EP3555845A1

EP3555845A1 - Means for producing and/or checking a part made of composite materials

Info

Publication number: EP3555845A1
Application number: EP16829103.7A
Authority: EP
Inventors: Serge Luquain
Original assignee: Techni-Modul Engineering
Current assignee: Techni-Modul Engineering
Priority date: 2016-12-14
Filing date: 2016-12-14
Publication date: 2019-10-23
Also published as: WO2018109286A1; CN110088800A; US20200027224A1

Abstract

The invention relates to a method for producing and/or checking a part made of composite materials formed from at least one fabric having a surface whose texture exhibits a main orientation, comprising the following steps: -obtaining a first image representing the texture of the fabric; -determining an estimation relating to the main orientation of the texture, by: determining, for each pixel of the first image, an orientation of gradients relating to the luminance level of said pixel; determining an estimation of a global distribution of the orientations of gradients of the pixels of the first image; determining the main orientation as a function of the estimation of the global distribution of the orientations of gradients of the pixels of the first image; -determining a deviation between the estimation relating to the main orientation and a setpoint value; -producing the part as a function of the deviation and/or emitting a check signal dependent on the deviation.

Description

Means for producing and / or controlling a composite material part

The present invention relates to the field of production of composite material parts, as well as quality control processes of said parts. More particularly, the invention relates to the production and / or controls of preforms, adapted to be used to manufacture composite material parts.

The use of parts made of composite materials continues to increase, particularly in the field of transport, due in particular to the potential gains in terms of weight, strength, stiffness or even life. For example, composite materials are widely used in both the secondary and primary structures of civil and military aircraft, but also for the realization of many elements in motor vehicles or in railway equipment.

Among the known processes for manufacturing parts made of composite materials, the manufacturing method comprising a draping step followed by a cooking step in an autoclave, an oven or a press is widely used. Such a process often requires the use of a preform-set of compacted tissues. To manufacture a preform, it is known to cut, position and form layers of fabric, manually and non-automatically. One operator in particular must ensure the proper main orientation of the various tissue layers, relative to each other.

Also, such an approach has the defect of low repeatability when producing a series of complex preforms, each manual intervention being subject to various errors or defects. In addition, the productivity of such a process is limited by the necessary controls and by the difficulty of implementing said steps. It is therefore generally desirable to automate the step of producing complex preforms. It is understood that the choice of automation is conditioned by other criteria also in particular by the number of parts to be produced (profitability of the process), by the shape of the part, etc.

However, to allow the automation of such a process, for example by deploying robots to manufacture preforms, it is necessary to be able to determine and / or control the main orientation of the tissue layers used, each layer having a directional texture. . A texture is described as directional when elementary structures identifiable in said texture are arranged substantially in a main orientation. Various image analysis methods are known for determining the main orientation of a texture. The document Da-Costa, JP "Statistical analysis of directional textures, application to the characterization of composite materials", Thesis of the University of Bordeaux 1, defended on December 21, 2001, describes such a process. However, in order to determine the main orientation of a texture, the known methods analyze the texture image in a dense manner, that is to say by using in an identical manner any pixel or any point of said image. This results in a significant sensitivity to the noise present in the image, which negatively affects the reliability and accuracy of the determination of the main orientation.

There is therefore still a need for means capable of enabling the automated production of composite material parts, especially preforms, which is economically efficient and which guarantees improved quality, repeatability and accuracy compared to manual processes.

One of the objects of the invention is to provide means capable of enabling the automated production of composite material parts. Another object of the invention is to provide means capable of allowing the determination of the main orientation of a texture, which are not very sensitive to the noise present in the image representing the texture. Another object of the invention is to provide a method for determining an estimate of the main orientation of a particularly robust texture, of very good quality even in the presence of a significant noise in the image representing the texture. Another object of the invention is to propose a method for determining a principal orientation error of a texture with respect to a reference orientation, which is particularly robust, of very good quality even in the presence of a large noise in the field. image representing the texture.

One or more of these objects are filled by the method to determine an estimate for a main orientation of a texture. The dependent claims further provide solutions to these objects and / or other advantages.

More particularly, according to a first aspect, the invention relates to a method for producing and / or controlling a composite material part formed from at least one fabric having a surface whose texture has a main orientation. The method comprises the following steps:

obtaining a first image formed by a plurality of pixels and in which a luminance level can be determined for each pixel, representing the texture of the fabric; - determine an estimate of the main orientation of the texture:

in a first step, determining, for each pixel of the first image, an orientation of gradients relative to the luminance level of said pixel;

in a second step, by determining an estimate of an overall distribution of the gradient orientations of the pixels of the first image;

in a third step, determining the main orientation as a function of the estimate of the overall distribution of the gradient orientations of the pixels of the first image;

- determining a difference between the estimate of the main orientation and a set value;

- produce the part according to the deviation and / or issue a control signal depending on the deviation.

