Method and device for calibrating an imaging device
The calibration process employs pyramidal decomposition and high-frequency analysis to efficiently correct spatial noise in photodetector matrices, updating offset parameters dynamically to enhance image quality in imaging devices.
Patent Information
- Application Number
- FR2023004572
- Authority / Receiving Office
- FR · FR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2023-05-11
- Publication Date
- 2025-04-25
- Estimated Expiration
- 2043-05-11
AI Technical Summary
Existing calibration processes for photodetector matrices in imaging devices, such as infrared cameras, are inefficient in correcting spatial noise like 'Sunburn' defects and require extensive image capture, leading to interruptions in video flow and residual defects due to time-invariant offset tables and temperature-based corrections.
A calibration process using pyramidal decomposition to determine offset parameters through high-frequency analysis during relative movement, updating the correction base with elementary supplements of offset parameters, and iteratively refining these parameters to correct spatial defects efficiently.
The process quickly corrects spatial noise in a reduced time frame, optimizing image quality by dynamically updating offset parameters, particularly for infrared cameras, with minimal disruption to the video flow.
Smart Images
Figure 00000017_0000 
Figure 00000017_0001 
Figure 00000018_0000
Abstract
Description
Title of the invention: Method and device for calibrating an imaging device Technical field
[0001] The invention relates to the field of imaging and more particularly relates to the field of calibrating a photodetector array of an imaging device, for example, a camera. A photodetector array is also known to those skilled in the art under the designation of sensor.
[0002] In a known manner, a matrix of photodetectors provides a set of response signals, designated pixels, forming an image. A succession of images advantageously forms a video stream.
[0003] In practice, in a photodetector array, the pixel value of each photodetector is not uniform at the same input optical intensity natively. This is particularly the case when the photodetector array belongs to an uncooled infrared camera, also called LWIR for "LongWave InfraRed" meaning long wavelength infrared.
[0004] To obtain uniform pixel values for the same optical intensity, it is necessary to correct the value of the pixels of each photodetector by applying a gain parameter or an offset parameter called “offset”.
[0005] Gain correction is conventionally performed using a gain table, in which each photodetector is associated with a gain parameter. Such a gain table depends essentially on the input optics of the imaging device and remains substantially constant throughout the life of the imaging device. In practice, the gain table is estimated only once following assembly of the imaging device, for example, at the factory.
[0006] On the other hand, the offset parameter specific to each photodetector depends on many factors that change over time, including ambient temperature and aging. Consequently, the use of a time-invariant offset table is not appropriate.
[0007] To eliminate this drawback, it has been proposed to carry out, regularly over time, a so-called "one-point" calibration. Such a one-point calibration consists of mechanically generating, in particular by means of a cache or defocusing of the input optics of the imaging device, a homogeneous scene in order to estimate the offset error of each photodetector. A new offset table, leading to a corrected image that is also homogeneous, is then deduced. In a In such a case, the resulting offset table is used without modification until the next calibration.
[0008] Furthermore, when the imaging device, or its usage protocol, does not allow the creation of a homogeneous scene, it is known to perform a real-time calibration using information relating to a temperature of the sensor. Such a real-time calibration, also called photometric non-uniformity correction, consists of modeling the offset defect by a polynomial of the temperature of the sensor, for example a polynomial of degree 3, the temperature being measured by a thermal sensor positioned close to the sensor.
[0009] However, such calibration methods are not entirely satisfactory. Indeed, single-point calibration results in a cutoff of the video stream delivered by the imaging device each time it is implemented. Furthermore, in certain use cases, such calibration retains a residual defect. Furthermore, the photometric non-uniformity correction only makes it possible to correct the contribution of temperature to the offset defect.
[0010] A photodetector array is furthermore subject to a sunburn defect, known by its English name "sunburn", which corresponds to a remanence of the sun. This type of defect is similar to a defect in the offset of the values of the response signals of the photodetectors.
[0011] Document FR3118558A1 discloses a method for calibrating an imaging device in which high-frequency filtering is used to determine a defect detectable via movement of the imaging device.
