WO2023110922A1

WO2023110922A1 - Multispectral "snapshot" (instantaneous) acquisition camera

Info

Publication number: WO2023110922A1
Application number: PCT/EP2022/085693
Authority: WO
Inventors: Philippe Foubert; Thierry TOUATI; Clotilde PEYROT; Nicolas Roux
Original assignee: Safran Electronics & Defense
Priority date: 2021-12-14
Filing date: 2022-12-13
Publication date: 2023-06-22
Also published as: FR3130369A1

Abstract

The invention relates to an acquisition device (1) including a mosaic of filters (2) comprising super-macropixels (SM), each including a plurality of macropixels (M) which each comprise a plurality of elementary pixels (Pe), each super-macropixel being such that: o each macropixel of said super-macropixel forms a general band-pass filter allowing a general spectral band to pass therethrough, the general spectral bands being separate from and consecutive to one another; o each elementary pixel of each macropixel forms an elementary band-pass filter allowing an elementary spectral band (Bel, Be2, Be3, Be4) to pass therethrough, the elementary spectral bands being separate from and consecutive to one another; o for any pair of two macropixels having a side that is shared or a side portion that is shared, the general spectral bands associated with the two macropixels are not adjacent to one another.

Description

“SNAPSHOT” MULTISPECTRAL ACQUISITION CAMERA

(INSTANT)

The invention relates to the field of multispectral cameras, and, in particular, multispectral cameras used to carry out decamouflage operations. BACKGROUND OF THE INVENTION

It is known, to carry out an operation of decamouflage of a target (an armored vehicle for example), to use the technique of multispectral imaging.

Multispectral imaging consists of discretely acquiring the energy reflected or emitted by a surface in a plurality of spectral bands, whether contiguous or not (classically between 3 and 20 spectral bands).

The acquisition is carried out by a multispectral acquisition device comprising a multispectral sensor capable of measuring spectra (of reflectance or luminance in particular) in ranges of wavelengths corresponding to spectral bands located for example in the visible range. and/or in the infrared range.

To improve the performances of the decamouflage, it is advantageous to also use images from an image acquisition device of the VJC type (for Voie Jour Couleurs). A VJC acquisition device provides images very close to what a human being sees. These images are particularly relevant for knowing the “context” of the decamouflage, that is to say for restoring the characteristics of the scene (of the vegetation in particular) in which the target is located. It has therefore been envisaged, in order to achieve the decamouflage, to use an acquisition system comprising a multispectral acquisition device, a VJC acquisition device, and a processing unit. The multispectral acquisition device produces multispectral images, the VJC acquisition device produces VJC images, and the processing unit combines the multispectral images and the VJC images to achieve decamouflage.

However, the operational use of a system comprising two separate cameras is not easy. The system is both bulky and has high power consumption.

Moreover, the combination of these images is complex to achieve.

Multispectral acquisition and VJC acquisition indeed present very different radiometric behaviors.

As the two acquisition devices are not integrated in the same camera, there is no support channel for decamouflage, which leads to difficulties in harmonizing and correcting the VJC and multispectral channels, as well as timing difficulties spatio / temporal.

Moreover, the VJC acquisition technique on small pixels generates calibration difficulties. VJC images are very sensitive to noise due to the size of the pixels.

OBJECT OF THE INVENTION

The subject of the invention is an acquisition device making it possible to acquire multispectral images and VJC type images in a combined manner, and which does not have the drawbacks which have just been mentioned.

SUMMARY OF THE INVENTION With a view to achieving this aim, an acquisition device is proposed comprising:

- a mosaic of filters comprising identical super-macropixels, each super-macropixel comprising a plurality of macropixels each comprising a plurality of elementary pixels, each super-macropixel being such that: o each macropixel of said super-macropixel forms a band-pass filter global allowing a global spectral band to pass, the global spectral bands being distinct and successive; o each elementary pixel of each macropixel of said super-macropixel forms an elementary band-pass filter allowing an elementary spectral band to pass, the elementary spectral bands being distinct and successive; o for any pair of two macropixels belonging to said super-macropixel and having a common side or a common side portion, the global spectral bands associated with the two macropixels are not adjacent;

- a matrix of sensors each associated with an elementary pixel; a processing unit arranged to produce a multispectral image from output signals from the sensors, the multispectral image comprising hyperpixels each associated with an elementary pixel, each hyperpixel comprising spectral components each corresponding to a distinct elementary spectral band.