Typically, the piece of composite material is for example composed of a plurality of superimposed fabrics so that the main orientation of each fabric respects a predetermined sequence. The part may be in particular a preform adapted to be used to manufacture composite material parts, in particular by implementing a process comprising a draping step followed by a cooking step in an autoclave, a furnace or a hurry. The method is particularly adapted to be implemented by automated devices, such as robots provided with calculation means and an image pickup device, adapted to arrange the fabrics relative to one another according to one another so that the main orientation of each tissue substantially respects a predetermined sequence.

The set value is for example obtained from a predefined value or calculated as a function, for example, of the orientation of at least one of the other fabrics forming the part. Alternatively, the setpoint may be a range of acceptable values.

During the production step of the workpiece, the position of the fabric can be modified to reduce the deviation to a value below a predefined threshold. For this purpose, the previous steps may be repeated as often as necessary.

During the step of transmitting the control signal as a function of the difference 40b, the control signal can be generated so as to allow the signaling, to a operator-controller or a quality control device, a potential defect of orientation of the fabric, when the difference is greater than a tolerance value greater than a predefined threshold.

During the first step, for each pixel of the first image, the orientation of gradients relative to the luminance level of said pixel can be obtained by:

Determining a first response of a first Sobel-type convolution filter applied at the luminance level in a first direction;

A second response of a second Sobel-type convolution filter applied at the luminance level, in a second direction orthogonal to the first direction;

• calculating the argument of a vector whose horizontal component corresponds to the first response and the vertical component corresponds to the second response.

During the second step, the estimation of the overall distribution of the gradient orientations of the pixels of the first image can be determined by constructing a discrete histogram, comprising a plurality of classes relating to different ranges of possible values for the gradient orientations. pixels of the first image. The histogram is for example constructed by determining:

· For each pixel of the first image, a contribution; and,

• for each class, a height equal to the sum of the contributions of all the pixels of the first image whose orientation is included in said class.

During the third step, the main orientation can be determined by identifying, in the histogram, the class whose height is maximum. Prior to the first step, the method may comprise:

A step, during which, for each pixel of the first image, a score relating to the membership of said pixel in a luminance contour is determined;

A step, during which, for each pixel of the first image, a probability of importance is determined, using the score of said pixel;

and wherein, in the second step, the contribution of each pixel is determined according to the importance probability associated with said pixel. For each pixel of the first image, the score is for example a calculated Harris score, based on the luminance levels of the pixels of the first image. Thus, the presence of noise - for example "pepper and salt" - is filtered by the use of the Harris score and the probability of importance. Advantageously, the corners of the first image used to calculate the score of each pixel of the first image may correspond to a break between luminance levels of the first image in a single direction. For each pixel of the first image, the score is for example an estimate of the luminance gradient amplitude, based on the luminance levels of the pixels of the first image. For each pixel of the first image, the probability of importance can be determined using a sigmoid function and the score of said pixel.

Prior to the third step, the method may include a step in which the luminance level of each pixel of the first image is filtered so as to reduce the noise present in the luminance information of the first image.

In one embodiment,

During the first step, for each pixel of a reference image of a textured surface, said reference image being formed by a plurality of pixels and in which a luminance level can be determined for each pixel, an orientation gradients relating to the luminance level of said pixel is determined;

During the second step, a reference estimate of an overall distribution of the gradient orientations of the pixels of the reference image is determined;

During the third step, an error of the main orientation is determined according to the estimation of the global distribution of the gradient orientations of the pixels of the first image and according to the reference estimate of the global distribution gradient orientations of the pixels of the first image. The difference between the estimate for Primary orientation and the setpoint is determined by the error of the main orientation. In particular, the reference value can be chosen according to the reference estimate of the overall distribution of the gradient orientations of the pixels of the reference image.

During the second step, the reference estimate of the overall distribution of the gradient orientations of the pixels of the reference image can be determined by constructing a discrete reference histogram, comprising a plurality of classes relating to different ranges of values. possible for the pixel gradient orientations of the reference image, and in which in the third step, the error of the main orientation is determined according to a maximum correlation between the histogram and the histogram of reference. In the third step, the maximum correlation between the histogram and the reference histogram can be determined by calculating a measure of the correlation between the histogram and the reference histogram for a plurality of offsets of the histogram. relative to the reference histogram by an offset angle within a specified range. In the third step, the measurement of the correlation between the histogram and the reference histogram can be determined according to a probabilistic distance of Bhattacharyya, a quality index being determined according to the value of the probabilistic distance. of Bhattacharyya and the error, and / or an estimate of conformity according to the probabilistic distance of Bhattacharyya, during a fourth stage.