[0012] Such a calibration algorithm is not very efficient, in particular because it requires a very large number of images (several tens of thousands) to enable convergence towards a correction value and therefore a significant amount of time.
[0013] An aim of the invention is therefore to propose a calibration method which makes it possible to correct spatial noise in a reduced time, in particular, a defect of the “sunburn” type. PRESENTATION OF THE INVENTION
[0014] The invention relates to a method for calibrating an imaging device configured to observe a scene, the imaging device comprising an array of photodetectors configured to capture a raw image, the raw image comprising a plurality of response signals, each photodetector being configured to provide a response signal as a function of the optical intensity of the scene, the method having access to a correction base comprising an offset parameter for the array of photodetectors.
[0015] The method is remarkable in that it comprises steps consisting of: • Correct the raw image from the correction base in order to obtain a corrected image, • Upon relative movement of the imaging device with respect to the scene, calculating an offset parameter complement for the photodetector array, the calculation comprising steps consisting of: • Decompose the corrected image in a pyramidal manner into a plurality of elementary spatial resolutions, • Determine, for each elementary spatial resolution, a plurality of high frequencies, • Statistically determine, for each elementary spatial resolution, an elementary complement of the offset parameter determined from the plurality of high frequencies, • Determine, for the photodetector matrix, the offset parameter complement from the elementary offset parameter complements obtained for each elementary spatial resolution, • Update the photodetector array offset parameter with the offset parameter complement.
[0016] Thanks to the invention, advantage is taken of the pyramidal decomposition to determine several frequency bands and thus determine an offset parameter complement from a plurality of elementary offset parameter complements. Each image is thus used in oversampling to determine the characteristics of an offset defect. This makes it possible to correct an offset parameter very quickly with a reduced number of corrected images. The correction is thus very reactive. This is particularly advantageous for correcting a “sunburst” type defect.
[0017] According to one aspect, the calibration method comprises a step of measuring a quantity of movement in the corrected image and stopping the calibration method if the measured quantity of movement is less than a predetermined movement threshold. A movement makes it possible to highlight an offset defect. Preferably, the quantity of movement is determined by comparison with a previous corrected image.
[0018] According to one aspect, the elementary complement of the shift parameter comprising a plurality of coefficients respectively associated with a response signal, the method comprises a step consisting of comparing the high frequencies of a response signal, in absolute value, with a predetermined high threshold and determining an elementary complement coefficient of the shift parameter of zero value in the event of overshoot for said response signal. This makes it possible to remove the edges of objects in the scene which would disrupt the highlighting of the defect.
[0019] By processing each coefficient individually, a correction specific to each photodetector is determined.
[0020] According to one aspect, the elementary complement of the shift parameter comprising a plurality of coefficients respectively associated with a response signal, the method comprises a step of multiplying by a convergence rate constant, between 0 and 1, the elementary complement coefficient of the shift parameter for said response signal so as to average the elementary complement coefficient of the shift parameter for said response signal over time. This convergence rate constant makes it possible to define a compromise between the precision and the desired convergence speed. A value close to 1 accelerates the convergence while a value close to 0 makes the estimation more precise.
[0021] According to one aspect, the elementary complement of the shift parameter comprising a plurality of coefficients respectively associated with a response signal, the method comprises a step of saturating the elementary complement coefficient of the shift parameter for said response signal so as to limit the variations of the shift parameter for said response signal over time. A high saturation constant accelerates the convergence while a value close to 0 makes the estimation more precise.
[0022] According to one aspect, the calibration method comprises a step of scaling the elementary complement of the offset parameter as a function of the elementary spatial resolution with which it is associated. This makes it possible to rescale the offset defect after oversampling via the pyramidal decomposition. The offset defect is thus adapted to the dimensions of the photodetector array. Preferably, the scaling is a function of the resolution ratio. This makes it possible to obtain elementary complements of the offset parameter having the same dimension
[0023] According to one aspect, said steps of the method are iterated so as to carry out several updates of the offset parameter of the photodetector matrix in the correction base. This makes it possible to converge the correction base to correct the defects optimally.
