PROCEEDS FROM THREE-DIMENSIONAL CONTROL

Abstract

A method for three-dimensional inspection of a part from multiple radiographic images using an image analysis device comprising electronic circuitry adapted to implement the three-dimensional inspection method; the method comprises at least the following steps: - Step 1, Acquire a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part; - Step 3, Align each image to a view of a digital model of the part; - Step 4, Detect indications on each image and position the detected indications with bounding boxes; - Step 5, Construct a model of the multi-view geometry using the alignment of each image to a respective view of the digital model of the part;- Step 6, Consolidate the detected features with bounding boxes, using the multi-view geometry model by constructing, in step 6.1, a set of candidate groups, each group containing projections of a potential three-dimensional feature, and selecting, in step 6.2, a candidate group to retain; - Step 7, Locate each three-dimensional feature; - Step 8, Characterize the dimensions of each feature in a relevant image selected from the set of images using the bounding boxes used to position the detected features; - Step 9, Use the characterization of each feature for further processing by a third-party device.

Description

Title of the invention: THREE-DIMENSIONAL CONTROL METHOD technical field

[0001] The invention relates to the field of non-destructive testing of parts. STATE OF PRIOR ART

[0002] In the field of non-destructive testing, particularly of metal parts, it is well known to use computer vision methods and conventional image processing methods or methods based on statistical learning methods.

[0003] A non-destructive testing method for metal parts is known, which uses a module for the precise classification and localization of indications. It is specified that in this document, an indication is understood to be a characteristic of an image that requires a penalty. An indication can be penalized as an anomaly or as benign. The module is specific to each area of ​​the part that presents distinctive characteristics. This method is particularly suitable for inspecting heterogeneous parts. This method uses neural network training with two-dimensional radiographic images to localize the image indications.

[0004] Thus, it is known to automatically recognize features on a two-dimensional image. However, the transition from characterizing a feature on a two-dimensional image to characterizing that same feature in a three-dimensional aeronautical part is traditionally performed manually by operators expert in X-ray image analysis. Traditionally, this characterization follows a procedure described in a work instruction established for the part's certification. Typically, multiple images are taken of each part. Each image describes the X-ray transmission properties of the part's material from a different viewpoint. In this document, "view" refers to the viewpoint of the inspected part from which its radiographic image is acquired.Once all the multi-view images have been acquired, the operators analyze all the images to retrieve all the indications. Next, the operators determine which indications in the images correspond to projections of the same indication within the room (also called three-dimensional indications). Then, the operator determines the location of the three-dimensional indications within the room, and they are also characterized in their most relevant projection.

[0005] This method relies mainly on the expertise and talent of the operators, and requires a lot of time.

[0006] In this context, it is necessary to provide a three-dimensional control method for a part which allows for the automatic characterization of a three-dimensional indication from multi-view images in two dimensions, to help in the reliable and rapid detection of indications in aeronautical parts. Description of the invention

[0007] To this end, according to a first aspect, a three-dimensional inspection method for a part is proposed, based on several radiographic images, using an image analysis device comprising electronic circuitry adapted to implement the three-dimensional inspection method. The method comprises at least the following steps: - (Step 1) Acquire a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part; - (Step 3) Align each image with a view of a digital model of the part; - (Step 4) Detect indications on each image and position the detected indications with bounding boxes; - (Step 5) Construct a multi-view geometry model using the alignment of each image on a respective view of the digital model of the part; - (Step 6) Consolidate the detected indications with bounding boxes, using the multi-view geometry model by constructing (step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (step 6.2) a candidate group to retain; - (Step 7) Locate each three-dimensional indication; - (Step 8) Describe the dimensions of each detail in an image relevant selected from all images based on bounding boxes used to position detected indications; - (Step 9) Use the characterization of each indication for further treatment by a third-party device.

[0008] According to a particular arrangement, alignment step 3 includes estimating (step 3.1) an optimal rigid body transformation of each view of the digital model of the part with respect to the source-detector system to obtain an image of the projection of each view of the digital model.

[0009] According to a particular arrangement, a projection operator is used to carry out step 3.1, the operator is a matrix of the form . [ R [t ] ' with K a matrix described by intrinsic projection parameters, and with R and t respectively the rotation and translation of each view of the numerical model.