Other features and advantages of the present invention will become apparent from the following description of embodiments with reference to the accompanying drawings, in which:

FIG. 1 is a block diagram of the steps of a method for determining an estimate of the main orientation of a texture according to one embodiment of the invention;

FIG. 2 is a block diagram of the steps of a method for determining a main orientation error of a texture with respect to a main reference orientation, according to an embodiment of the invention;

FIG. 3a is a reproduction of an image of a texture whose main orientation is substantially equal to 5 ° in the reference frame R;

FIG. 3b is a reproduction of a reference image of a texture whose main orientation is substantially equal to 0 ° in the reference frame R;

Figure 4 is a block diagram of the steps of a method for producing and / or controlling a composite material part, according to one embodiment of the invention.

With reference to FIG. 4, a method for producing and / or controlling a composite material part will now be described. The composite material part is produced from at least one fabric having a surface whose texture has a main orientation O. Typically, the composite material part is composed of a plurality of superimposed fabrics so that the orientation principal of each fabric respects a predetermined sequence. The The piece may in particular be a preform adapted to be used to manufacture composite material parts, in particular by implementing a process comprising a draping step followed by a cooking step in an autoclave, a furnace or a hurry. The method is particularly adapted to be implemented by automated devices, such as robots provided with calculation means and an image pickup device, adapted to arrange the fabrics relative to one another according to one another so that the main orientation of each tissue substantially respects a predetermined sequence.

In the remainder of the description, for purposes of illustration, it is considered the case where it is desired to obtain the main orientation information of a fabric in a given reference system so as to control or modify the positioning of said fabric. in this repository. For example, this situation can be encountered when a robot moves a fabric from a storage location to a manufacturing area of composite material parts, and that said fabric is to be positioned in a particular main orientation. However, during the manufacture of a piece of composite material, this operation can be repeated as often as necessary to obtain the final piece, typically at least as often as the number of tissues forming the final piece. The orientation of each of the tissues may be predetermined from a recorded sequence. The orientation of each of the tissues may also be predetermined from the orientation of at least one of the other tissues forming the workpiece.

The method includes a step of obtaining a first image I REQ formed by a plurality of pixels and wherein a luminance level can be determined for each pixel, representing the texture of the fabric. The image I _RE Q of the texture comprises at least luminance information. Thus, the image I _RE Q may be a digital image, generally referred to as a luminance image, in which at each pixel x, y is associated with at least one value l (x, y) corresponding to a luminance level. The image I REQ. can be a so-called gray level image, each gray value corresponding to a luminance level.

The method comprises a step 20 of determining an estimate relating to the main orientation O of the texture, as a function of the luminance information included in an image of said texture.

The method comprises a step 30 for determining a difference D between the main orientation O and a set value. The reference value is for example obtained from a predefined value or calculated according to, for example the orientation of at least one of the other tissues forming the part. Alternatively, the setpoint may be a range of acceptable values.

The method comprises a step 40a for producing the part as a function of the difference D and / or a step 40b for transmitting a control signal as a function of the difference.

During step 40a, the position of the fabric can be modified so as to reduce the deviation D to a value lower than a predefined threshold. For this purpose, steps 10 to 40 may be repeated as often as necessary.

During step 40b, the control signal can be generated so as to enable the signaling to a control operator or a quality control device of a potential defect of orientation of the tissue, when the deviation E is greater than a tolerance value greater than a predefined threshold.

With reference to FIG. 1 and FIG. 3a, a first method for determining an estimate of the main orientation of a texture according to one embodiment of the invention will now be described. The first method is particularly adapted to allow the determination of an estimate relating to the main orientation O of the texture, as a function of the luminance information included in an image of said texture, in the production and / or control method. of a piece made of composite materials.

The first method is to determine a principal orientation in a texture based on the luminance information included in an image of said texture. More particularly, the first method aims at determining a principal orientation in the texture as a function of a spatial derivative of the luminance information included in the image of said texture. In the remainder of the description, the term "gradient" refers to the spatial derivative of the luminance information included in the image of said texture.

The first method is adapted in particular to estimate, in real time, the main orientation O of a texture, from an image IREQ of said texture, the main orientation O being determined, for a measurement interval M relatively to a reference frame R of the image IREQ. Typically, the measurement interval M is [-5 ° .. 5 °]. The first method also makes it possible to calculate a quality index Q of the estimate of the main orientation O, relating to a degree of confidence or reliability of the estimate of the main orientation O.