[0024] The invention also relates to a computer program type product, comprising at least one sequence of instructions stored and readable by a processor and which, once read by this processor, causes the steps of the method as presented previously to be carried out.
[0025] The invention further relates to a computer-readable medium comprising the computer program product as presented above.
[0026] The invention further relates to a device for calibrating an imaging device configured to observe a scene, the imaging device comprising a matrix of photodetectors configured to capture a raw image, the raw image comprising a plurality of response signals, each photodetector being configured to provide a response signal as a function of the optical intensity of the scene, the calibration device having access to a correction base comprising an offset parameter for the matrix of photodetectors, the calibration device comprising a calculator configured to: • Correct the raw image from the correction base in order to obtain a corrected image, • Upon relative movement of the imaging device with respect to the scene, calculating an offset parameter complement for the photodetector array, the calculation comprising steps consisting of: • Decompose the corrected image in a pyramidal manner into a plurality of elementary spatial resolutions, • Determine, for each elementary spatial resolution, a plurality of high frequencies, • Statistically determine, for each elementary spatial resolution, an elementary complement of the shift parameter determined from the plurality of high frequencies. • Determine, for the photodetector matrix, the offset parameter complement from the elementary offset parameter complements obtained for each elementary spatial resolution, • Update the photodetector array offset parameter with the offset parameter complement.
[0027] The invention also relates to an imaging system comprising an imaging device configured to observe a scene, the imaging device comprising an array of photodetectors configured to capture a raw image, the raw image comprising a plurality of response signals, each photodetector being configured to provide a response signal as a function of the optical intensity of the scene, and a calibration device as presented previously connected to the imaging device. PRESENTATION OF THE FIGURES
[0028] The invention will be better understood on reading the description which follows, given by way of example, and referring to the following figures, given by way of example: non-limiting, in which identical references are given to similar objects.
[0029] [Fig.l] is a schematic representation of an imaging system according to one embodiment of the invention.
[0030] [Fig.2] is a schematic representation of an example of implementation of a calibration method according to the invention.
[0031] [Fig.3] is a schematic representation of a pyramidal decomposition step in order to obtain a complement of shift parameter.
[0032] [Fig.4] is a schematic representation of several spatial resolutions resulting from the pyramidal decomposition.
[0033] [Fig.5] is a schematic representation illustrating substeps for determining an elementary offset parameter complement for a spatial resolution.
[0034] [Fig.6] is a schematic representation of the calculation of a coefficient of an elementary complement of a pixel shift parameter for several images obtained over time for a given spatial resolution.
[0035] [Fig.7] is a schematic representation of the evolution of the offset parameter complement of a pixel for several images obtained over time.
[0036] It should be noted that the figures set out the invention in detail to implement the invention, said figures can of course be used to better define the invention where appropriate. DETAILED DESCRIPTION OF THE INVENTION
[0037] With reference to [Fig. 1], there is shown schematically an imaging device 1 comprising a matrix of photodetectors 10, for example, a camera, a telescope, binoculars, in particular, of the infrared type. Each photodetector 10 is configured to provide a response signal as a function of the optical intensity of a scene SC. In practice, the values of the response signals are not uniform for the same optical intensity and it is necessary to correct the value of the response signal of each photodetector 10 by applying a correction, for example, a gain parameter and / or an offset parameter designated “offset”. The imaging device 1 comprises an output from which a video stream is provided, that is to say, a series of raw images IMb. In known manner, each raw image IMb comprises a plurality of pixels, each pixel having a value which corresponds to the value of the response signal corresponding to a photodetector 10.In a known manner, the imaging device 1 is associated with a display screen 4 in order to allow an operator to view the images of the video stream.
[0038] With reference to [Fig. 1], a calibration device 2 is connected to the output of the imaging device 1 to receive the video stream and provide a corrected video stream to the display screen 4. In this example, the calibration device 2 and the imaging device 1 are integrated into the same imaging system 3. Thus, as will be presented later, the imaging device 1 can be calibrated directly during its use. Dynamic calibration is advantageous since it allows an operator to use the imaging device 1 while performing a calibration in real time.