[0010] According to a particular provision, step 3 includes defining (step 3.2) a similarity metric between the calculated image of each view of the digital model and each acquired image.

[0011] According to a particular provision, step 6.1 of constructing a set of candidate groups includes analyzing a location of indications.

[0012] According to a particular provision, step 6.2 of selecting a candidate group includes selecting candidate groups containing projections of the same three-dimensional indication.

[0013] According to a particular arrangement, the localization (step 7) includes determining an average of point clouds per candidate group containing several indications.

[0014] According to a particular arrangement, the characterization step includes selecting (step 8.1) a relevant image to estimate a length and width of an indication, the relevant image being selected based on a three-dimensional location of the indication.

[0015] According to another aspect, an image analysis device is proposed which includes electronic circuitry to implement a three-dimensional control method for a part from several radiographic images, the method comprising at least the following steps: (Step 1) Acquire a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part; - (Step 3) Align each image with a view of a digital model of the part; - (Step 4) Detect indications on each image and position the detected indications with bounding boxes; - (Step 5) Construct a multi-view geometry model using the alignment of each image on a respective view of the digital model of the part; - (Step 6) Consolidate the detected indications with bounding boxes, using the multi-view geometry model by constructing (step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (step 6.2) a candidate group to retain; - (Step 7) Locate each three-dimensional indication; - (Step 8) Characterize the dimensions of each indication in a relevant image selected from the set of images using bounding boxes to position the detected indications; - (Step 9) Use the characterization of each indication for further treatment by a third-party device.

[0016] According to another aspect, a computer program product is proposed comprising program code instructions for executing the process according to the invention.

[0017] According to another aspect, a non-transient storage medium is proposed on which is stored a computer program comprising program code instructions to execute the process according to the invention, when said instructions are read from said non-transient storage medium and executed by a processor.

[0018] BRIEF DESCRIPTION OF THE DRAWINGS.

[0019] The features of the invention mentioned above, as well as others, will become clearer upon reading the following description of at least one exemplary embodiment, said description being made in relation to the accompanying drawings, among which:

[0020] [Fig.1] schematically illustrates the sequence of a control process;

[0021] [Fig.2] schematically illustrates the construction of a multi-view geometry;

[0022] [Fig.3] schematically illustrates an epipolar constraint for the projection of a point in a three-dimensional room;

[0023] [Fig.4] schematically illustrates a cutout of a digital model of a part;

[0024] [Fig.5] illustrates a candidate group of four indications;

[0025] [Fig.6] illustrates a first phase of creation of candidate groups;

[0026] [Fig.7] illustrates a second phase of creation of candidate groups;

[0027] [Fig.8] illustrates the construction of a three-dimensional score;

[0028] [Fig.9] illustrates the use of a three-dimensional score for a candidate group;

[0029] [Fig. 10] illustrates a proposed solution for dimensional characterization;

[0030] [Fig. 11] illustrates the application of a Huber regression model to estimate a length;

[0031] [Fig.12] illustrates the application of a second-degree polynomial regression model for estimating a width;

[0032] [Fig. 13] schematically illustrates a computer system adapted to implement the process.

[0033] DETAILED DESCRIPTION OF IMPROVEMENTS

[0034] Three-dimensional control method.

[0035] With reference to [Fig. 1], according to a first aspect, a method is proposed for three-dimensional control of a part from several radiographic images. by an image analysis device comprising electronic circuitry 200 adapted to implement the three-dimensional control method 100. The method 100 comprises at least the following steps: - (Step 1) Acquire a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part; - (Step 3) Align each image with a view of a digital model of the part; - (Step 4) Detect indications on each image and position the detected indications with bounding boxes; - (Step 5) Construct a multi-view geometry model using the alignment of each image on a respective view of the digital model of the part; - (Step 6) Consolidate the detected indications with bounding boxes, using the multi-view geometry model by constructing (step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (step 6.2) a candidate group to retain. - (Step 7) Locate each three-dimensional indication; - (Step 8) Describe the dimensions of each detail in an image relevant selected from all images based on bounding boxes used to position detected indications; - (Step 9) Use the characterization of each indication for further treatment by a third-party device.

[0036] Step 1 - Image Acquisition

[0037] As previously stated, the control method 100 includes a first step (step 1) of acquiring a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part.