The image IREQ. of the texture comprises at least luminance information. Thus, the image IREQ may be a digital image, generally referred to as a luminance image, in which at each pixel x, y is associated at least with a value l (x, y) corresponding to a luminance level. The image IREQ can thus be a so-called gray level image, each gray value corresponding to a luminance level.

In a first optional step S110, a filtered image I IF LT is determined, based on the luminance information I (x, y) of the image IREQ, by implementing a noise reduction method. The filtered image I FI LT can be determined in particular, by reducing or eliminating the components relating to luminance information l (x, y) of the image IREQ. whose spatial frequency is greater than a cutoff frequency FC. A spatial convolutional filter with a Gaussian core can be used for this purpose. At the end of the first optional step S110, the filtered image I FI LT thus obtained comprises at least the filtered luminance information IFILT (x, y) of the image IREQ.

Advantageously, the first method comprises a second optional step S120 and a third optional step S130.

During the second optional step S120, for each pixel x, y of the filtered image I FI LT or of the image IREQ, a score SH relating to the belonging of said pixel to a luminance contour is determined. In particular, a Harris score is calculated from luminance information l (x, y) or filtered luminance information IFILT (x, y). The algorithm known as "Harris and Stephen" for calculating SH (x, y) is described in detail in Harris, C. & Stephens, M. (1988), "A Combined Corner and Edge Detector", in 'Proceedings of the 4th Alvey Vision Conference', pp. 147-151. At the end of the second optional step S120, a first pseudo-image SH is obtained, in which the value of each pixel corresponds to the Harris score associated with the pixel whose coordinates correspond in the filtered image I FI LT or in the image. IREQ image. The Harris score SH (x, y) of an x, y pixel of the filtered image I F1 LT or of the image IREQ is all the more important since the proximity of said pixel x, y is large in terms of luminance with a corner of the I FI LT filtered image or the IREQ image. The term "corner", also commonly referred to as "corner", refers to an area of an image where a discontinuity in the luminance information is present, typically when a sudden change in luminance level is observed between adjacent pixels.

Advantageously, in the filtered image I FI LT or the IREQ image, where each gray value corresponds to a luminance level, the corners taken into account to determine the Harris score SH (x, y) of the pixel x, y , can be the corners corresponding to a break between gray levels in a single direction, the Harris score SH (x, y) of the pixel x, y, then being less than zero. Thus, frank breaks in a single direction can be favored.

In the second optional step S120, for each pixel x, y of the filtered image IFILT or the image IREQ, an estimate EGRAD (x, y) of the amplitude the luminance gradient can be determined from the luminance information l (x, y) or the filtered luminance information IFILT (x, y). A description of a method for determining Γ EGRAD estimate (x, y) is detailed in the document I. Sobel and G. Feldman, entitled "A 3x3 isotropic gradient operator for image processing", presented at the conference "Stanford Artificial Project ", 1968.

In the optional third step S130, for each pixel x, y of the IFILT filtered image or of the IREQ image, a probability of importance p (x, y) is determined, using the Harris SH score. (x, y) said pixel x, y or the estimate EGRAD (x, y) of the gradient amplitude of said pixel x, y. For this, a calibration method can be implemented to obtain the probability of importance p (x, y), for each pixel x, y of the IFILT filtered image or of the image IREQ., From the score Harris SH (x, y) of said pixel or estimate EGRAD (x, y) of the gradient of said pixel x, y.

The calibration method may for example comprise a step in which a sigmoid function, whose parameters have been determined empirically, is applied to the first pseudo-image SH, as described for example in the document "Platt, J. ( 1999). Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in large margin classifiers ", 10 (3), pages 61-74. At the output of the calibration method, a second pseudo-image PI is obtained, in which the value of each pixel corresponds to the probability of importance p (x, y) associated with the pixel whose coordinates correspond in the IFILT filtered image or in the image IREQ.

In a fourth step S140, for each pixel x, y of the IFILT filtered image or of the IREQ image, an orientation OG (x, y) of gradients relative to the luminance level is determined. The orientation OG (x, y) of gradients is densely determined, i.e. using the information of all x, y pixels of said image. For example, to determine the orientation OG (x, y) of the pixel x, y, we calculate:

A first response of a first Sobel-type convolution filter applied at the luminance level in a first direction; A second response of a second Sobel core-type convolution filter applied at the luminance level, in a second direction orthogonal to the first direction.

The orientation OG (x, y) of the pixel x, y is then obtained by forming a vector whose horizontal component corresponds to the first response and the vertical component corresponds to the second response. The argument of the vector thus formed is thus estimated for an interval [0; 2π]. At the end of the fourth step S140, a third pseudo-image OG is obtained, in which the value of each pixel corresponds to the orientation OG (x, y) of gradients associated with the pixel whose coordinates correspond in the image. filtered IFILT or in the image IREQ.

