pattern. The temporal evolution of this non-uniformity is slow compared to the presentation frequency of the near-field source (7). To limit the influence of temporal noise, the recorded non-uniformity is averaged over time by the module (6) before the module (6) performs the correction on all image portions. The portion of the panoramic image in which the source (7) is imaged is advantageously declared to the system by the module (6). The reference source (7) is visible in each panoramic image.

[0078] To avoid losing the continuous component of the signal, the non-uniformity is subtracted pixel by pixel from each image portion by module (6), and the average value of the non-uniformity is added. After stitching the image portions together by module (6), a quasi-panoramic image corrected for the internal noise within the image sensor (1), which is caused by the lens, is obtained by module (6). This is a quasi-real-time correction of a point in the image.

[0079] Depending on the variant considered: The near source (7) is not thermostatically controlled and its temperature is unknown, while the calibration source (8) is thermostatically controlled. Sources (7) and (8) are therefore distinct. The calibration source (8) is present in a portion of the panoramic image: its position is known by declaration to the system (100), as is its temperature. The apparent temperature of the calibration source (8) is chosen to be close to that of the objects (11) of interest, and in particular to an alarm threshold. The calibration source (8) is configured to correct any residual measurement bias after uniformity correction and improve detection accuracy by comparing the scene signal to the signal of the calibration source (8) to trigger an alarm; or the near source (7) is thermostatically controlled. The near source (7) thus fulfills the function of the calibration source (8).Sources (7) and (8) are therefore combined into a single source located at the location of reference (7) of the . figure 1 or 2 The near source (7) is part of the same hardware assembly as the image sensor (1). Source (7) then serves both as a non-uniformity correction source (7) and as a temperature measurement reference source (8). The system (100) therefore no longer requires a calibration source (8) and is thus easier to deploy. Alternatively, the near source (7) may not be thermostatically controlled, but its instantaneous temperature is known, and a dynamic calibration of the image sensor (1) is performed. The near source (7) thus fulfills the function of the calibration source (8). Sources (7) and (8) are therefore combined into a single source located at the reference point (7) of the image sensor (1). figure 1 If the gain is stable, such a system (100) is simpler to implement and eliminates the need for a heating, supply and control loop system.

[0080] The correction of atmospheric transmission effects is optionally handled by module (6) with a digital transmission model parameterized by temperature and humidity measurements accessible via sensors integrated into the image sensor (1). "Integrated sensors" refers to sensors such as thermometers, hygrometers, and / or weather stations.

[0081] The processing module (6) is configured and / or programmed to determine the distance between the object (11) and the image sensor (1) based on a dimension of the object (11) in one of the acquired images. The processing module (6) is also configured and / or programmed to determine the position of the object (11) within a field of an image acquired by the image sensor (1). Finally, the processing module (6) is configured and / or programmed to correct the temperature of the object (11) based on the determined distance between the object (11) and the image sensor (1) and / or based on the position of the object (11) within a field of an image acquired by the image sensor (1).

[0082] For the image sensor (1), the signal delivered from an object (11) with a fixed apparent temperature depends on the distance of the object (11) from the sensor (1) and the angular size of the object (11). The angular size of the object (11) is determined by calculating the number of pixels covered by the object (11). Knowing the dimensions of the object (11), the distance of the object (11) from the sensor (1) is then calculated. The signal from the object (11) is corrected based on its distance from the object (11). The correction performed by the module (6) is more or less refined to best match the calibration points. This ranges from a simple linear correction based on distance to a combination of more complex corrections, such as exponential or power functions.

[0083] It is possible to train a neural network implemented by module (6) to recognize object (11) and extract it from the distinct or panoramic image. The influence of atmospheric transmission on the temperature measurement depends on the ambient temperature and humidity. The use of a numerical model and real-time local measurements by system (100) adjusts this correction.

[0084] Furthermore, the signal level of the background visible around the areas of interest influences the temperature measurement. The correction based on the angular size of the object (11) is advantageously parameterized by the signal difference relative to the local background. An additional temperature correction of the object (11) based on the object's position within the field of view of the thermographic system (100) is implemented by module (6). This correction is applied to each image acquired by the sensor (1) of the scene, excluding the image taken from the nearby reference source (7), which is already corrected. This correction consists of adding to the detected objects (11) a portion of the thermal contrast between the object (11) and the background around the object (11) using a parabolic function centered on the image.The additional correction compensates for the fact that on some image sensors (1), the source does not have the same signal in the far field depending on whether it is placed in the center or at the periphery of the field, despite good non-uniformity correction in the near field.

[0085] The processing module (6) is configured and / or programmed to determine surface temperature data for the object (11). The system (100) further includes means for measuring the ambient temperature around the object (11). The processing module (6) is configured and / or programmed to determine an internal temperature of the object (11) based on the surface temperature data for the object (11) and the measured ambient temperature. The measuring means are typically thermometers and / or hygrometers.