[0039] The calibration device 2 is configured to determine a correction for the photodetector matrix 10. The correction is in the form of a matrix comprising a plurality of coefficients, each being associated with a photodetector 10. As indicated previously, each correction comprises a gain parameter G and an offset parameter OF. Preferably, as presented previously, the gain parameter G is determined at the factory outlet for the photodetector matrix 10.
[0040] As illustrated in [Fig.l], the calibration device 2 comprises a correction base DB in which the corrections of the photodetectors 10 are recorded and at least one calculation unit 20, for example a processor or an FPGA, configured to apply the corrections and provide a video stream comprising corrected images IMc. This advantageously allows real-time processing.
[0041] As will be presented later, the calculation unit 20 is also configured to implement a method for calibrating the photodetector matrix 10 in order to update the correction base DB in order to dynamically correct any defect which could appear on the photodetector matrix 10 during its life.
[0042] An example of implementation of such a calibration method will now be presented.
[0043] An example of implementation of a calibration method will be presented with reference to figures 2 to 5. In this example, the method will be presented for a raw image IMb from the imaging device 1 but it goes without saying that it can be applied to a video stream comprising a plurality of raw images IMb.
[0044] With reference to [Fig.2], the method comprises a step E0 consisting of correcting the raw image IMb from the photodetector matrix 10 from the correction base DB so as to obtain a corrected image IMc.
[0045] At the start of the implementation, the correction base DB may be empty or comprise only the gain parameters. In this example, the correction base DB comprises a correction for the photodetector array 10, in particular, a correction coefficient for each photodetector 10. In practice, each correction coefficient comprises a parameter coefficient of gain determined at the factory and an offset parameter coefficient which is determined over time. Subsequently, only the OF offset parameter will be presented, that is, a matrix grouping all the offset parameter coefficients.
[0046] With reference to [Fig. 2], the method comprises a step E1 consisting of calculating a complement of the offset parameter dOF for each photodetector of the photodetector matrix 10 from one or more corrected images IMc, hereinafter “calculation step E1”. The complement of the offset parameter dOF is then supplied to the correction base DB in order to update the offset parameter OF of each correction of the correction base DB. In other words, the offset parameter OF is updated dynamically during use of the imaging device 1.
[0047] Preferably, the calculation step E1 is implemented only in the presence of a movement, that is to say, when there is a variation in the response signals of the photodetectors 10. This movement may result from a movement of the imaging device 1 (modification of the direction of view) or from a movement of elements in the scene SC.
[0048] For this purpose, the method comprises a step E2 consisting of measuring a quantity of movement in the corrected image IMc by comparison with a previous corrected image (in particular the last image) and stopping the calibration method if the measured quantity of movement is less than a predetermined movement threshold. Preferably, the movement can be measured as a variation in the value of the pixels of the corrected image IMc. It goes without saying that the quantity of movement could be measured in a different way, for example, by inertial measurements, line-of-sight control or the like. The movement can be estimated globally from an external estimator (setpoints if the line of sight of the imaging device 1 is controlled by a mechanical system or estimation by a gyroscope), globally or locally from image processing functions.The movement is assumed to be sufficient if the translation of the pixels in space, the rotation of the line of sight or the variation of the intensities of the pixels is sufficient. The acceptance thresholds are to be defined according to the compromise between the robustness of the desired calibration and the desired convergence speed.
[0049] As will be presented later, the temporal processing of high frequencies is based on the independence of high spatial frequencies as a function of time. When the scene SC is fixed, the high frequency is not independent as a function of time. The calculation step El is therefore implemented only on the corrected images IMc with movement.
[0050] [Fig.3] is a schematic representation of the sub-steps of the EL calculation step
[0051] The calculation step E1 comprises a step E3 consisting of decomposing the corrected image IMc in a pyramidal manner into a plurality of elementary spatial resolutions Ri. In this example, the corrected image IMc is decomposed into n elementary spatial resolutions Ri with i being an integer between 0 and n. In this example, each resolution Ri is twice as low as the previous resolution Ri-1. It goes without saying that the resolution ratio could be different. The transition from one spatial resolution to another corresponds to low-frequency filtering.