In a fifth step S150, an estimate of the global distribution of the gradient orientations OG (x, y) of the pixels x, y of the filtered image IFILT or of the image IREQ., Is determined. The estimation of the overall distribution of the gradients OG (x, y) is, for example, a discrete histogram H, comprising a plurality of classes relating to different ranges of possible values for the orientations of the gradients OG (x, y). The histogram H is constructed by determining:

For each pixel x, y a contribution C (x, y); and,

• for each class, a height equal to the sum of the contributions C (x, y) of all the pixels whose orientation OG (x, y) is included in said class.

The contribution C (x, y) of each pixel x, y can be chosen constant, for example equal to 1.

Advantageously, in the case where the second pseudo-image PI is available, the contribution C (x, y) of each pixel x, y may be a function of the probability of importance p (x, y) associated with said pixel.

In a sixth step S160, the main orientation O is determined as a function of the estimation of the global distribution of the gradient orientations, the pixels x, y of the IFILT filtered image or the IREQ image. In particular, the main orientation O can be determined by identifying, in the histogram H, the class whose height is maximum.

With reference to FIGS. 2, 3a and 3b, a second method for determining a principal orientation error E of a texture with respect to a main reference orientation, according to an embodiment of the invention, will now be described. The second method is particularly adapted to allow the determination of an estimate relating to the main orientation O of the texture, as a function of the luminance information included in an image of said texture, in the production and / or control process. a piece of material composites. The second method is adapted to determine the main orientation error E in the texture as a function of the luminance information included in an IREQ image of said texture, and according to the luminance information included in a reference image IREF. The second method also makes it possible to determine a quality index Q of the estimate of the error E, relating to a degree of confidence or reliability of the estimate of the error E, and a conformity estimate CF. In the remainder of the description, the term "gradient" refers to the spatial derivative of the luminance information included in the image of said texture. From an image of the textured surface of said piece, and an image of a textured surface considered to be a reference to be reached in terms of principal orientation of the fibers of the part to be manufactured, it is then possible to determine an estimate of the error E and undertake the corrective steps that may then be necessary during the step 40a.

The image IREQ. of the texture comprises at least luminance information. So the image IREQ. may be a digital image, generally referred to as a luminance image, in which at each pixel x, y is associated with at least one value l (x, y) corresponding to a luminance level. The image IREQ can thus be a so-called gray level image, each gray value corresponding to a luminance level.

The reference image IREF is an image of a textured surface considered to be a reference to which the IREQ image should be compared in terms of principal orientation of the fibers. The reference image IREF comprises at least luminance information. Thus, the reference image IREF may be a digital image, generally referred to as a luminance image, in which at each pixel x, y is associated with at least one value l (x, y) corresponding to a luminance level. The reference image IREF can thus be a so-called gray level image, each gray value corresponding to a luminance level.

In a first optional step S210, a filtered image I IF LT is determined, based on the luminance information I (x, y) of the IREQ image, by implementing a noise reduction method. The filtered image I FI LT can be determined in particular, by reducing or eliminating the luminance information components l (x, y) of the IREQ image whose spatial frequency is greater than a cutoff frequency FC. A spatial convolutional filter with a Gaussian core can be used for this purpose. At the end of the first optional step S210, the filtered image I FI LT thus obtained comprises at least the filtered luminance information IFILT (x, y) of the image IREQ. During the first optional step S210, a IFILT-REQ reference filtered image is determined, based on the luminance information I (x, y) of the reference image IREF, by implementing a noise reduction method. The filtered reference image IREF can be determined in particular by reducing or eliminating the luminance information components l (x, y) of the reference image IREF whose spatial frequency is greater than a cutoff frequency FC. A spatial convolutional filter with a Gaussian core can be used for this purpose. At the end of the first optional step S210, the filtered IFILT-REF reference image thus obtained comprises at least the filtered luminance information IFILT-REF (x, y) of the reference image IREF.

Advantageously, the second method comprises a second optional step S220 and an optional third step S230.

During the second optional step S220, for each pixel x, y of the IFILT filtered image or of the IREQ image, a Harris score is calculated, based on the luminance information l (x, y) or filtered luminance information IFILT (x, y). The algorithm known as "Harris and Stephen" for calculating SH (x, y) is described in detail in Harris, C. & Stephens, M. (1988), "A Combined Corner and Edge Detector", in 'Proceedings of the 4th Alvey Vision Conference', pp. 147-151. At the end of the second optional step S220, a first pseudo-image SH is obtained, in which the value of each pixel corresponds to the Harris score associated with the pixel whose coordinates correspond in the IFILT filtered image or in the image The Harris score SH (x, y) of an x, y pixel of the IFILT filtered image or of the IREQ image is all the more important as the proximity of said pixel x, y is large in terms of luminance with a corner of the IFILT filtered image or the IREQ image. The term "corner", also commonly referred to as "corner", refers to an area of an image where a discontinuity in the luminance information is present, typically when a sudden change in luminance level is observed between adjacent pixels.