[0086] The neural network implemented by module (6) is configured and / or programmed to recognize the human face of interest (11) in the scene. The size of a human face is known with sufficient accuracy to perform the correction (apparent size, distance) specified above. To determine if a person has a fever, it is necessary to be able to assess their core body temperature. However, only surface temperature is accessible to the infrared thermography system (100).

[0087] Module (6) models a difference between internal and external temperatures. The external temperature of an object (11), such as a face, depends in practice on the ambient air temperature. The ambient air temperature, measured locally by the image sensor (1), is used by module (6) to adjust for this difference.

[0088] The temperature of an object (11) is not uniform. For example, the face, eyes, nasal cavities, temples, the canthus of the eye, and an open mouth are warm, while the nose is rather cool. The warmest details have small angular dimensions. These details are resolved by the image sensor (1) with a near-field source (7) and / or with a far-field source (8).

[0089] Body temperature is determined by module (6) from temperature measurements on a non-uniform face.

[0090] The warmest parts of the face are considered visible, either in the current image or in subsequent images. The same person is potentially seen multiple times by the image sensor (1) as it is scanned, from different angles. Faces or heads are detected by the neural network implemented by module (6). The warmest area of ​​the face is recorded by module (6). To avoid excessive variations due to the resolution of warm details, module (6) performs a filtering by average or median among a proportion of the warmest pixels of the face. The face is also partially obscured by a hat, scarf, or mask in certain situations. In this case, the recorded temperature is lower than the temperature of the usual hot spots that are obscured here.By evaluating the temperature, through the application of a single correction (normally applied to visible hot spots), the assessment of the internal temperature is reduced.

[0091] The system (100) is arranged and / or programmed to track detected objects (11) from one image to the next, as the video stream progresses. Body temperature is correctly evaluated when the warmest parts of the face are presented to the image sensor (1), even intermittently during movement, for example.

[0092] The processing module (6) is arranged and / or programmed to determine different temperature values ​​of the object (11) over time and determine the maximum value among all the previously determined values.

[0093] For the same face tracked in the video stream by the sensor (1), the actual body temperature is the maximum value of a series of recent measurements. A series of recent measurements of the same object (11) tracked by the sensor (1) is stored in the module (6) in order to retain the maximum value for evaluating the internal temperature of the object (11).

[0094] The neural network implemented by module (6) is arranged and / or programmed to recognize the nature of the imaged facial portions and differentiate the presence or absence of covering elements on the face. To this end, the neural network is trained to detect these objects (11).

[0095] The discrepancies between the object's internal temperature (11) and its surface temperature are modeled separately. The correction performed by module (6) depends on the recognition process used to determine the correct body temperature. Regardless of the area detected by system (100), the body temperature is correctly evaluated.

[0096] The system (100) includes means for generating a visual or audible alert if the temperature of the object (11) is above a threshold temperature stored by the processing module (6), or a temperature deviation greater than a threshold deviation from the temperatures of other objects (11) imaged by the thermographic system (100).

[0097] The means for generating a visual or audible alert are arranged and / or programmed to filter out false alarms by comparing different measured temperatures of the object (11). The means for generating a visual or audible alert are typically of the type of displayed message indicating the presence of a feverish individual in a color of red, for example.

[0098] In the presence of a large flow of people, it is unlikely that a majority of individuals will have a fever. Therefore, only the differential measurements within a group are processed by module (6) using statistical modeling. The fever criterion is defined as an abnormally high temperature in one or a few individuals compared to a typical temperature: mean, median, or median / mean among the central values. However, it is not always possible to have a significant number of people simultaneously within the field of view. The statistical data obtained by system (100) over a rolling period are used.

[0099] A comparison between the signal on one face and that of another face seen by the image sensor (1) in the same portion of the field of view at a different time within a sliding period is possible. Variable relative infrared signal difference thresholds within the field are calculated by the module (6) and continuously updated based on measurements taken in the same area of ​​the field.

[0100] A hybrid mode is implemented, performing fever detection based on an absolute temperature measurement on the one hand and a relative measurement with respect to the typical local infrared signal on the other.

[0101] A confidence level for the local relative measurement, based on the local density of samples obtained over a rolling period, is calculated by module (6). If too few samples are obtained in an area, only the absolute measurement is used in that area. When both absolute and relative modes are available, they are combined by relaxing the individual detection thresholds to optimize a good detection rate while limiting the false positive rate.

[0102] The processing module (6) is arranged and / or programmed to eliminate hot spots on the object (11) and / or background surrounding the object (11) from the images acquired by the image sensor (1) before determining the temperature data of said object (11).

[0103] The image is cleaned by module (6) to remove hot spots by superimposing an element with a higher temperature, such as a hot coffee cup, onto the object (11) of interest. The background surrounding the face can be removed by positioning a background or an object (11) with a higher temperature around the object (11) of interest, such as a face. Module (6) then analyzes the histogram and determines the temperature of the object (11) by averaging the percentages of the highest points in the cleaned image.