The acquisition device according to the invention is therefore capable, by using relevant global spectral bands, of producing on a single channel both multispectral images and color and infrared images. We can therefore implement an operation of effective decamouflage by using a single camera in which the acquisition device is integrated; a reduction in cost, volume and consumption is therefore obtained.

The particular configuration of each super-macropixel, and the relative positions of the macropixels, make it possible to significantly reduce crosstalk and noise problems, which improves the accuracy and reliability of decamouflage.

There is also proposed an acquisition device as previously described, in which each super-macropixel has the shape of a square and comprises four macropixels each also having the shape of a square, the macropixels comprising a first macropixel allowing a first global spectral band Bg1 , a second macropixel letting through a second global spectral band Bg2 , a third macropixel letting through a third global spectral band Bg3 and a fourth macropixel letting through a fourth global spectral band Bg4 , the macropixels being such that: Bg1 < Bg2 < Bg3 < Bg4 .

We also propose an acquisition device as previously described, in which, when the mosaic of filters is seen from the front and oriented according to a predefined orientation, each super-macropixel is arranged so that the first macropixel is located at the top at left, the second macropixel is located at the bottom right, the third macropixel is located at the bottom left, and the fourth macropixel at the top right of said super-macropixel.

We also propose an acquisition device as previously described, in which the first global spectral band is included in a blue spectral band, the second global spectral band is included in a green spectral band, the third global spectral band is included in a red spectral band, and the fourth global spectral band is included in a near infrared spectral band.

An acquisition device as previously described is also proposed, in which each macropixel comprises a first elementary pixel allowing a first elementary spectral band Bel to pass, a second elementary pixel allowing a second elementary spectral band Be2 to pass, a third elementary pixel letting pass a third elementary spectral band Be3 , and a fourth elementary pixel allowing a fourth elementary spectral band Be4 to pass, the elementary pixels being such that:

Be1 < Be2 < Be3 < Be4 .

We also propose an acquisition device as previously described, in which, when the mosaic of filters is seen from the front and oriented according to the predefined orientation, each macropixel is arranged so that the first elementary pixel is located at the top left, the second elementary pixel is located at the top right, the third elementary pixel is located at the bottom left, and the fourth elementary pixel is located at the bottom right of said macropixel.

There is also proposed an acquisition device as previously described, in which the processing unit is arranged to confer, on each spectral component of the hyperpixel associated with a particular elementary pixel of a particular super-macropixel, a value equal to the output signal of the sensor associated with the elementary pixel belonging to said particular super-macropixel and passing the elementary spectral band corresponding to said spectral component.

There is also proposed an acquisition device as previously described, in which the processing unit is arranged to confer, on each spectral component of the hyperpixel associated with a particular elementary pixel of a particular super-macropixel, a value obtained by interpolation of the output signals from the sensors associated with elementary pixels neighboring the particular elementary pixel and allowing the elementary spectral band corresponding to said spectral component to pass.

There is also proposed an acquisition device as previously described, in which the processing unit is arranged to confer, on each spectral component of the hyperpixel associated with a particular elementary pixel of a particular super-macropixel, a value equal to the output signal of the sensor associated with an elementary pixel, allowing the elementary spectral band corresponding to said spectral component to pass, said elementary pixel belonging to a sliding window having the size of a super-macropixel and to which the elementary pixel belongs particular .

An acquisition device as described above is also proposed, the processing unit being further arranged to produce a color and infrared image from the output signals of the sensors.

We also propose an acquisition device as previously described, in which the processing unit is arranged, to produce the image in color and infrared, to implement an operation of spatial grouping of elementary pixels, then an algorithm color reproduction, the pixel grouping operation elements consisting in associating a single intermediate value with each macropixel of each super-macropixel, said intermediate value being obtained by weighting the output signals of the sensors associated with the elementary pixels of said macropixel.

There is also proposed an acquisition device as described above, further comprising a notch filter, positioned at the level of a pupil of a camera in which the acquisition device is integrated, the notch filter being designed to cut a boundary spectral band located between the third global spectral band and the fourth global spectral band.

We also propose an acquisition device as previously described, in which the frontier spectral band belongs at least partially to the red spectral band, and in which the third global spectral band is shifted so as not to present any overlap with the red spectral band. frontier spectral band.