Advantageously, in the IFILT filtered image or the IREQ image, where each gray value corresponds to a luminance level, the corners taken into account to determine the Harris score SH (x, y) of the pixel x, y, may to be the corners corresponding to a break between gray levels in a single direction, the Harris SH score (x, y) of the pixel x, y, then being less than zero. Thus, frank breaks in a single direction can be favored.

In the second optional step S220, for each pixel x, y of the filtered image IFILT or the image IREQ, an estimate EGRAD (x, y) of the amplitude of the luminance gradient can be determined from luminance information l (x, y) or filtered luminance information IFILT (x, y).

During the second optional step S220, for each pixel x, y of the reference IFILT-REF reference image or of the reference image IREF, a Harris SH-REF (x, y) score is calculated, at from luminance information IREF (x, y) or filtered luminance information IFILT-REF (x, y). At the end of the second optional step S220, a first pseudo-reference image SH-REF is obtained, in which the value of each pixel corresponds to the Harris score associated with the pixel whose coordinates correspond in the filtered reference image. IFILT-REF or in the reference image IREF. The Harris SH-REF score (x, y) of an x, y pixel of the IFILT-REF reference image or of the reference image IREF is all the more important as the proximity of said pixel x, y is large in terms of luminance with a corner of the IFILT-REF reference filtered image or the IREF reference image.

Advantageously, in the IFILT-REF reference filtered image or the reference image IREF, where each gray value corresponds to a luminance level, the corners taken into account to determine the Harris SH-REF score (x, y ) of the pixel x, y, may be the corners corresponding to a break between gray levels in a single direction, the Harris score SH (x, y) of the pixel x, y, then being less than zero. Thus, frank breaks in a single direction can be favored.

As an alternative to Harris score calculations, in the second optional step S220, for each pixel x, y of the reference IFILT-REF reference image or the IREF reference image, an estimate EGRAD-REF ( x, y) of the luminance gradient amplitude can be determined from the luminance information IREF (x, y) or the filtered luminance information IFILT-REF (x, y).

In the third optional step S230, for each pixel x, y of the filtered image I FI LT or of the image IREQ., A probability of importance p (x, y) is determined, using the score of Harris SH (x, y) of said pixel x, y or the estimate EGRAD (x, y) of the gradient amplitude of said pixel x, y. For this, a calibration method can be implemented to obtain the probability of importance p (x, y), for each pixel x, y of the filtered image I FI LT or of the image IREQ., From Harris score SH (x, y) of said pixel or EGRAD estimate (x, y) of the gradient amplitude of said pixel x, y.

The calibration method may for example comprise a step in which a sigmoid function, whose parameters have been determined empirically, is applied to the first pseudo-image SH. At the output of the calibration method, a second pseudo-image PI is obtained, in which the value of each pixel corresponds to the probability of importance p (x, y) associated with the pixel whose coordinates correspond in the filtered image IFILT or in the image IREQ.

In the third optional step S230, for each pixel x, y of the IFILT-REF reference image or of the reference image IREF, a probability of importance pREF (x, y) is determined, using Harris SH-REF (x, y) score of said pixel x, y or EGRAD-REF estimate (x, y) of the gradient amplitude of said pixel x, y. For this, a calibration method can be implemented to obtain the probability of importance pREF (x, y), for each pixel x, y of the IFILT-REF reference filtered image or of the reference image IREF from the Harris SH-REF (x, y) score of said pixel or from the EGRAD-REF (x, y) estimate of the gradient amplitude of said pixel x, y.

The calibration method may for example comprise a step in which a sigmoid function, the parameters of which have been determined empirically, is applied to the first pseudo-image SH-REF. At the output of the calibration method, a second pseudo-image PI-REF is obtained, in which the value of each pixel corresponds to the probability of importance pREF (x, y) associated with the pixel whose coordinates correspond in the filtered image reference IFILT-REF or in the reference image IREF.

In a fourth step S240, for each pixel x, y of the IFILT filtered image or the IREQ image, an orientation OG (x, y) of gradients relative to the luminance level is determined. The orientation OG (x, y) of gradients is densely determined, i.e. using the information of all x, y pixels of said image. For example, to determine the orientation OG (x, y) of the pixel x, y, we calculate:

A first response of a first Sobel-type convolution filter applied at the luminance level in a first direction;

A second response of a second Sobel core-type convolution filter applied at the luminance level, in a second direction orthogonal to the first direction.