[0038] According to one embodiment, the images are acquired using an X-ray radiography method. Thus, in the example presented here, the X-ray acquisition takes place in a tomography booth. The tomography booth consists of an X-ray generator that emits a beam passing through a part to be examined before being analyzed, after attenuation, by an X-ray detection system. An acquired image, or an image obtained after calibration for any inhomogeneities in the detector and / or extraction of any background information, is called a projection. It should be noted that calibration can be performed for a wide variety of of phenomena. A set of images of the same view is acquired without moving the room and the images are averaged in order to reduce acquisition noise.

[0039] According to a particular arrangement, several X-ray images are obtained from different views of the part in order to obtain an optimized view of the part to be inspected. In other words, several images are acquired from different orientations around the part. The views (i.e., the different orientations) are selected to have an optimized signal-to-noise ratio, enabling the detection of indications with the lowest radiographic contrasts.

[0040] Step 2 - Pretreatments

[0041] According to a particular provision, preprocessing (step 2) may optionally be applied to the images to facilitate subsequent analysis:

[0042] - Filtering: use of filters to enhance characteristic patterns of Image guidance and noise reduction.

[0043] - Masking a region of interest on each image. According to a layout In particular, masking is achieved through automated detection of air and collimator.

[0044] - Data normalization: in order to facilitate learning and obtain a faster convergence, the data is normalized to have a mean of 0 and a standard deviation of 1.

[0045] Step 3 - Alignment

[0046] As previously stated, the inspection process then includes a step 3 consisting of aligning each image, which describes the X-ray transmission properties of the material for a view of the inspected part, with a corresponding view of a digital model of the part. The alignment aims to match each image with the digital model of the part.

[0047] It is specified that by view of the digital model of the part, it is understood a viewpoint of the digital model of the part, for example, the digital model of the ideal part.

[0048] To this end, according to a particular arrangement, alignment step 3 comprises estimating (step 3.1) an optimal rigid body transformation of each view of the digital model of the part with respect to the source-detector system to obtain an image by projection of each view of the digital model. This image is referred to herein as the computed image. This transformation is described by a rotation and a translation in three-dimensional space.

[0049] In a particularly advantageous manner, a projection operator is used to carry out step 3.1. According to a particular arrangement, the projection operator is a matrix of the form ar R ], with K a matrix described by Intrinsic projection parameters, such as the focal length and position of the projected source in each image, and R and t, respectively the rotation and translation of each view of the digital model of the part, thus its rigid body transformation. The calculated image is then obtained by a digital radiography calculation algorithm, for example a ray tracing algorithm or the Siddon Jacobs algorithm, from the view of the digital model of the part, the digital model of the part, and the projection operator.

[0050] In other words, to estimate the transformation of the digital model view of the part, a projection operator is required to project the three-dimensional information of the digital model onto a two-dimensional plane of a two-dimensional image. The matrix K can be calibrated beforehand using calibration procedures specific to part inspection, or it can be constructed from intrinsic parameters prescribed within a part inspection range.

[0051] Advantageously, alignment step 3 also includes defining (step 3.2) a similarity metric between the calculated image of each view of the digital model and each image from a view of the inspected part. It is specified that a similarity metric is a value that describes the similarity between two images. This similarity can, for example, be evaluated from a difference in gray levels, or from the correlation of the two images, or from information that one image contains about the other. In other words, in step 3.2, a similarity metric is defined between the projected information of the digital model and the information in the view of the part to be inspected.In a particularly ingenious way, it is the optimization of this similarity metric with respect to the rotation and translation parameters of the digital model of the part that allows the alignment of a view of the inspected part with a view of the digital model of the part, in other words, the alignment of an acquired image with the digital model of the part.

[0052] According to a particular arrangement, a first realization of the alignment uses a similarity measure between a projection of three-dimensional salient points onto the two-dimensional plane of the image and corresponding points estimated in the image. In this method, the three-dimensional salient points are pre-selected, view by view, in the digital model of the part. The projection is performed view by view from the matrix K and a set of initial parameters for R and t. This initialization can be based on parameters prescribed by a control range, for example, the pose of the part prescribed for image acquisition. The estimation of corresponding points in the image can be performed by a local correlation between an image calculated from a view of the digital model and the corresponding acquired image.The image can, for example, be calculated from the digital model of the part with parameters prescribed by a control range. using the projection operator defined for the view of the digital model. The so-called local correlation corresponds to a correlation operator, for example, the normalized two-dimensional cross correlation, between a neighborhood centered around a salient point projected into the computed image and a prescribed search region around the same point in the acquired image. The estimated correspondence of the salient point in the image corresponds in this case to the position where the local correlation is maximum. Once the salient points have been projected into the computed image, and the corresponding points have been estimated in the acquired image, the optimization consists of minimizing their distance with respect to the parameters of a rigid body transformation of the three-dimensional positions of the salient points. This can be done, for example, with an implementation of the PnP (Perspective and Point) method.