A camera is also proposed comprising an acquisition device as previously described.

The invention will be better understood in the light of the following description of a particular non-limiting mode of implementation of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

Reference will be made to the appended drawings, including:

FIG. 1 represents an acquisition device according to the invention;

FIG. 2 represents a mosaic of filters and a super-macropixel;

FIG. 3 represents a super-macropixel; FIG. 4 illustrates a sub-sampling method that can be implemented to obtain hyperpixels associated with elementary pixels;

FIG. 5 illustrates an interpolation method that can be implemented to obtain hyperpixels associated with elementary pixels;

FIG. 6 illustrates a sliding window method that can be implemented to obtain hyperpixels associated with elementary pixels;

FIG. 7 represents spectral bands of the resins and of the elementary pixels;

FIG. 8 represents the frontier spectral band cut by a Notch filter.

DETAILED DESCRIPTION OF THE INVENTION

Referring to FIG. 1, the acquisition device according to the invention 1 comprises a mosaic of filters 2 , a photodetector 3 and an internal processing unit 4 .

The acquisition device 1 is integrated in a camera 6 .

An external processing unit 5 is connected to the internal processing unit 4 . Here, the external processing unit 5 is located at a distance from the camera 6, but it could perfectly be integrated into the camera 6 (and for example share electronic or software resources with the internal processing unit 4).

The mosaic of filters 2 contains n*m elementary pixels.

The photodetector 3 integrates a matrix of sensors also comprising n*m sensors, each sensor forming a physical pixel and being associated with a distinct elementary pixel.

The sensors are for example CCD sensors (for Charged Coupled Device) or CMOS (for Complementary Metal Oxide Semiconductor).

The internal processing unit 4 is an electronic and software unit. The internal processing unit 4 acquires the output signals produced by the sensors and analyzes them to produce multispectral images and color and infrared images.

The external processing unit 5 is also an electronic and software unit.

The external processing unit 5 acquires the multispectral images and the color and infrared images and performs the decamouflage operation.

The internal processing unit 4 comprises at least one processing component 8 suitable for executing program instructions.

The processing component 8 is for example a microcontroller, a conventional processor, a GPU (for Graphics Processing Unit, which can be translated as "graphics processor"), a DSP (for Digi tal Signal Processor, which can can be translated as "digital signal processor"), or else a programmable logic circuit such as an FPGA (for Field Programmable Gate Arrays) or an ASIC (for Application Specific Integrated Circuit).

The internal processing unit 4 also comprises at least one memory 9 making it possible in particular to store the instructions of the programs which have just been mentioned.

Likewise, the external processing unit 5 comprises at least one processing component 10 and at least one memory 11 .

We are now more particularly interested in the mosaic of filters 2 .

With reference to Figures 2 and 3, the mosaic of filters 2 includes SM super-macropixels which are all identical.

Each super-macropixel SM comprises a plurality of macropixels M each comprising a plurality of elementary pixels Pe.

Here, each super-macropixel SM comprises four macropixels M each comprising four elementary pixels Pe.

Each super-macropixel SM and each macropixel M each have the shape of a square.

In each super-macropixel SM, each macropixel M of said super-macropixel SM forms a global band-pass filter allowing a global spectral band to pass, the global spectral bands being distinct and successive.

By "distinct and successive", we mean that it is possible to order the N global spectral bands of a super-macropixel Bg1 , Bg2 , ...Bgi , ..., BgN according to a list such as: Bg1 < Bg2 < ... < Bgi < ... < BgN .

This means that the maximum wavelength of Bg1 is lower than the minimum wavelength of Bg2, that the maximum wavelength of Bg2 is lower than the minimum wavelength of Bg3, and so on.

The macropixels M therefore here comprise a first macropixel M1 letting through a first global spectral band Bg1, a second macropixel M2 letting through a second global spectral band Bg2, a third macropixel M3 letting through a third global spectral band Bg3 and a fourth macropixel M4 letting through pass a fourth global spectral band Bg4, the macropixels M being such that: Bg1<Bg2<Bg3<Bg4.

The global spectral bands are equidistributed by subdomains Blue, Green, Red, Near Infrared.

The first global spectral band Bg1 is included in a blue spectral band, the second global spectral band Bg2 is included in a green spectral band, the third global spectral band Bg3 is included in a red spectral band, and the fourth global spectral band Bg4 is included in a near infrared (NIR) spectral band.