[0052] [Fig.3] represents several elementary spatial resolutions R0-R10 in which the elementary spatial resolution RO corresponds to the corrected IMc image of the VGA type (640x480 pixels). All successive resolutions correspond to the corrected IMc image successively reduced by 2 (according to each dimension) until an image of size 1 pixel is obtained (Elementary spatial resolution RIO). The passage from one elementary spatial resolution to the next has a low-frequency filtering effect. The set of elementary spatial resolutions R0-R10 of the pyramid makes it possible to represent all the frequencies of the corrected IMc image.
[0053] The method comprises a step E4 consisting of determining, for each elementary spatial resolution Ri, a plurality of high frequencies HFi, that is to say, the relevant frequencies in each frequency band.
[0054] According to one aspect, the plurality of high frequencies is obtained by extracting the static part of each elementary spatial resolution Ri from the corrected image IMc. The static part can be easily removed during a movement of the imaging device 10 when it scans the scene to be observed SC. For example, a high frequency is obtained for each pixel by subtracting from this pixel the average of its spatial neighborhood. The neighborhood can be freely defined, for example the 8 neighboring pixels (3x3 square). High frequencies HFi are thus obtained for each elementary spatial resolution Ri.
[0055] The set of high frequencies HFi for the set of elementary spatial resolutions Ri makes it possible to represent the complement of the shift parameter dOF. With reference to [Fig.3], the method comprises a step E5 consisting of statistically determining, for each elementary spatial resolution Ri, an elementary complement of the shift parameter dOFi (in particular a matrix) determined from the corrected images IMc obtained during a movement.
[0056] The elementary complement of the offset parameter dOFi comprises a plurality of coefficients. Each coefficient is associated with a photodetector 10.
[0057] The statistical determination step E5 is presented in [Fig.5] and comprises steps consisting, for each spatial resolution Ri, for each response signal associated with a photodetector 10, of: • E51: Compare the high frequencies HFi for a response signal for a spatial resolution Ri, in absolute value, to a predetermined high threshold SH and determine an elementary complement coefficient of the shift parameter dOFi of zero value in case of overshoot. This makes it possible to remove the edges of the objects from the scene SC. Preferably, the high threshold SH is defined as a function of the expected amplitude of the defect to be corrected and the scene SC. This high threshold SH can be automatically determined statistically on the corrected images IMc. • E52: Multiply by a convergence rate constant Kv, included between 0 and 1, the elementary complement coefficient of the shift parameter dOFi so as to average the values of the elementary complement coefficient of the shift parameter dOFi over time. This convergence rate constant Kv makes it possible to define a compromise between the precision and the desired convergence speed. A value close to 1 accelerates the convergence while a value close to 0 makes the estimation more precise. The convergence rate constant Kv is adjustable by the operator. • E53: Saturate the values of the elementary complement coefficient of dOFi offset parameter to limit variations in the dOF offset parameter. In this example, the values are limited to the interval [-SAT, +SAT] where the SAT saturation constant is used to define a compromise between accuracy and the desired convergence speed. A high saturation constant speeds up convergence while a value close to 0 makes the estimation more accurate. The SAT saturation constant is adjustable by the operator.
[0058] Steps E51, E52, E53 are preferably all implemented, in particular successively, to determine the elementary complement coefficients of the shift parameter dOFi. It nevertheless goes without saying that only certain steps E51, E52, E53 could be implemented.
[0059] As illustrated in [Fig.3], step E5 makes it possible to determine an elementary complement of shift parameter dOFi which is raw for each elementary spatial resolution Ri. The calibration method comprises a step E6 consisting of rescaling the elementary complement of shift parameter dOFi as a function of the elementary spatial resolution Ri with which it is associated, in particular, as a function of the resolution ratio. Each elementary complement of shift parameter dOFi is rescaled. Each elementary complement of shift parameter dOFi is thus in the form of a matrix of the same dimension.