The orientation OG (x, y) of the pixel x, y is then obtained by forming a vector whose horizontal component corresponds to the first response and the vertical component corresponds to the second response. The argument of the vector thus formed is thus estimated for an interval [0; 2π]. At the end of the fourth step S240, a third pseudo-image OG is obtained, in which the value of each pixel corresponds to the orientation OG (x, y) of gradients associated with the pixel whose coordinates correspond in the image. filtered IFILT or in the image IREQ .. During the fourth step S240, for each pixel x, y of the IFILT-REF reference image or of the reference image IREF, an orientation OG-REF (x, y) of gradients relative to the luminance level is determined. The orientation OG-REF (x, y) of gradients is determined densely, that is to say using the information of all pixels x, y of said image. For example, to determine the orientation OG-REF (x, y) of the pixel x, y, we calculate:

The orientation OG-REF (x, y) of the pixel x, y is then obtained by forming a vector whose horizontal component corresponds to the first response and the vertical component corresponds to the second response. The argument of the vector thus formed is thus estimated for an interval [0; 2π]. At the end of the fourth step S240, a third pseudo-image OG-REF is obtained, in which the value of each pixel corresponds to the orientation OG-REF (x, y) of gradients associated with the pixel whose coordinates correspond to in the IFILT-REF reference filtered image or in the IREF reference image.

In a fifth step S250, an estimate of the global distribution of the orientations of the gradients OG (x, y) of the pixels x, y of the filtered image IFILT or of the image IREQ., Is determined. The estimation of the global distribution of the gradients OG (x, y) is, for example, a discrete histogram H, comprising a plurality of classes relating to different ranges of possible values for the orientations of the gradients OG (x, y). The histogram H is constructed by determining:

For each pixel x, y a contribution C (x, y); and,

During the fifth step S250, a reference estimate of the global distribution of the orientations of the gradients OG-REF (x, y) of the pixels x, y of the reference filtered image IFILT-REF or of the reference image IREF, is determined. The reference estimate of the global gradient distribution OG-REF (x, y) is, for example, a discrete HREF reference histogram comprising a plurality of classes relating to different ranges of possible values for the OG-REF gradient orientations ( x, y). The reference histogram HREF is constructed by determining:

• for each pixel x, y a contribution CREF (x, y); and,

• for each class, a height equal to the sum of the CREF (x, y) contributions of all the pixels whose orientation OG-REF (x, y) is included in said class.

The contribution CREF (x, y) of each pixel x, y can be chosen constant, for example equal to 1.

Advantageously, in the case where the second pseudo-image PI-REF is available, the contribution CREF (x, y) of each pixel x, y may be a function of the probability of importance pREF (x, y) associated with said pixel.

In a sixth step S260, the main orientation 0 is determined according to:

Estimating the global distribution of the gradient orientations, the pixels x, y of the filtered image F1 LT or the image IREQ; and,

Reference estimation of the global distribution of the orientations of the gradients OG-REF (x, y) of the pixels x, y of the reference filtered image IFILT-REF or of the reference image IREF.

In particular, a maximum correlation between the histogram H and the reference histogram HREF can be sought. Thus, it is possible, for example, to evaluate a measure of the correlation between the histogram H and the reference histogram HREF, by varying the offset of the histogram H with respect to the reference histogram HREF by an angle shift in steps of 0.1 ° between -10 ° and 10 °. The measurement of the correlation between the histogram H and the reference histogram HREF is, for example, determined as a function of the probabilistic distance of Bhattacharyya, described in particular in the document entitled Kailath, T., (1967), "The Divergence and Bhattacharyya". Distance Measures in Signal Selection ", IEEE Transactions on Communication Technology, Vol. 15, No. 1, 1967, p. 52-60. The estimate of the error E is then equal to the offset angle for which a maximum correlation is observed. The quality index Q. is then a function of the value of the Bhattacharyya distance associated with the estimate of the error E.

In a seventh step S270, a conformity estimate CF based on a statistical distance between the histogram H and the reference histogram HREF is determined. Thus, it is possible, for example, to determine the probabilistic distance of Bhattacharyya between the histogram H and the reference histogram HREF. The conformity estimate CF is then obtained as a function of the probabilistic distance of Bhattacharyya between the histogram H and the reference histogram HREF, typically by subtracting 1 said probabilistic distance from Bhattacharyya.