The first macropixel M1 is made with a blue resin (ReB). The second macropixel M2 is made with a green resin (ReV). The third M3 macropixel is made with a red resin (ReR). The fourth macropixel M4 is made with a PIR (ReP) resin.

When the mosaic of filters 2 is seen from the front and oriented according to a predefined orientation, which corresponds in this case to the orientation visible in FIG. 2, each super-macropixel SM is arranged so that the first macropixel M1 is located in top left, the second macropixel M2 is located bottom right, the third macropixel M3 is located bottom left, and the fourth macropixel M4 is located top right of said super-macropixel SM.

It is noted that, in each super-macropixel SM, for any pair of two macropixels M belonging to said super-macropixel and having a common side or a common side portion, the global spectral bands associated with said two macropixels M are not adjacent.

By “adjacent global spectral bands”, we mean two bands which follow one another in the list mentioned earlier.

Similarly, for each macropixel M of a super-macropixel SM, each elementary pixel Pe of said macropixel M forms a elementary band-pass filter allowing an elementary spectral band to pass, the elementary spectral bands being distinct and successive.

Thus, for the first macropixel M1 (blue), comprising the group of elementary pixels {B1, B2, B3, B4}, we have:

Be_B1 < Be_B2 < Be_B3 < Be_B4, where Be_B1 is the elementary spectral band of elementary pixel B1, Be_B2 is the elementary spectral band of elementary pixel B2, Be_B3 is the elementary spectral band of elementary pixel B3, and Be_B4 is the elementary spectral band of elementary pixel B4.

For the second macropixel M2 (green), comprising the group of elementary pixels {V1, V2, V3, V4}, we have:

Be_V1 < Be_V2 < Be_V3 < Be_V4, where Be_V1 is the elementary spectral band of the elementary pixel V1, Be_V2 is the elementary spectral band of the elementary pixel V2, Be_V3 is the elementary spectral band of the elementary pixel V3, and Be_V4 is the elementary spectral band of the elementary pixel V4.

For the third macropixel M3 (red), comprising the group of elementary pixels {R1, R2, R3, R4}, we have:

Be_R1 < Be_R2 < Be_R3 < Be_R4, where Be_R1 is the elementary spectral band of the elementary pixel R1, Be_R2 is the elementary spectral band of the elementary pixel R2, Be_R3 is the elementary spectral band of the elementary pixel R3, and Be_R4 is the elementary spectral band of the elementary pixel R4.

For the fourth macropixel M4 (near infrared), comprising the group of elementary pixels {PIR1, PIR2, PIR3, PIR4}, we have:

Be_PIR1 < Be_PIR2 < Be_PIR3 < Be_PIR4, where Be_PIR1 is the elementary spectral band of the pixel elementary PIR1 , Be_PIR2 is the elementary spectral band of elementary pixel PIR2 , Be_PIR3 is the elementary spectral band of elementary pixel PIR3 , and Be_PIR4 is the elementary spectral band of elementary pixel PIR4 .

Thus, each macropixel M comprises a first elementary pixel allowing a first elementary spectral band Bel to pass, a second elementary pixel allowing a second elementary spectral band Be2 to pass, a third elementary pixel allowing a third elementary spectral band Be3 to pass, and a fourth elementary pixel allowing a fourth elementary spectral band Be4 to pass, the elementary pixels being such that: Be1<Be2<Be3<Be4.

The Bel, Be2, Be3, Be4 depend on the micropixel concerned.

When the mosaic of filters 2 is seen from the front and oriented according to the predefined orientation, each macropixel M is arranged so that the first elementary pixel of said macropixel M is located at the top left, the second elementary pixel is located at the top right , the third elementary pixel is located at the bottom left, and the fourth elementary pixel is located at the bottom right of said macropixel M .

The elementary band-pass filter of each elementary pixel Pe typically has an FWHM (Full Width at Half Maximum, or “width at half-height”) equal to about 20 nm.

The global band-pass filter of each macropixel M typically has an FWHM equal to about 80 nm.

The relative arrangement of the macropixels M is important for limiting the geometric contiguity of adjacent spectral bands, and therefore for limiting crosstalk.

We now describe the processing carried out by the internal processing unit 4 to produce a multispectral image (also called a hypercube) from the sensor output signals.