[0104] However, infrared imaging does not allow for the identification of targeted individuals during face detection. For example, if a person is detected by the sensor (1) in the middle of a group, it is difficult to identify them and approach them. It is very useful to associate an image in the visible spectrum with the image.

[0105] The system (100) includes at least one visible light image sensor (i.e. arranged to capture and image at least one radiation, preferably all radiations, having a wavelength between 450 nm and 750 nm), color or black and white, arranged to capture the object (11) simultaneously with the infrared image sensor (1).

[0106] The at least one visible light image sensor (color or black and white) comprises a ring of 15 fixed visible image sensors (preferably high resolution) observing a wide panorama of at least 180° and preferably up to 360°, and is associated with the image sensor (1). The stream is continuously recorded by the sensors over a sliding time period at least longer than the delay between the infrared image capture and the end of information processing. Knowing precisely the time of the infrared and visible image captures, the object (11) detected in the infrared channel is also detected by the module (6) in the recent visible recording, the image being captured in the same direction, at the same time, and for the same object (11). Identification by the module (6) with the visible channel is performed following, for example, the detection of an abnormal temperature in the infrared channel.

[0107] The visible light stream from the visible light image sensor (color or black and white) is displayed in real time by the control and communication systems (9). The operator finds it easier to orient themselves in the visible stream than in the infrared. The visible image has a faster refresh rate than the infrared image. The infrared image provided by fixed sensors has a frequency greater than 10 Hz, whereas the panoramic infrared image formed by rotating a sensor has a more limited rate (on the order of Hertz).

[0108] When the system (100) detects an object (11) with a precise time, it associates it with the track where the object (11) was near the infrared detection point at the same time. The temperature and fever status are then directly associated and displayed by the control and communication means (9) in the panoramic image visible next to the detected moving faces, for example.

[0109] To achieve optimal detection results with the neural network implemented by module (6), image preparation is performed by module (6). The acquired images have a dynamic range of fourteen bits, but only a portion of this range is used locally. Neural networks function best when the full dynamic range is utilized. Therefore, a tone-mapping process implemented by module (6) reduces the dynamic range to eight bits and improves its utilization. Temporal and / or temperature information is used by module (6) to maximize contrast on the objects (11) of interest. In the case where the objects (11) are faces, they are distinguished from the background by the fact that they are not stationary and / or are at a different temperature.If the temperature is estimated from the intensity of a pixel, an intensity interval of fourteen bits of interest [m, M] (m and M being parameters of the algorithm which are numbers which define the interval of expected intensities of the objects: m being the minimum value of the interval and M being the maximum value) which contain the expected intensities of the objects (11) is defined, and by affine conversion projects this fourteen-bit interval onto an eight-bit interval, for example [102, 230] (The notation [a, b] here corresponds to the definition of an interval between a and b, in this case the interval of values ​​between 102 and 230, the interval being therefore included in that of [0, 255] of the 8-bit values). The rest of the image is added optionally, by converting by affine transformation the interval described by the minimum in a neighborhood of the pixel and m to [0, 102], and the interval described by M and the maximum in a neighborhood of the pixel to [230, 255].The conversion is thus piecewise affine.

[0110] Alternatively, a background image is determined by module (6), for example, by taking the median of each pixel across a series of images taken at certain time intervals and updated regularly. The background image, if well-formed, should not contain any objects (11) of interest. Furthermore, it is assumed that the background is cooler than the objects (11), at least in areas where they might be visible. This background image is used by module (6) to determine a transformation of the current distinct image that will both reveal the background and highlight areas that are locally warmer than the background. This implementation involves module (6) calculating the background image Im, for example, by taking the median of each pixel across a series of distinct images taken at certain time intervals.A low-resolution version L of this image is then constructed by module (6) using a Gaussian pyramid. This L is resized to the original size and subtracted from the background image Im to obtain image H. This removes low-frequency information from the original image and compresses the dynamic range. The 1% and 90% quantiles of image H are then calculated by module (6). To be more robust, the asymmetry is adjusted to account for the presence of hot objects (11) in the background. Images m and M are then calculated by adding the previously calculated quantiles of H to image L, respectively. Finally, for the current image, for each pixel, if the intensity is between the minimum in a neighborhood of the image and m, we convert the intensity using an affine transformation in the interval [0, 25].Similarly, we convert an intensity in [m, M] to the interval [25, 102] using an affine transformation, and the interval [M, maximum in a neighborhood] to [102, 255]. Before being provided to the neural network implemented by module (6), the image is recentered around 0.

[0111] The images were previously calibrated in intensity using the systems and / or methods described in the present invention.