[0053] According to another embodiment, it is possible to use a similarity measure between an image calculated from a view of the digital model of the part and the image of the part to be inspected. In this case, the optimization consists of maximizing the similarity between these two projections with respect to the parameters of a rigid body transformation of the digital model of the part. One implementation of this optimization consists of using the procedure that minimizes the residual (difference) between the image calculated, for example by a simulation method, and the acquired image. Another implementation of this optimization consists of using a similarity measure such as those derived from information theory, for example, implementations of mutual information metrics based on entropy. The estimation of the similarity measure can be restricted to a prescribed region of interest. The optimization is performed view by view from K and a set of initial parameters.In one embodiment, the initial parameters may be derived from the parameters prescribed by the control range, for example, the positioning of the part prescribed for image acquisition. In another embodiment, the initial parameters may be derived from the result of the first realization of the alignment described in the preceding paragraph.

[0054] Step 4 - Detection of indications

[0055] As previously stated, the method includes (step 4) detecting indications on each image and positioning the detected indications with bounding boxes.

[0056] It is specified that by bounding box, it is understood that a rectangle encompasses the indication. A bounding box is defined by a height, a width and a position.

[0057] According to a particular arrangement, the detection of indications is carried out in two stages: a learning stage 4.1 and a prediction stage 4.2 on a production line. According to one embodiment, the learning stage 4.1 can be implemented using a neural network. According to this approach, to perform the detection of two-dimensional features, a model is learned in a supervised manner. Supervised learning involves a database of images annotated by an operator. Each feature present in an image is labeled as follows: - position of the indication with an encompassing box; - label of the indication; - dimensions of the indication (the dimension consists of the width and length of the identified indication, as will be described below this information will be used in particular in step 7 of dimensional characterization).

[0058] According to a specific provision, it is possible to label all indications with a single class or with several classes depending on the type of indication present (for example: inclusion, excess thickness, residue, or excess thickness). This labeling method allows not only the detection of indications but also their labeling according to predefined classes (e.g., inclusions, excess thickness, under-thickness). It is specified that this database can consist of images representative of the variability of images in production. Preferably, the number of images with indications to be inserted into the database is sufficient to allow the system to learn representative characteristics of each indication.

[0059] According to a particular arrangement, a database balancing phase may optionally be performed. Indeed, it may be more difficult to obtain samples of certain types of indications and / or certain areas to be inspected. A highly unbalanced number of samples can be detrimental during statistical learning. To compensate for this imbalance, transformations may be added to the images (for example, rotations, adding noise or blur, modifying contrast, changing brightness, or cropping). It is also possible to opt for the use of simulated images, provided they are representative of real data. According to another arrangement, it is possible to opt for the use of loss functions during learning that more heavily penalize errors on minority samples.

[0060] After applying the preprocessing steps mentioned in step 2, the X-ray images are acquired, and the annotations manually made by the operators (nature and two-dimensional location of each detail) are given as input to the neural network. In the example presented here, these entire images are 1024x1024 pixels in size. However, other sizes can be used.

[0061] According to a particular arrangement, the neural network used is a convolutional network designed for the classification and localization of patterns of interest on In a specific configuration, network training is performed using classical gradient descent algorithms (an optimization algorithm that finds the minimum of any convex function by progressively converging towards it). The two tasks (classification and two-dimensional localization with bounding boxes) are performed in parallel by backpropagation of two combined loss functions, where 'two' refers to the two tasks performed in parallel: classification and two-dimensional localization. This process is repeated several times to iteratively update the network weights in an increasingly optimal manner.