The multispectral image comprises, for each elementary pixel Pe, a hyperpixel comprising a plurality of spectral components, ie one spectral component per elementary spectral band. Each hyperpixel therefore comprises here 16 spectral components each associated with a distinct elementary spectral band.

Contrary to other optical architectures, for which the formation of the multispectral image consists for example of “stacking” the imagettes, the use of the mosaic of filters has the consequence that the filtered pixels are not spatially coherent.

Consequently, as for a so-called “Bayer” matrix color sensor, the internal processing unit 4 uses an algorithm to calculate the spectral values corresponding to each hyperpixel.

The optics and the multispectral image formation algorithm are dimensioned to maximize the performance of the acquisition device 1, and in particular, to optimize the detection range (or the apparent size of the smallest detectable target). This dimensioning must be based on a prediction model of this detection range, or, at the very least, on an analytical model making it possible to position in relative terms the ranges given by different architectures of sensor, optics and shaping. spectral data.

Indeed, the range is intimately linked to the angular extent of the zone which contributes to the formation of a hyperpixel, to the spatial sampling step of the hyperpixels, but also to the angular difference separating two points to avoid a "mixing spectral” in the evaluation of their respective spectrum. Contributors are both the optics (resolving power) and the hypercube formation algorithm.

Three modes of reconstruction of the multispectral image are presented here.

FIG. 4 illustrates a first reconstitution mode. We see four SM super-macropixels in Figure 4.

The first reconstitution mode consists, for the internal processing unit 4, in conferring, on each spectral component 14 of the hyperpixel HP associated with a particular elementary pixel Pe of a particular super-macropixel SM, a value equal to the signal sensor output associated with the elementary pixel Pe belonging to said particular super-macropixel SM and allowing the elementary spectral band corresponding to said spectral component 14 to pass.

The four hyperpixels HP of FIG. 4 are therefore distinct, each hyperpixel HP being associated with all the elementary pixels Pe of the same super-macropixel SM.

In order not to mix the spectral bands (condition of non-mixing), it is however necessary that the apparent size of the target that one seeks to disambiguate is greater than or equal to 2×2 macropixels M to always encompass at least one complete super-macropixel SM .

The first reconstruction mode introduces a relatively strong under-sampling. However, this reconstitution mode has the advantage of not impacting the radiometry, since no interpolation or smoothing type processing is carried out. The volume of data used to produce the multispectral image corresponds to the initial volume.

With reference to FIG. 5, a second reconstitution mode consists, for the internal processing unit 4, in giving, to each spectral component of the hyperpixel HP associated with a particular elementary pixel Pe of a particular super-macropixel, a value obtained by interpolation of the output signals from the sensors associated with elementary pixels neighboring the particular elementary pixel and allowing the elementary spectral band corresponding to said spectral component to pass.

In this second reconstitution mode, the internal processing unit 4 therefore restores, for each elementary pixel, a hyperpixel by calculating the value of the various spectral bands by interpolation from the values measured on the neighboring physical pixels corresponding to this band, as would do, for a color image, a “Bayer” type treatment.

This solution, in principle, therefore makes it possible to calculate as many hyperpixels as elementary pixels, and therefore to restore a “high resolution” hypercube.

In order not to mix the spectral bands (condition of non-mixing), the target should however have a width and a height which cover the elementary pixels entering into the interpolation algorithm.

The drawing on the left of FIG. 5 illustrates a case where the interpolation involves the elementary pixels of the same spectral band surrounding the hyperpixel to be calculated.

The choice of the second reconstitution mode, and therefore of the interpolation algorithm, can decrease the impact of the furthest physical pixel, and the non-mixing zone can tend to approach the size of the super-macropixel.

With reference to FIG. 6, a third reconstitution mode consists, for the internal processing unit 4, in giving, to each spectral component of the hyperpixel associated with a particular elementary pixel of a particular super-macropixel, a value equal to output signal of the sensor associated with an elementary pixel, allowing the elementary spectral band corresponding to said spectral component to pass, said elementary pixel belonging to a sliding window 15 having the size of a super-macropixel and to which the particular elementary pixel belongs.

Thus, in FIG. 6, the hyperpixel HP1 is associated with the elementary pixel Pel while the hyperpixel HP2 is associated with the elementary pixel Pe2.