[0112] We will now describe an infrared thermographic method for temperature detection of the object (11), implemented by the system (100) comprising a collection of infrared radiation by the thermal infrared image sensor (1) and the construction of at least one image from this radiation. The method includes rotating the sensor (1) around the axis of rotation (10) by the drive support (2), the image sensor (1) being fixed to the drive support (2) such that, during the rotation of the drive support (2), the sensor (1) images different distinct areas surrounding the thermographic system comprising said sensor (1) and said drive support (2), the image sensor (1) acquiring several distinct images such that the combination of these different images forms a continuous panorama of at least 180 degrees around the axis of rotation of the drive support (2).The process also includes a determination by the processing module (6) of temperature data of said object (11) from the images acquired by the image sensor (1).

[0113] At least one of the distinct images in each panorama is captured by the near-field infrared reference source (7), referred to as the non-uniformity correction source. The processing module (6) corrects any spatial non-uniformity in the images acquired by the sensor (1) according to the non-uniformity correction source.

[0114] The non-uniformity correction source is located at a distance from the image sensor (1) such that at least one acquired image, called the reference image, only images the non-uniformity correction source.

[0115] The processing module (6) corrects spatial non-uniformity by subtracting, pixel by pixel, on each of the acquired images, the spatial non-uniformity of the reference image imaging the non-uniformity correction source and adding to it the average value of this non-uniformity.

[0116] The process also includes temperature control (the calibration source (8) is preferably temperature-controlled at a temperature above 33 degrees and / or below 40 degrees) and / or a temperature measurement of the infrared reference source, referred to as the calibration source (8), as well as at least one of the separate images from each panorama imaging this calibration source (8). The processing module (6) calibrates the acquired images of the same panorama for temperature based on this calibration source.

[0117] The processing module (6) determines a distance between the object (11) and the image sensor (1) based on the dimensions of the object (11) in one of the acquired images and / or determines the position of the object (11) within a field of an image acquired by the image sensor (1). The processing module (6) corrects the temperature of the object (11) based on the determined distance between the object (11) and the image sensor (1) and / or based on the position of the object (11) within a field of an image acquired by the image sensor (1).

[0118] The processing module (6) determines surface temperature data for said object (11). The method further includes an ambient temperature measurement around the object (11), the processing module (6) determining an internal temperature of the object (11) as a function of the surface temperature data for said object (11) and the measured ambient temperature.

[0119] In addition, the processing module (6) determines different temperature values ​​of the object (11) over time and determines the maximum value among all the previously determined values.

[0120] The method includes generating a visual or audible alert if the temperature of the object (11) exceeds a threshold temperature stored by the processing module (6). The method also includes generating a visual or audible alert if the temperature of the object (11) deviates from a threshold temperature relative to the temperatures of other objects (11) imaged by the thermographic system (100).

[0121] The generation of the visual or audible alert filters out a false alarm by comparing different measured temperatures of the object (11).

[0122] The method includes capturing, by at least one visible light image sensor (color or black and white), the object (11) simultaneously with the infrared image sensor (1).

[0123] The processing module (6) eliminates hot spots on the object (11), and / or a background surrounding the object (11) from the images acquired by the image sensor (1) before determining the temperature data of said object (11).

[0124] The system (100) includes a 640 by 480 element thermal infrared microbolometer image sensor (1) equipped with a 25mm focal length f / 1.1 aperture lens, model TAU 2.7 640 supplied by FLIR Systems.

[0125] The horizontal field of view of this sensor (1) is approximately 25°. The vertical field of view is approximately 20°, which, after geometric corrections, allows us to obtain the image of a panoramic strip with a height of 18°.

[0126] The optical axis can be manually indexed in elevation to positions of 0° and ±6° relative to the horizontal, allowing for rapid adjustment of the viewing angle to suit the installation situation: horizon, sky, or ground aiming, for example. If the positions are unsuitable, continuous indexing is possible, up to ±18°. The advantage of indexed positions is that they allow the image sensor orientation (1) to be defined with sufficient precision to apply acceptable geometric corrections without additional calibration.

[0127] The system (100) also includes the drive support (2) arranged to drive the sensor (1) in rotation around a rotation axis (10).

[0128] The drive motor (4) is a compact, low-power, high-efficiency, high-torque brushless servomotor, model TC 40 0.32 from the manufacturer's MPC range, equipped with a high-resolution resolver. The choice of resolver is dictated by the need for high angular resolution, compatible with the expected positioning accuracy. The motor control electronics consist of an ELMO WHISTLE module with a resolver input (10 to 15 bits).

[0129] Sixteen images of successive angular sectors are captured by the sensor (1) and constitute a 360° infrared panoramic image. The panoramic image has a field of view of 360° by 18°. Its format is 16 by 640 columns by 512 rows, which is equal to 10240 columns by 512 rows, or 5,242,880 pixels. Its spatial resolution is on the order of 0.6 mrad (18° / 512).

[0130] The panoramic image refresh rate of sensor (1) is 0.44 Hz, resulting in a new 360° image every 2.3 seconds. This rate is limited by the frame rate of the infrared microbolometer image sensor (1), which is capped at 7.5 Hz for exportability reasons. The 7.5 Hz frame rate corresponds to an image period of 1000 / 7.5, or 133 ms per angular sector, to be divided between the imaging and jump phases. The time required for the imaging phase is at least equal to the thermal response time of the microbolometer detector, i.e., 3τ, which corresponds to 36 ms, plus the readout time of 33 ms, for a total minimum of 69 ms.