[0062] In more detail, the procedure for updating weights (network parameters) through gradient descent algorithms generally follows the following formula:

[0063] et+1=et-a7()\(x,y,et)

[0064] Where represents the network parameters at time t, a(x, y; 0) represents the loss function specific to the problem to be solved with x the input data and y the expected output, and a represents a hyperparameter that manages the convergence rate.

[0065] For training the classification heads of the network to determine the type of target indication, traditional loss functions such as "focal loss" or cross-entropy can be used. For training the two-dimensional localization heads of the network to locate the target object in the image, functions commonly used for regression problems can be employed (e.g., Smooth L1 or L1).

[0066] Step 4.2 of prediction with the learned neural network is applied to all the parts to be inspected. Consequently, a few X-ray transmission radiographic images of the part to be inspected are used as input to the model learned in step 4.1 in order to obtain the prediction results, i.e. the classification and two-dimensional localization (with bounding boxes) of any indications present in each image.

[0067] According to a particular arrangement, the output of this network is given for each image of the part individually and there is no joint analysis of the different projections of the same part nor quantification of the position of the indications detected in the part (three-dimensional localization).

[0068] Finally, all the information found with the model is used as input for the following steps.

[0069] Step 5 - Construction of a multi-view geometry model

[0070] Next, the process includes a step 5 which consists of constructing a multi-view geometry model using the alignment of each image on a respective view of the digital model of the part.

[0071] As schematically shown in Fig. 2, the multi-view geometry of the scanner-part system is constructed, which will subsequently allow the exploitation of the epipolar constraint for the multi-view consolidation described below. This geometry is constructed from the part's pose, given by a rotation p0) and a translation modeled by view 1 of the part, estimated by step 3 of image alignment calculated with the acquired images.

[0072] According to the embodiment shown schematically in Fig. 2, the multi-view geometry is constructed such that the coordinate system of the digital model of the ideal part, described by the unit vectors j and the origin °, is the reference system. The radiographic images corresponding to each view are placed around the part. For this purpose, the orientations of its horizontal and vertical axes, and of the normal vector to the plane containing them, are calculated with respect to the coordinate system of the digital model of the ideal part. The position of the X-ray source (C) per view with respect to the coordinate system of the digital model of the ideal part is also calculated. It should be noted that the embodiment shown schematically in [Fig. 2] illustrates a three-view geometry.

[0073] Let / (yO)) be the intensity at the image position a3j)c P2 of the image of the view \ hereafter referred to as image \ Since the X-ray detector is discrete in nature, the positions x of a sensor element (pixel) are defined by the coordinates (xu, xY) of its center. Let G p2 be the projection of the three-dimensional point XG P3 onto the image 2

[0074] with pü) _ . £ p0) | j the projection matrix, ph) a matrix with the Intrinsic projection parameters of the CT scanner per view (e.g., constructed with values ​​prescribed by the control range such as focal length). The projection of X in image j, i.e., the correspondence of X in image j, must lie on a straight line called the epipolar line:

[0075]

[0076] with pü'jj the fundamental matrix of image 2 to image j. This matrix is ​​calculated from ph), pU) the projection matrices for images 2, j, and £>0) the position of the X-ray source for image 2, using for example a formalization known for stereovision.

[0077] Step 6 - Multi-view consolidation

[0078] Step 6 includes consolidating the detected indications with bounding boxes, using the multi-view geometry model by constructing (step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (step 6.2) a candidate group to retain. In other words, step 6 consists of consolidating the indications detected on each image into groups containing the image indications that correspond to the projection of the same three-dimensional indication in the room.

[0079] According to a particular arrangement, step 6.1 of constructing a set of candidate groups includes analyzing a location of the indications. This arrangement allows each candidate group to contain two-dimensional indications that are potentially projections of the same three-dimensional indication.

[0080] According to a particular provision, step 6.2 of selecting a candidate group includes selecting candidate groups comprising projections of the same indication.

[0081] Fig. 3 illustrates the problem of correspondence between boxes which encompass the image indications which correspond to the projections of the same three-dimensional indication in the part.

[0082] As previously stated, according to a particular arrangement, consolidation step 6 comprises two steps. In the first step, 6.1, a set of candidate projection groups is constructed by analyzing the location of the two-dimensional indications. In the second step, 6.2, the candidate groups containing the projections of the same three-dimensional indication are selected. The selection uses a three-dimensional proximity criterion and the location relative to the digital model of the part of the three-dimensional position estimated for each group. We will detail each of these two steps later.