This mode therefore restores a hyperpixel for each physical pixel by assigning, for each elementary spectral band, the value of the closest physical pixel corresponding to this band. This therefore amounts to moving on the sensor, from pixel to pixel, the window 15 , and assigning to the corresponding hyperpixel the reordered values taken from this window 15 .

This solution makes it possible to calculate as many hyperpixels as physical pixels, and therefore to restore a “high resolution” spectral image. This “nearest neighbour” logic is the one that limits the interpolation medium the most: whatever the position of the window 15, all the wavelengths are represented in a close neighborhood.

In order not to mix the spectral bands (condition of non-mixing), the target should however have a width and a height close to those of a super-macropixel SM.

We are now interested in the synthesis of the VJC and PIRL (near wide band infrared) color pathways.

The internal processing unit 4 produces a color and infrared image from the output signals of the sensors. By "color and infrared image" is meant an image forming a colored composition and produced from blue, red, green, and infrared (here near infrared) spectral bands.

To produce the color and infrared image, the internal processing unit 4 first implements an operation of spatial binning of elementary pixels Pe (that is to say an operation of spatial grouping of elementary pixels Pe) . Spatial binning makes it possible to obtain an intermediate image.

Spatial binning consists of associating a single intermediate value with each macropixel M of each super-macropixel SM, said intermediate value being obtained by weighting the output signals from the sensors associated with the elementary pixels of said macropixel.

So we have :

where B _{Super-macropixel} (i ₀ , j ₀ ) is the intermediate value of an elementary pixel Pe belonging to a first macropixel M1 (blue), θ _B is a first weighting coefficient, and the Pixels(i,j) are the output signals from the sensors associated with the elementary pixels Pe belonging to said first macropixel M1;

where V _{Super-macropixel} (i ₀ , j ₀ ) is the intermediate value of an elementary pixel Pe belonging to a second macropixel M2 (green), θ _V is a second weighting coefficient, and the Pixels(i,j) are the output signals from the sensors associated with the elementary pixels Pe belonging to said second macropixel M2;

where R _{Super-macropixel} (i ₀ , j ₀ ) is the intermediate value of an elementary pixel Pe belonging to a third macropixel M3 (red), θ _R is a third weighting coefficient, and the Pixels(i,j) are the output signals from the sensors associated with the elementary pixels Pe belonging to said third macropixel M3;

where PIR _{Super-macropixel} (i ₀ , j ₀ ) is the intermediate value of an elementary pixel Pe belonging to a fourth macropixel M4 (PIR), θ _PIR is a fourth weighting coefficient, and the Pixels(i,j) are the output signals from the sensors associated with the elementary pixels Pe belonging to said fourth macropixel M4.

A 2x2 spatial binning (by macropixel) is therefore implemented according to the groups.

For elementary pixels of size 4 μm×4 μm to 6 μm×6 μm, equivalent pixels are therefore obtained having a size of 8 μm×8 μm to 12 μm×12 μm.

A gain in sensitivity is thus obtained (because the colorimetric calibration is sensitive to noise in images with small pixels and in images with low flux). The image is obtained by a single reading for 4 elementary pixels.

In addition, a spreading of the PSF (for Point Spread. Function, or point spreading function) is obtained by adjusting the optics along 2×2 elementary pixels (elementary pixel greater than or equal to 4 μm×4 μm), and a stress relief on optics in terms of optical cut-off frequency. The optical resolution is reduced and the optical spot is spread.

Advantageously, the spatial grouping is carried out in hardware, that is to say by electronic components (and not software). This allows an improvement of the sensor signal-to-noise ratio. The accumulated charges of each physical pixel are for example accumulated by using summation gates to obtain the intermediate values of the intermediate image.

The color and infrared image is then reconstituted from the intermediate image, and therefore from the synthesized bands of the macropixels of each super-macropixel.

We see in Figure 7 the first global spectral band Bg1 (optical transmission) of the first resin (blue), the second global spectral band Bg2 of the second resin (green), the third global spectral band Bg3 of the third resin (red ) and the fourth spectral band Bg4 of the fourth resin (PIR).

We also see, for each global spectral band, the associated elementary spectral bands Be1 , Be2 , Be3 , Be4 .

The resins may have imperfections producing rebounds in the near infrared range. It may be relevant to improve the spectral zone of transition between the visible domain (B, G, R) and the near infrared domain, to prevent the transmission profiles of the resins from being disturbed.