[0131] A dynamic adjustment of the sensor's movement (1) is initially performed to test the system's response limits (100) A jump phase is performed in about 40ms, An imaging phase therefore lasts up to about 90ms (i.e. 133-40), which is more than the 69ms required for the formation and reading of the image on the microbolometer detector, The return phase lasts about 120ms for an almost complete cycle.

[0132] For this setting, the average power consumption of system (100) remains very low, on the order of 5W. System (100) is configured to capture a panoramic image in 1.6s, or at 0.62Hz for a complete panorama.

[0133] In practice, to further limit consumption and reduce the inertia required for the fixed part, the sequence of phases is linked as follows, in the case of imaging a complete 360° panorama: Duration of a jump phase: Ts = 57 ms Duration of an imaging phase: Ti = 76 ms (= 133 - 57) Duration of the return phase: Tr = 190 ms

[0134] The 190ms duration corresponds to the loss of a complete image during the return phase (57+76+57=190). Under these conditions, the average power consumption of the system (100) is approximately 4W. The analysis time for a complete 360° panorama by module (6) is 17 times 133, or 2261s, which corresponds to a frame rate of 0.44Hz for a complete panorama. This power consumption is compatible with the power supply capabilities of the integrated and embedded equipment.

[0135] The system (100) is designed to switch from a panoramic mode to a narrow field of view but high frame rate mode (or "staring"). This is achieved by orienting the image sensor (1) in the desired direction of azimuth and maintaining a constant rotation angle. The frame rate is then increased to the maximum rate of the image sensor (1), which is 7.5 to 30 Hz depending on the image sensor model chosen.

[0136] It is advantageous to take advantage of the high frame rate to lock the tracking onto a target (11) of interest in the center of the field using a conventional image correlation algorithm. This design operates in intermediate modes and monitors a useful area only with a scan reduced to a few sectors that do not represent 360°. It is possible to electronically adjust the sensor's (1) scan origin to further optimize utilization and coverage.

[0137] The system (100) is compatible with the use of visible or near-infrared image sensors, provided the port is adapted. In this case, the time required for image acquisition by the sensor (1) is no longer subject to the same constraints as for a microbolometer detector, since it is reduced to a few milliseconds. It is therefore possible to achieve frame rates on the order of 1 Hz for a complete panoramic image.

[0138] The system (100) includes the near-field reference source (7), consisting of a passive or active mask depending on the variant. The polyamide mask is painted with a matte black paint to provide good infrared emissivity. This source (7) completely masks the first of the sixteen sectors (also called the "distinct image") scanned by the image sensor (1), so the usable panoramic image is at most 337.5°. Depending on the variant, a distant calibration source (8) at 37°C, developed by HGH under the name CN37, is placed a few meters away from the image sensor (1) to cover at least 10 by 10 pixels.

[0139] The thermal microbolometer image sensor (1) is implemented with its factory calibration taking into account ambient parameters such as temperature for example and a one-point correction on the internal shutter.

[0140] All the calculations and signal processing described below are carried out by module (6).

[0141] The image sensor (1) therefore acquires sixteen sector images in succession: the first is the reference image (image of the source 7) which is used for additional offset correction; the other fifteen follow and display the panorama. The image of the first sector is used to determine the additional offset correction to be applied. This image has the following values: S 0 m n The following images have the following values: S i m n , i ∈ 1 , 15 . The images corrected for the additional offset become: Sc i m n = Sc i m n − S 0 m n + S 0 m n ¯ i ∈ 1 , 15

[0142] To limit the impact of temporal noise S 0 on the correction, a moving time average of: S 0 The average is calculated over, for example, ten images before applying the additional correction, as the fixed spatial noise evolves slowly over time. The first sector, once corrected, should be almost uniform, meaning that its standard deviation is almost equal to the temporal noise. Adding the average to the additional correction prevents the loss of the continuous component in the processing. Once applied, the correction is perfect on a scene at the ambient temperature of the nearby reference source (7). The advantage of this correction is that it is performed in near real-time and on a reference source external to the system (100).

[0143] In one of the sector images acquired by the sensor (1), the image of the calibration source (8) appears. It is sufficiently large to allow for spatial averaging, which is then combined with a temporal averaging to eliminate the temporal noise component before subtracting it from the image signal. The response of the image sensor (1) and the apparent temperature of the source (8) are sufficiently stable over time to be considered constant and averaged to reduce temporal noise. The average level recorded is compared to any point in the image to determine the temperature of that point. To do this, the system (100) knows the position of the calibration source (8) in the stable image from one panoramic image to the next, and the temperature of the source, which is also stable.