[0083] Step 6.1: Selection of a candidate group:

[0084] The epipolar constraint given by the multi-view geometry and a zoning of the digital model of the ideal part are used to calculate a set of candidate groups of image features. Indeed, each candidate group contains two-dimensional features that are potentially the projections of the same three-dimensional feature in the part. Information from the digital model of the ideal part and / or a partitioning of the digital model of the part into zones of interest (see [Fig. 4]) is added to the epipolar constraint. It should be noted that in [Fig. 4], these zones are referenced Z1 to Z7.

[0085] The candidate groups are obtained from the analysis of the location of the two-dimensional features in the projections. Fig. 3 illustrates the correspondence of two projections of the same three-dimensional feature 300 in the case where the two-dimensional feature is contained by a bounding box 301. The correspondence is achieved by matching the two-dimensional points of the different images, for example, the centers 302 of the bounding boxes. For this, the distance

[0086]

[0087] The epipolar function is exploited. The epipolar line 304 of a projection of a bounding box center 305 in image 1 is drawn in image j using the fundamental matrix given by the multiview geometry, and the distance 303 to this epipolar line 304 for the box centers in image j is calculated. An epipolar confidence score is given to each center in image j. This score is calculated by the complementary error function: pd, With d the epipolar distance, j(J) a detected indication in image j, the mean, and &d the standard deviation value of a normal distribution of epipolar distances. d is calculated for a point in j(J), for example, the center of its bounding box. Pj is nonzero in the case where the points are The corresponding points are not exactly the projection of the same point in the three-dimensional coordinate system, as illustrated in Figure 5. The standard deviation &d is calibrated to the expected variability in epipolar distances. This variability is explained by the expected uncertainty in the results of an alignment chain. In one embodiment, Ud and ®d are estimated per view from the analysis of variations of the part's installation (see Section Other Methods of Implementation). An indication in image 1 is correlated with jU), an indication in image J if p(dP thd AND if the two image indications are intersected with the

[0088]

[0089]

[0090]

[0091]

[0092] projection of the same area in the digital model of the ideal room: if p(^P) > a 0) for at least one Zk (0, otherwise with thd = 0.05 the threshold for the distance d to be within the 95% confidence interval, the projection onto image 1 of the k-th zone of the digital model of the part, and ^j(i) q 2^) the intersection between the indication jU) and the zone £©. In one implementation, jCO p 2^) is defined as the number of pixels contained by the bounding box of jW and by the projection onto image 1 of the k-th zone of the model digital of the piece. Finally, the following equation is used to construct a consolidation matrix between the indications detected for a part: f for i, i {1, A}, i^j, Lmejl.....L}, l*m JU l 0, otherwise with N the number of views and L the number of indications detected for a piece. f indicates whether the 1st indication in image 1 and the nth indication in image j can be matched. Therefore, it allows for the estimation of candidate groups.

[0093] It should be noted that the consolidation matrix fjm is not symmetric: it is possible O116 f ( B® # f ( B^ ) • Furthermore, the areas in which the part is cut can be quite large. This implies that the constraints of the equation used to construct a consolidation matrix can be satisfied, but that the association is not correct (see example in [Fig. 5]). For these reasons, a final three-dimensional consolidation step is proposed.

[0094] Once the first candidate groups are obtained, all possible combinations of detection associations are calculated. This is illustrated in Figs. 6 and 7. First, the first candidate groups are obtained according to the method previously described ([Fig. 6]). According to the example presented here, it can be observed that the consolidation matrix is ​​not symmetrical: detection 1 (column “Id detection”) was associated with detection 2 (column “Match id”). However, this latter detection was associated with detections 4 and 5.

[0095] Consequently, all possible combinations of detection associations are then calculated from these first candidate groups ([Fig.7]). With reference to [Fig.7], this allows the following candidate groups to be created: {[1,2,4,5],[1,2,4],[1,2,5], [1,4,5],[2,4,5],[1,2],[1,4],[1,5],[2,4],[2,5],[4,5]}.

[0096] It should be noted that combinations where more than one bounding box of the same image is present are not accepted. For example, detections 2 and 3 cannot be present in the same group, because they are located on the same image.