A Notch filter (band-stop filter) positioned at the level of the pupil of the camera 6 is therefore used.

With reference to figure 8, the Notch filter makes it possible to cut a border spectral band Bf located between the third global spectral band Bg3 (red) and the fourth global spectral band Bg4 (near infrared) .

The response of the Notch filter has a downward slope from a wavelength greater than 680nm, and an upward slope from a wavelength less than 730nm.

Advantageously, provision is made for the frontier spectral band Bf to be situated in the red spectral band Br, as can be seen in FIG. 8. The third global spectral band Bg3 is then shifted so that the third global spectral band Bg3 and the spectral band Bf border do not show any overlap.

It can be seen in FIG. 8 that the central wavelength of the third global spectral band Bg3 is slightly reduced, which shifts the third global spectral band to the left.

The red components are thus much better separated from the near infrared components.

Advantageously, a quartz blade forming a low-pass filter is also used. The low-pass filter is positioned just before the matrix of sensors and makes it possible to reduce the interference produced by the mosaic of filters.

Following these processing operations, the internal processing unit 4 produces a combined image from the multispectral image and the color and infrared image, and directly transmits the combined image to the external processing unit 5. Alternatively , the external processing unit 5 acquires the multispectral image and the color and infrared image.

The external processing unit 5 then implements the supra decamouflage to attempt to detect the presence of the target. In the case of a normal light environment, the acquisition device 1 is configured to carry out a defocusing and a spreading of the PSF over the 4×4 elementary pixels Pe of each super-macropixel SM.

The hyperspectral image is formatted in absolute luminance with a calibration process.

A contrast enhancement method is then implemented. The contrast enhancement method possibly uses transparency and/or afterglow effects (see for example document FR 3 011 663 B1).

A metric, representative of a contrast in the scene, is then used to perform the decamouflage. The metric is calculated from the 16 elementary spectral bands.

The color and infrared image is reconstructed to constitute the support channel (no binning for the application). Calibration is done with binning. Bands B, V, R and PIRL are digitally synthesized.

We do not use binning for the VJC application, but a "custom" weighting (personalised, adapted to the application) of the bands to constitute the VJC, which is less sensitive to color metamerism. The weighting coefficients are stored in a specific saturated color look-up table (LUT).

In the case of a light environment with low solar illumination, the acquisition device 1 is configured to carry out a defocusing and a spreading of the PSF on the 2×2 macropixels M of the super-macropixel SM. A gain in sensitivity is obtained because the analysis is made on the basis of the macropixel (2×2 elementary pixels).

The hyperspectral image is formatted with a VJC + PIRL calibration process. A contrast enhancement method is then implemented. The contrast enhancement process possibly uses transparency and/or afterglow effects.

A metric, representative of a contrast in the scene, is then used to perform the decamouflage. The metric is calculated on 4 spectral bands only (due to binning).

The color and infrared image is reconstructed to constitute the support channel (binning is used for the application). Calibration is done with binning. Bands B, V, R and PIRL are digitally synthesized.

Of course, the invention is not limited to the embodiment described but encompasses any variant falling within the scope of the invention as defined by the claims.

The overall spectral bands may well be different from those described here. Any band in the visible, near infrared PIR or NIR (Near InfraRed) domains, or short infrared SWIR (Short Wavelength Infrared), could for example be used.

Super-macropixels and macropixels may be different from those presented here: different number of elementary pixels or macropixels, different shape of macropixels or super-macropixel, etc.

Claims

1 . Acquisition device ( 1 ) comprising:

- a mosaic of filters (2) comprising identical super-macropixels (SM), each super-macropixel comprising a plurality of macropixels (M) each comprising a plurality of elementary pixels (Pe), each super-macropixel being such that: o each macropixel of said super-macropixel forms a global band-pass filter allowing a global spectral band (Bg1, Bg2, Bg3, Bg4) to pass, the global spectral bands being distinct and successive; o said super-macropixel (SM) comprises four macropixels, the macropixels comprising a first macropixel (M1) letting through a first global spectral band Bg1, a second macropixel (M2) letting through a second global spectral band Bg2, a third macropixel (M3 ) allowing a third global spectral band Bg3 to pass and a fourth macropixel (M4) allowing a fourth global spectral band Bg4 to pass, the macropixels being such that: Bg1 <Bg2 <Bg3 <Bg4; o when the mosaic of filters (2) is seen from the front and oriented according to a predefined orientation, said super-macropixel is arranged so that the first macropixel is located at the top left, the second macropixel is located at the bottom right, the third macropixel is located at the bottom left, and the fourth macropixel at the top right of said super-macropixel; o each elementary pixel of each macropixel of said super-macropixel forms an elementary band-pass filter allowing an elementary spectral band (Bel, Be2, Be3, Be4) included in the overall spectral band of said macropixel to pass, the elementary spectral bands being distinct and successive;