[0144] To deduce the temperature of each point in the image, the response of the image sensor (1) must be calibrated around the temperature of interest, which is the temperature of the calibration source (8). Depending on the application, it is necessary to calibrate the image sensor (1) at the factory by presenting it with a blackbody whose temperature varies from ambient to several hundred degrees Celsius. For a fever measurement application, calibration of a few points, for example two or three, around the reference temperature and a linear relationship are sufficient. In this latter case, the image sensor (1) exhibits a gain in digital levels per Kelvin around the reference temperature. The module (6) is configured and / or programmed to calculate the temperature of each pixel as a function of the temperature of the reference source and the signal difference between the pixel and the source: T i m n = Sc i m n − S ref k l t 1 t 2 ¯ / Gain + Tref i ∈ 1 , 15

[0145] In the variant where there is no calibration source (8), the nearby source (7) is thermostatically controlled or of known temperature and acts as the reference source 8. The following calculation applies: T i m n = Sc i m n − S ref m n t 1 t 2 ¯ / Gain + Tref i ∈ 1 , 15

[0146] In the variant where there is no calibration source (8) or thermostatically controlled source, prior calibration of the image sensor (1) is required to perform the calculation: T i m n = Sc i m n − S ref ¯ / Gain + Tref i ∈ 1 , 15

[0147] The measured temperature of an object (11) depends on the apparent angular size of that object (11) and its distance from the image sensor (1). The signal is expressed by the following formula: Signal DL = Signal 0 DL × SourceAngulareSize SourceAngularSize 0 Power × e − Sigma × distance − distance 0

[0148] More simply, for an object (11) of known size, for example a face, an affine distance correction is applied to the signal of the pixels whose distance to the image sensor (1) is known before the temperature conversion: T i m n = Sc i m n − S ref k l t 1 t 2 ¯ + a ∗ D i m n − D ref + b Gain + Tref i ∈ 1 , 15

[0149] In the variant where there is no calibration source (8), the nearby source (7) is thermostatically controlled and acts as the reference source. The following calculation applies: T i m n = Sc i m n − S ref m n t 1 t 2 ¯ + a ∗ D i m n + b / Gain + Tref i ∈ 1 , 15

[0150] According to the variant where there is no calibration source (8) or thermostatically controlled source, prior calibration of the image sensor (1) is required to perform the calculation: T i m n = Sc i m n − S ref ¯ + a ∗ D i m n + b / Gain + Tref i ∈ 1 15

[0151] To determine the distance of the object (11) of interest from the sensor (1), it must be detected within the panoramic image. A neural network implemented by module (6) is used to learn to recognize objects (11) of interest; here, we will choose faces in the infrared. It is common to do this in the visible spectrum. Detection is performed in the infrared to isolate the presence of faces in the panoramic image. The network is pre-trained on visible images converted to grayscale and refined on annotated infrared images. The network is learned and then applied by module (6) to sectors of the image. The results are combined to obtain detection across the entire panoramic image.

[0152] The neural network thus learned to detect and isolate areas corresponding to faces in the panoramic image. The size of an area allows the distance to the face to be determined, assuming, for example, that the person is of average height 1.75 meters and that the width of the face is, for example, 13 percent of the subject's height. The distance correction is applied by module (6) to the pixels of the area corresponding to the face, and the temperature of each pixel is determined.

[0153] A single-stage anchor-free detection network is used, and more specifically a training-time-friendly network (TTFNet). (paper: https: / / arxiv.org / abs / 1909.00700Detection networks are typically single-stage or two-stage. Among single-stage networks, anchor-based networks are distinguished, which specialize several detection heads for fixed sizes and shapes of target objects (each head is assigned a rectangle—or "anchor box"—which can be interpreted as the typical size and shape of an object for that detection head). Anchor-free networks do not perform this specialization, and recent work such as "Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection," CVPR 2020, Zhang et al., demonstrates that they achieve performance identical to anchor-based networks.

[0154] In terms of network architecture, the head end of the network is the part of the network that is not part of the core network (or backbone). The core network operates in grayscale; it takes a single-channel image as input, instead of a color image.

[0155] The network is trained with the ADAM learning algorithm ("Adam: A method for stochastic optimization", 2014, Diederik Kingma and Jimmy Ba). The learning rate is halved as soon as the training score does not improve over three consecutive periods.

[0156] A dataset augmentation adapted to shades of gray and infrared in particular is used. Regarding the visual perturbation of the inputs, horizontal rotations, scaling changes, and translations are performed as is standard practice in detection networks. In addition, a five-point piecewise linear histogram transformation is applied, pulled uniformly over [0,255]. The purpose of this dataset augmentation step is to promote greater insensitivity to the contrast and light intensity level of the scene and target objects (11) for the network. Finally, Gaussian noise with a standard deviation pulled uniformly over [0,40] is added.

[0157] The object detection network (11) is first trained on visible data in shades of gray, then on infrared data. The network learns better because there is more visible data. Since infrared data has a dynamic range of fourteen bits, it is not directly used by the network. Neural networks frequently use images whose available dynamic range is normalized between -1 and 1 or -127.5 and 127.5, for example. The normalized dynamic range between -127.5 and 127.5 is used. A tone-mapping process implemented by module (6) is used to reduce the dynamic range to eight bits, while making the best use of the available dynamic range.