[0060] Thus, for the elementary spatial resolution RI in which each dimension of the image is reduced by 2 (resolution ratio), the elementary complement of the shift parameter dOFl determined for the elementary spatial resolution RI is multiplied by 2 when it is rescaled. In other words, there is a reinterpolation step to the original size of the corrected image IMc.
[0061] The method comprises a step E7 consisting of determining, for the photodetector matrix 10, the sum of the elementary complement of the offset parameter dOFi in order to determine the complement of the offset parameter dOF.
[0062] Thanks to the invention, it is possible to fully reconstruct the offset parameter complement dOF of the photodetector matrix 10 of an imaging device 1 from a pyramid of high frequencies of the corrected image IMc. This makes it possible to update the correction base DB as illustrated in [Fig.2].
[0063] [Fig. 6] is a representation over time t of the high frequency measurements HFi for a plurality of images obtained consecutively by a photodetector of the imaging device 1 (a single pixel) for a resolution Ri. In this [Fig. 6], the different results of the statistical determination step E5 are represented. The black stars correspond to the high frequencies which have been removed by the thresholding step E51 since they do not belong to the range [-SH; SH]. The white stars correspond to the high frequencies which have been saturated at the values -SAT or +SAT since they do not belong to the range [-SAT; +SAT]. The black circles correspond to the high frequencies which belong to the range [-SAT; +SAT] and which have been multiplied by the convergence rate constant Kv.For information, the moments which are not represented by one of the markers correspond to images for which the quantity of movement is insufficient.
[0064] [Fig.7] schematically represents the evolution of the offset parameter OF for a photodetector over time t after processing several images. It is observed that the addition of offset parameter complements dOF over time t makes it possible to converge, in a reduced number of iterations, the offset parameter OF towards the real offset parameter OFr which corresponds to the optimal correction of the photodetector matrix 10. The calibration method allows an incremental improvement of the image quality in the video stream.
[0065] In practice, when a defect appears on a matrix of photodetectors 10, in particular a defect of the “sunburn” type, an OF offset parameter is obtained quickly and reliably due to the pyramidal decomposition which makes it possible to obtain a large variety of high spatial frequencies HFi characteristic of the defect, which makes it possible to converge quickly and precisely towards the OF offset parameter. real offset OFr. This gives a correction base DB which optimally corrects the photodetector matrix 10.
[0066] Advantageously, thanks to the present invention, a spatial defect can be corrected in a practical manner in a short time, of the order of a few tens of seconds, by simple movement of the line of sight of the imaging device. This is particularly the case when the photodetector matrix 10 belongs to an uncooled infrared camera, also called LWIR for “LongWave InfraRed” meaning long wavelength infrared.
Claims
1. Claims Method for calibrating an imaging device (1) configured to observe a scene (SC), the imaging device (1) comprising a matrix of photodetectors (10) configured to capture a raw image (IMb), the raw image (IMb) comprising a plurality of response signals, each photodetector (10) being configured to provide a response signal as a function of the optical intensity of the scene (SC), the method having access to a correction base (DB) comprising an offset parameter (OF) for the matrix of photodetectors (10), method characterized in that it comprises steps consisting of: • Correct (E0) the raw image (IMb) from the correction base (DB) in order to obtain a corrected image (IMc), • During a relative movement of the imaging device (1) with respect to the scene (SC), calculating (El) a complement of offset parameter (dOF) for the matrix of photodetectors (10), the calculation comprising steps consisting of: • Decompose (E3) the corrected image (IMc) in a pyramidal manner into a plurality of elementary spatial resolutions (Ri), • Determine (E4), for each elementary spatial resolution (Ri), a plurality of high frequencies (HFi), • Determine (E5) for each elementary spatial resolution (Ri), an elementary complement of offset parameter (dOFi) determined from a plurality of high frequencies (HFi), the elementary complement of offset parameter (dOFi) comprising a plurality of coefficients associated respectively with a response signal, the method comprising steps consisting of: • compare (E51) the high frequencies (HF i) of a response signal, in absolute value, to a predetermined high threshold (SH) and determine an elementary complement coefficient of the shift parameter (dO Fi) of zero value in the event of overshoot for said response signal, the high threshold (SH) being automatically determined statistically on the corrected images (IMc) Saturate (E53) the elementary complement coefficient of the offset parameter (dOFi) for said response signal so as to limit the variations of the offset parameter (OF) for said response signal over time, • Determine (E7), for the photodetector matrix (10), 1st complement of shift parameter (dOF) from the elementary complements of shift parameter (dOFi) obtained for each elementary spatial resolution (Ri), • Update the shift parameter (OF) of the photodetector matrix (10) with the complement of shift parameter (dOF).