[0097] Step 6.2 of selecting a candidate group:

[0098] In step 6.2 of candidate group selection, the candidate group to be retained (called the consolidated group) is selected. For this purpose, a consolidation score r (or "3D Score") is calculated for each of the candidate groups. This score r corresponds to the radius of the smallest sphere containing the closest points between all the lines extending from the three-dimensional position of the center of the two-dimensional indications in the images to the positions of the corresponding X-ray sources (see Figs. 8 and 9). The average of these points, called Xm, gives the estimated 3D position of each candidate group.

[0099] According to the example presented here, step 6.2 of consolidation begins with the group(s) containing the largest number of indications. Thus, in the example of Figs. 6 and 7, there are two groups with four indications: [1,2,4,5] and [1,3,4,5]. Among these groups, the group with the minimum 3D score is chosen (in the example presented here, this is group [1,3,4,5]), and then its 3D score is compared with a threshold. In one embodiment, the threshold is estimated by analyzing the variations of the The part is positioned (see Section Other Implementation Methods). If r < thr, this group is retained. Otherwise, the group containing the second-largest number of indications is chosen. The same procedure is followed, comparing r with thr. This is done until a candidate group is found that satisfies the constraint: r ≤ thr. If no group is retained, the detections may remain unconsolidated.

[0100] Step 7 - Localization of three-dimensional indications

[0101] As previously stated, the process then includes a step 7 consisting of locating each three-dimensional indication.

[0102] The localization (step 7) includes determining an average of point clouds per candidate group containing several two-dimensional indications, which are projections of the three-dimensional indication.

[0103] The location of the three-dimensional indications corresponds to the averages Xm (described in step 6.2) of the point clouds by consolidated group containing several two-dimensional indications (see Figs 8 and 9).

[0104] Step 8 - Characterization of dimensions

[0105] Next, the method includes (step 8) characterizing the dimensions of each indication in a relevant image selected from the set of images from the bounding boxes used to position the detected two-dimensional indications.

[0106] According to a particular arrangement, the characterization step includes identifying a relevant image to measure the length and width of an indication using the length and width of the bounding box of that indication.

[0107] Since the same feature can be visible on several images, it is necessary to identify the image from the most relevant view for performing this measurement. This image, known as the master image, is identified using the three-dimensional localization described in step 7. Indeed, the image to be used depends on the three-dimensional localization of the feature to be characterized.

[0108] Once the main image is identified, the process proceeds to a measurement step (length and width). To do this, the process exploits the strong relationship between the size of the bounding boxes (obtained in step 4) and the measurement of the indications. Depending on a particular arrangement, a regression model (e.g., linear, polynomial, or Huber model) is used. Indeed, the regression model allows us to describe the relationship between one or more independent variable(s) (e.g., the height or width of the box) and a dependent variable (e.g., the length or width of the indication in mm). Depending on a particular arrangement, a regression model is used for each dimension of the indication: a first model to estimate the length of the indication and a second model to estimate its width (see [Fig. 10]).

[0109] According to a specific arrangement, the two models are pre-fitted with annotated features (length and width) by inspectors and detected by the detection model (see Figs. 11 and 12). It should be noted that this data is obtained from step 4.1. Once the regression models are fitted, they are used in the production line on the detected features located in the main image (information obtained through the three-dimensional localization of step 7).

[0110] Step 9 - Use

[0111] The process then includes using (step 9) the characterization of each indication for further treatment by a third-party device.

[0112] Image analysis device

[0113] According to another aspect, an image analysis device is proposed comprising electronic circuitry (computer system 200) adapted to implement a process 100.

[0114] As schematically shown in [Fig. 13], the computer system 200 may include, connected by a communication bus 210: a processor 201; a random access memory 202; a read-only memory 203, for example of type ROM (Read Only Memory) or EEPROM (Electrically-Erasable Programmable Read Only Memory); a storage unit 204, such as a hard disk drive (HDD) or a storage media reader, such as an SD card reader (Secure Digital); and an input / output interface manager 205.

[0115] The processor 201 is capable of executing instructions loaded into RAM 202 from ROM 203, external memory, a storage medium (such as an SD card), or a communication network. When the computer system 200 is powered on, the processor 201 is capable of reading instructions from RAM 202 and executing them. These instructions form a computer program enabling the processor 201 to implement process 100.