[0158] The fast training network (or TTFNet) is adapted to meet various detection needs. The loss function is modified to perform well on different target image sizes. Without modification, the ratio between the parts that make up the loss function depends on the size of the network's input and the number of classes. The loss function is also modified so that during training, the network tries equally to reduce the number of false positives in images with many or few targets, while the fast training network (or TTFNet) divides the penalty by the number of targets.

[0159] Furthermore, a term was added to the loss function to more heavily penalize the highest object presence score (11) per class and image of the mini-batch in areas without objects (11). This has the effect of slightly reducing the presence of false detections.

[0160] The loss function, using the notation of the fast-training network (or TTFNet) https: / / arxiv.org / abs / 1909.00700 ), is therefore: L = L pos + L neg + L neg 2 + 5 L reg

[0161] Either B the size of the batch, and N p the size of the mini-batch heatmap N p = B . W . H r 2 , SO L pos = − 1 M ∑ bijc 1 − H bijc ^ 2 log H bijc ^ 1 H bijc = = 1 L neg = − 1 e 5 N p . c ∑ bijc 1 − H bijc 4 H bijc ^ 2 log 1 − H bijc ^ 1 H bijc < 1 WH corresponds to the size of the image analyzed during training and subsequently for detection (identical to the size of a sector). R corresponds to the "output stride" factor, i.e., the detection step size set to 4. As a reminder, M is the number of objects (11) present in the images of the mini-batch. M = ∑ bijc 1 H bijc = = 1 b is the index on the batch. To link this to the fast training network (or TTFNet), L loc = L pos + N p . c 1 e 5 1 M L neg L_pos encourages the network to predict 1 on the heat map if an object (11) is present. The present invention divides the contribution of each relevant pixel by the number of objects (11). L _ neg encourages the network to predict 0 on the heatmap if an object (11) is not present (using the technique, introduced in "Cornernet: Detecting objects as paired keypoints", ECCV 2018, Hei Law and Jia Deng, to reduce the penalty around the centers of objects (11) to account for the imprecision in the definition of the center of objects (11)). We normalize by the number of pixels in the heatmap where 0 must be predicted. The heatmap has c channels (one channel per class) and N p pixels, we normalize by N p . c . M is subtracted, but M is very negligible compared to N p Finally, we multiply by one hundred thousand in order to give much greater importance to L neg compared to L pos This ratio results in the optimal detection threshold of the final network being approximately 0.20. Without these modifications, the ratio between L pos And L neg depends on the number of pixels in the image, and therefore the optimal threshold depends on the size of the image, as well as the number of classes.

[0162] Finally L neg 2 , which contributes slightly to the loss function, is defined by: L neg 2 = 1 B . c ∑ bc MAX ij 1 − H bijc 4 H bιjc ^ 2 log 1 − H bιjc ^ 1 H bijc < 1

[0163] Of course, the invention is not limited to the examples just described and many modifications can be made to these examples without departing from the scope of the invention.

[0164] Of course, the various features, forms, variants, and embodiments of the invention can be combined in various ways, provided they are not incompatible or mutually exclusive. In particular, all the variants and embodiments described above are combinable.

Claims

1. Infrared thermographic system configured to measure the temperature of an object (11) comprising: - a thermal infrared image sensor (1) arranged to collect infrared radiation and construct at least one image based on this radiation, - a drive support (2) arranged to drive the sensor in rotation about an axis of rotation, the image sensor (1) being fastened on the drive support (2) such that, during a rotation of the drive support (2), the sensor images different distinct areas surrounding the thermographic system, the image sensor (1) being arranged to acquire several distinct images such that combining these different images forms a continuous panorama of at least 180 degrees about the axis of rotation of the drive support (2), - a processing module (6) arranged and / or programmed to measure a temperature of said object (11) based on the images acquired by the image sensor (1), characterized in that it comprises a near-field infrared reference source (7) called non-uniformity correction source and an optical system optically connecting the near-field infrared reference source (7) to the image sensor (1), the drive support (2) being arranged to drive the sensor in rotation about an axis of rotation, the image sensor (1) being fastened on the drive support (2) such that at least one of the distinct images of each panorama images the non-uniformity correction source, the processing module (6) being arranged and / or programmed to correct a spatial non-uniformity of the acquired images as a function of the non-uniformity correction source, and characterized in that the non-uniformity correction source is situated at a distance from the image sensor (1) such that at least one acquired image, called reference image, images only the non-uniformity correction source, and characterized in that the near-field infrared reference source (7) is close enough to the image sensor (1) to be situated outside the depth of field of the optical system optically connecting the source (7) to the sensor (1), such that, for at least one position of the drive support (2), the sensor (1) is arranged to intercept only light radiation originating from the source (7) in the focal plane of the optical system optically connecting the source (7) to the sensor (1).