Claims

A method for producing and / or controlling a composite material part formed from at least one fabric having a surface whose texture has a main orientation (0), characterized in that it comprises the following steps :

obtaining (10) a first image (IREQ.) formed by a plurality of pixels and wherein a luminance level can be determined for each pixel, representing the texture of the tissue;

determining (20) an estimate relating to the principal orientation (O) of the texture,

In a first step (S140; S240), determining, for each pixel of the first image, an orientation (OG (x, y)) of gradients relative to the luminance level of said pixel;

In a second step (S150; S250), by determining an estimate of an overall distribution of the gradient orientations of the pixels of the first image;

In a third step (S160; S260), determining the main orientation as a function of the estimate of the overall distribution of the gradient orientations of the pixels of the first image;

determining (30) a deviation (D) between the estimate relating to the main orientation (O) and a set value;

producing (40a) the workpiece according to the deviation and / or outputting (40b) a control signal depending on the deviation.

The method according to claim 1, wherein, during the first step (S140; S240), for each pixel of the first image, the orientation (OG (x, y)) of gradients relative to the luminance level of said pixel is obtained in:

A second response of a second Sobel-type convolution filter applied at the luminance level, in a second direction orthogonal to the first direction; • calculating the argument of a vector whose horizontal component corresponds to the first response and the vertical component corresponds to the second response.

The method according to any one of claims 1 to 2, wherein in the second step (S150; S250), estimating the overall distribution of the gradient orientations of the pixels of the first image is determined by constructing a discrete histogram (H), comprising a plurality of classes relating to different ranges of possible values for the orientation of the gradients of the pixels of the first image.

The method of claim 3, wherein the histogram (H) is constructed by determining:

For each pixel of the first image, a contribution; and,

· For each class, a height equal to the sum of the contributions of all the pixels of the first image whose orientation is included in said class.

5. The method of claim 4, wherein in the third step, the main orientation is determined by identifying, in the histogram, the class whose height is maximum.

6. Method according to any one of claims 4 or 5, further comprising, prior to the first step:

A step (120), during which, for each pixel of the first image, a score (SH) relating to the membership of said pixel in a luminance contour is determined;

A step (S130), during which, for each pixel of the first image, a probability of importance (p (x, y)) is determined, using the score of said pixel;

and wherein, in the second step, the contribution of each pixel is determined according to the importance probability associated with said pixel.

The method of claim 6, wherein, for each pixel of the first image, the score is a calculated Harris score, based on the luminance levels of the pixels of the first image.

The method of claim 7, wherein the corners of the first image used to calculate the score of each pixel of the first image correspond to a break between luminance levels of the first image in a single direction.

9. The method of claim 6, wherein for each pixel of the first image, the score is an estimate of the luminance gradient amplitude from the luminance levels of the pixels of the first image.

The method according to any one of claims 6 to 9, wherein for each pixel of the first image, the probability of importance is determined using a sigmoid function and the score of said pixel.

The method according to any one of claims 1 to 10, further comprising, prior to the third step, a step (S110; S210), in which the luminance level of each pixel of the first image is filtered from to reduce the noise present in the luminance information of the first image.

The method of any one of claims 1 to 11, wherein:

During the first step (S240), for each pixel of a reference image (IREF) of a textured surface, said reference image being formed by a plurality of pixels and in which a luminance level can be determined for each pixel, an orientation (OG-REF (x, y)) of gradients relative to the luminance level of said pixel is determined;

During the second step (S250), a reference estimate of an overall distribution of the gradient orientations of the pixels of the reference image is determined;

During the third step (S260), an error (E) of the main orientation is determined according to the estimate of the overall distribution of the gradient orientations of the pixels of the first image and according to the estimate reference of the global distribution of the gradient orientations of the pixels of the first image; and wherein the difference between the estimate of the main orientation (0) and the set value is determined according to the error (E) of the main orientation.

The method of claim 3 and claim 12, wherein, in the second step (S250), the reference estimate of the overall distribution of gradient orientations of the pixels of the reference image is determined by constructing a discrete reference histogram (HREF), having a plurality of classes relating to different ranges of possible values for the pixel gradient orientations of the reference image, and wherein in the third step (S260), Error (E) of the main orientation is determined based on a maximum correlation between the histogram (H) and the reference histogram (HREF).

The method of claim 13, wherein, in the third step (S260), the correlation maximum between the histogram (H) and the reference histogram (HREF) is determined by calculating a measure of the correlation. between the histogram (H) and the reference histogram (HREF) for a plurality of offsets of the histogram (H) relative to the reference histogram (HREF) at an offset angle within a specified range.

The method of claim 14, wherein, in the third step, measuring the correlation between the histogram (H) and the reference histogram (HREF) is determined according to a probabilistic distance of Bhattacharyya. , a quality index (Q.) being determined according to the value of the probabilistic distance of Bhattacharyya and the error (E), and / or an estimate of conformity (CF) according to the probabilistic distance of Bhattacharyya, to during a fourth stage (S270).