[0116] All or part of the process 100 can thus be implemented in software form by executing a set of instructions by a programmable machine, for example a DSP (Digital Signal Processor) or a microcontroller, or be implemented in hardware form by a dedicated machine or component, for example an FPGA (Field Programmable Gate Array) or ASIC (Application-Specific Integrated Circuit). Generally, the computer system 200 includes electronic circuitry adapted and configured to implement, in software and / or hardware form, the process in relation to the computer system 200 in question.

[0117] Computer program product

[0118] According to another aspect, a computer program product is also proposed, comprising program code instructions for executing process 100.

[0119] Storage medium

[0120] According to another aspect, a non-transient storage medium is also proposed on which is stored the computer program comprising program code instructions to execute process 100.

Claims

1.

2. Demands Method (100) for three-dimensional inspection of a part from several radiographic images by an image analysis device comprising electronic circuitry adapted to implement the three-dimensional inspection method, the method being characterized in that it comprises at least the following steps: - (Step 1) Acquire a plurality of radiographic images of a part by a source-detector system, each image including information on the radiographic transmission properties of the part; - (Step 3) Align each image with a view of a digital model of the part; - (Step 4) Detect indications on each image and position the detected indications with bounding boxes; - (Step 5) Construct a multi-view geometry model using the alignment of each image on a respective view of the digital model of the part; - (Step 6) Consolidate the detected indications with bounding boxes, using the multi-view geometry model by constructing (step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (step 6.2) a candidate group to retain; - (Step 7) Locate each three-dimensional indication; - (Step 8) Characterize the dimensions of each indication in a relevant image selected from the set of images using bounding boxes to position the detected indications; - (Step 9) Use the characterization of each indication for further treatment by a third-party device. Method (100) according to claim 1 wherein the alignment step 3 comprises estimating (step 3.1) an optimal rigid body transformation of each view of the digital model of the part with respect to the source-detector system to obtain an image of the projection of each view of the digital model.

3. Method (100) according to claim 2, wherein a projection operator is used to carry out step 3.1, the operator is a matrix of the form • [ R |t ] ' with K a matrix described by intrinsic projection parameters, and with R and t respectively the rotation and translation of each view of the digital model.

4. Method (100) according to any one of claims 2 or 3, wherein step 3 comprises defining (step 3.2) a similarity metric between the calculated image of each view of the digital model and each acquired image.

5. Method (100) according to any one of claims 1 to 4, wherein step 6.1 of constructing a set of candidate groups comprises analyzing a location of indications.

6. A method (100) according to any one of claims 1 to 5, wherein the candidate group selection step 6.2 comprises selecting candidate groups containing projections of the same three-dimensional indication.

7. A method (100) according to any one of claims 1 to 6 wherein the localization (step 7) comprises determining an average of point clouds per candidate group containing several indications.

8. A method (100) according to any one of the preceding claims, wherein the characterization step comprises selecting (step 8.1) a relevant image to estimate a length and width of an indication, the relevant image being selected based on a three-dimensional location of the indication.

9. Image analysis device characterized in that it comprises electronic circuitry (200) for implementing a method (100) for three-dimensional inspection of a part from several radiographic images, the method comprising at least the following steps: - (Step 1) Acquire a plurality of radiographic images of a part using a source-detector system, each image containing information on the radiographic transmission properties of the part; - (Step 3) Align each image to a view of a digital model of the part; - (Step 4) Detect indications on each image and position the detected indications with bounding boxes; - (Step 5) Construct a multi-view geometry model using the alignment of each image to a respective view of the digital model of the part; - (Step 6) Consolidate the detected indications with bounding boxes, using the multi-view geometry model by constructing (Step 6.1) a set of candidate groups, each group containing projections of a potential three-dimensional indication, and selecting (Step 6.2) a candidate group to retain; - (Step 7) Locate each three-dimensional feature; - (Step 8) Characterize the dimensions of each feature in a relevant image selected from the set of images using bounding boxes to position the detected features; - (Step 9) Use the characterization of each feature for further processing by a third-party device.

10. Computer program product comprising program code instructions to execute the process (100) according to any one of claims 1 to 9, when said computer program product is executed by a processor.

11. Non-transient storage medium on which is stored a computer program comprising program code instructions to execute the method (100) according to any one of claims 1 to 9, when said instructions are read from said non-transient storage medium and executed by a processor.

Images