2. Calibration method according to claim 1, comprising a step consisting of • Measuring (E2) a quantity of movement in the corrected image (IMc) and stopping the calibration method if the measured quantity of movement is less than a predetermined movement threshold.
3. Calibration method according to one of claims 1 to 3, wherein, the elementary complement of offset parameter (dOFi) comprising a plurality of coefficients respectively associated with a response signal, the method comprising a step consisting of: • Multiplying (E52) by a convergence rate constant (Kv), between 0 and 1, the coefficient of elementary complement of offset parameter (dOFi) for said response signal so as to average the coefficient of elementary complement of offset parameter (dOFi) for said response signal over time.
4. Calibration method according to one of claims 1 to 3, comprising a step consisting of: • Scale (E6) the elementary offset parameter complement (dOFi) according to the elementary spatial resolution (Ri) to which it is associated.
5. Calibration method according to one of claims 1 to 4, wherein said steps are iterated so as to carry out several updates of the offset parameter (OF) of each photodetector (10) in the correction base (DB).
6. Computer program type product, comprising at least one sequence of instructions stored and readable by a processor and which, once read by this processor, causes the steps of the method according to one of claims 1 to 5 to be carried out.
7. Calibration device (2) of an imaging device (1) configured to observe a scene (SC), the imaging device (1) comprising a matrix of photodetectors (10) configured to capture a raw image (IMb), the raw image (IMb) comprising a plurality of response signals, each photodetector (10) being configured to provide a response signal as a function of the optical intensity of the scene (SC), the calibration device (2) having access to a correction base (DB) comprising an offset parameter (OF) for the matrix of photodetectors (10), the calibration device (2) comprising a calculator (20) configured to: • Correct (E0) the raw image (IMb) from the correction base (DB) so as to obtain a corrected image (IMc), • During a relative movement of the imaging device (1) with respect to the scene (SC), calculate (El) a complement of parameters offset (dOF) for the photodetector array (10),the calculation comprising steps consisting of: • Decomposing (E3) the corrected image (IMc) in a pyramidal manner into a plurality of elementary spatial resolutions (Ri), • Determining (E4), for each elementary spatial resolution (Ri), a plurality of high frequencies (HFi), • Determining (E5) for each elementary spatial resolution (Ri), an elementary complement of offset parameter (dOFi) determined from 1,
8. a plurality of high frequencies (HFi), the elementary complement of offset parameter (dOFi) comprising a plurality of coefficients associated respectively with a response signal, the method comprising steps consisting of: • compare (E51) the high frequencies (HF i) of a response signal, in absolute value, to a predetermined high threshold (SH) and determine an elementary complement coefficient of offset parameter (dO Fi) of zero value in the event of overshoot for said response signal, the high threshold (SH) being automatically determined statistically on the corrected images (IMc) • Saturate (E53) the elementary complement coefficient of the offset parameter (dOFi) for said response signal so as to limit the variations of the offset parameter (OF) for said response signal over time,Determine, for the photodetector matrix (10), the offset parameter complement (dOF) from the elementary complements of the offset parameter (dOFi) obtained for each elementary spatial resolution (Ri), • Update the offset parameter (OF) of the photodetector matrix (10) with the offset parameter complement (dOF). Imaging system (3) comprising an imaging device (1) configured to observe a scene (SC), the imaging device (1) comprising an array of photodetectors (10) configured to capture a raw image (IMb), the raw image (IMb) comprising a plurality of response signals, each photodetector (10) being configured to provide a response signal as a function of the optical intensity of the scene (SC), and a calibration device (2) according to claim 9 connected to the imaging device (1).