2. Infrared thermographic system according to claim 1, characterized in that the optical system is situated between the near-field reference source (7) and the image sensor (1).

3. Infrared thermographic system according to any one of claims 1 to 2, characterized in that the processing module (6) is arranged and / or programmed to correct the spatial non-uniformity by pixel to pixel subtraction, on each of the acquired images, of the spatial non-uniformity of the reference image imaging the non-uniformity correction source.

4. Infrared thermographic system according to the any one of the preceding claims, characterized in that it comprises an infrared reference source, called calibration source (8), thermostatically controlled or equipped with means for measuring its temperature, at least one of the distinct images of each panorama imaging this calibration source, the processing module (6) being arranged and / or programmed for temperature calibration of the acquired images of one and the same panorama as a function of this calibration source.

5. Infrared thermographic system according to claim 4, characterized in that the calibration source is thermostatically controlled to a temperature above 33 degrees and / or below 40 degrees.

6. Infrared thermographic system according to any one of the preceding claims, characterized in that the processing module (6) is arranged and / or programmed to: - determine a distance between the object (11) and the image sensor (1) as a function of a dimension of the object (11) on one of the acquired images and / or determine a position of the object (11) in a field of an image acquired by the image sensor (1), - correct the temperature of the object (11) as a function of the determined distance between the object (11) and the image sensor and / or as a function of the position of the object (11) in a field of an image acquired by the image sensor (1).

7. Infrared thermographic system according to any one of the preceding claims, characterized in that it comprises at least one visible light image sensor (1), arranged to capture the object (11) simultaneously with the infrared image sensor (1).

8. Infrared thermographic system according to any one of the preceding claims, characterized in that the processing module (6) is arranged and / or programmed to measure a surface temperature of said object (11), the system further comprising means for measuring ambient temperature around the object (11), the processing module (6) being arranged and / or programmed to measure an internal temperature of the object (11) as a function of the surface temperature of said object (11) and of the measured ambient temperature.

9. Infrared thermographic system according to any one of the preceding claims, characterized in that the processing module (6) is arranged and / or programmed to determine different temperature values of the object (11) over time and to determine the maximum value among all the previously determined values.

10. Infrared thermographic system according to any one of the preceding claims, characterized in that it comprises means for generating a visual or acoustic warning signal if the temperature of the object (11): - is above a threshold temperature stored by the processing module (6), or - has a temperature difference greater than a threshold difference with respect to temperatures of other objects (11) imaged by the thermographic system.

11. Infrared thermographic system according to claim 10, characterized in that the means for generating a visual or acoustic warning signal are arranged and / or programmed to filter out a false warning signal by comparing different measured temperatures of the object (11).

12. Infrared thermographic system according to any one of the preceding claims, characterized in that the processing module (6) is arranged and / or programmed to eliminate, on images acquired by the image sensor (1): - hot spots on the object (11), and / or - a background surrounding the object (11) before measuring a temperature of said object (11).

13. Infrared thermographic method for measuring the temperature of an object (11) comprising: - collecting infrared radiation by means of a thermal infrared image sensor (1) equipped with an optical system and constructing at least one image based on this radiation, - driving the sensor in rotation about an axis of rotation by means of a drive support (2), the image sensor (1) being fastened on the drive support (2) such that, during the rotation of the drive support (2), the sensor images different distinct areas surrounding a thermographic system comprising said sensor and said drive support, the image sensor (1) acquiring several distinct images such that combining these different images forms a continuous panorama of at least 180 degrees about the axis of rotation of the drive support (2), - measuring, by a processing module (6), the temperature of said object (11) based on images acquired by the image sensor (1), at least one of the distinct images of each panorama imaging a near-field infrared reference source (7) called non-uniformity correction source, the processing module (6) correcting a spatial non-uniformity of the acquired images as a function of the non-uniformity correction source, the non-uniformity correction source being situated at a distance from the image sensors (1) such that at least one acquired image, called reference image, images only the non-uniformity correction source.

14. Infrared thermographic method according to claim 13, characterized in that the processing module (6) corrects the spatial non-uniformity by pixel to pixel subtraction, on each of the acquired images, of the spatial non-uniformity of the reference image imaging the non-uniformity correction source.

15. Infrared thermographic method according to any one of claims 13 to 14, characterized in that the processing module (6): - determines a distance between the object (11) and the image sensor (1) as a function of a dimension of the object (11) on one of the acquired images and / or determines a position of the object (11) in a field of an image acquired by the image sensor (1), - corrects the temperature of the object (11) as a function of the determined distance between the object (11) and the image sensor and / or as a function of the position of the object (11) in a field of an image acquired by the image sensor (1).

16. Infrared thermographic method according to any one of claims 13 to 15, characterized in that the processing module (6) eliminates on images acquired by the image sensor (1): - hot spots on the object (11), and / or - a background surrounding the object (11) before measuring a temperature of said object (11).