- a matrix of sensors (3) each associated with an elementary pixel;

- a processing unit (4) arranged to produce a multispectral image from output signals from the sensors, the multispectral image comprising hyperpixels (HP) each associated with an elementary pixel, each hyperpixel comprising spectral components (14) corresponding each has a distinct elementary spectral band.

2. Acquisition device according to claim 1, in which each super-macropixel (SM) and each macropixel have the shape of a square.

3 Acquisition device according to claim 1, in which the first global spectral band is included in a blue spectral band, the second global spectral band is included in a green spectral band, the third global spectral band is included in a red spectral band , and the fourth global spectral band is included in a near infrared spectral band.

4. Acquisition device according to one of the preceding claims, in which each macropixel comprises a first elementary pixel allowing a first elementary spectral band Bel to pass, a second elementary pixel allowing a second elementary spectral band Be2 to pass, a third elementary pixel allowing a second elementary spectral band to pass pass a third elementary spectral band Be3, and a fourth elementary pixel allowing a fourth elementary spectral band Be4 to pass, the elementary pixels being such that: Be1<Be2<Be3<Be4.

5. Acquisition device according to claim 4, in which, when the mosaic of filters (2) is seen from the front and oriented according to the predefined orientation, each macropixel is arranged so that the first elementary pixel is located at the top left, the second elementary pixel is located at the top right, the third elementary pixel is located at the bottom left, and the fourth elementary pixel is located at the bottom right of said macropixel.

6. Acquisition device according to one of the preceding claims, in which the processing unit (4) is arranged to give each spectral component (14) of the hyperpixel (15) associated with a particular elementary pixel d a particular super-macropixel, a value equal to the output signal of the sensor associated with the elementary pixel belonging to said particular super-macropixel and allowing the elementary spectral band corresponding to said spectral component to pass.

7. Acquisition device according to one of claims 1 to 5, in which the processing unit (4) is arranged to confer, on each spectral component of the hyperpixel associated with a particular elementary pixel of a super- particular macropixel, a value obtained by interpolation of the output signals of the sensors associated with elementary pixels neighboring the particular elementary pixel and allowing the elementary spectral band corresponding to said spectral component to pass.

8. Acquisition device according to one of claims 1 to 5, in which the processing unit (4) is arranged to confer, on each spectral component of the hyperpixel associated with a particular elementary pixel of a super- particular macropixel, a value equal to the output signal of the sensor associated with an elementary pixel, allowing the elementary spectral band corresponding to said spectral component to pass, said elementary pixel belonging to a sliding window having the size of a super-macropixel and to which the particular elementary pixel belongs to.

9. Acquisition device according to one of the preceding claims, the processing unit (4) being further arranged to produce a color and infrared image from the output signals of the sensors.

10. Acquisition device according to claim 9, in which the processing unit (4) is arranged, to produce the color and infrared image, to implement an operation of spatial regrouping of elementary pixels, then an algorithm color reproduction, the elementary pixel grouping operation consisting in associating a single intermediate value with each macropixel of each super-macropixel, said intermediate value being obtained by weighting the output signals of the sensors associated with the elementary pixels of said macropixel.

11. Acquisition device according to claim 10 and claim 3, further comprising a notch filter, positioned at the level of a pupil of a camera (6) in which the acquisition device is integrated, the notch filter being designed to cut off a frontier spectral band (Bf) located between the third global spectral band and the fourth global spectral band.

12. Acquisition device according to claim 11, in which the frontier spectral band (Bf) belongs at least partially to the red spectral band, and in which the third overall spectral band is shifted so as not to overlap with the frontier spectral band.

13. Camera (6) comprising an acquisition device (1) according to one of the preceding claims.