Image-guided multi-frame 4d millimeter wave radar point cloud 3d target detection method

Abstract

The application discloses a kind of 3D target detection methods of multi-frame 4D millimeter wave radar point cloud under image guidance.The application extracts optical flow between continuous image frames using deep learning neural network, combines the radial velocity observation of 4D millimeter wave radar point to recover the full speed of radar point, compensates the position of historical frame 4D millimeter wave radar point cloud point by point according to full speed, and uses the aggregated point cloud obtained by superposition after compensation to carry out 3D target detection.The application uses the radial velocity observation of 4D millimeter wave radar, recovers full speed with optical flow as matching condition, realizes accurate and robust full scene (including dynamic area) multi-frame 4D millimeter wave radar point cloud aggregation, compared with ordinary 4D millimeter wave radar point cloud aggregation mode, the point cloud obtained by aggregation of the application has better consistency and accuracy, and can better complete 3D target detection task.

Description

Technical Field

[0001] This invention relates to a 3D target detection method in the field of point cloud perception, and in particular to an image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method. Background Technology

[0002] 3D object detection is an important research direction in the field of point cloud perception. It involves taking point clouds as input and outputting the 3D position, orientation, and category of objects of interest. Compared to modalities such as images, point cloud inputs are sparse. To improve data density, multi-frame aggregation is a common approach widely used in 3D object detection to enhance performance. Common multi-frame aggregation methods require knowledge of pose changes between consecutive data frames. Rigid body transformations are used to transform historical frame point cloud data to the current frame, achieving multi-frame point cloud aggregation. However, this method can only align static parts of the scene; dynamic parts will exhibit trailing artifacts. A common solution is to add a dimension to the point cloud input to distinguish point cloud data from different frames (e.g., by directly using relative frame numbers) to reduce the impact of inconsistencies in dynamic regions.

[0003] 4D millimeter-wave radar is a newly emerging 3D sensor that can characterize the three-dimensional spatial coordinates and Doppler velocity information of targets in the environment. Compared to lidar, 4D millimeter-wave radar can observe the velocity information of targets in a scene. Furthermore, due to its longer operating wavelength, it can maintain reliable detection performance in harsh weather conditions such as fog, rain, and dust. Compared to 3D millimeter-wave radar, 4D millimeter-wave radar has greater angular and range resolution, and also has altitude measurement capabilities. Despite these unique advantages, 4D millimeter-wave radar also has significant drawbacks. Due to its relatively low spatial resolution, the point cloud of 4D millimeter-wave radar is very sparse, making it difficult to reflect the semantic information of the scene, posing a significant challenge to reliable 3D target detection. Summary of the Invention

[0004] This invention aims to address the problems existing in the background technology by providing an image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method, increasing the data density of the 4D millimeter-wave radar point cloud and improving the performance of 3D target detection. This invention proposes an image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method. First, optical flow between consecutive frame images is estimated using a deep learning neural network. Then, the 4D millimeter-wave radar point cloud is projected onto the image plane to obtain the corresponding optical flow. The full velocity of the 4D millimeter-wave radar point cloud is calculated based on the pose changes of the 4D millimeter-wave radar itself, the observed values ​​of the 4D millimeter-wave radar points, and the corresponding optical flow. Finally, the positions of the 4D millimeter-wave radar points in historical frames are compensated point-by-point based on the calculated full velocity, and the points are superimposed to obtain an aggregated point cloud. The aggregated 4D millimeter-wave radar point cloud is then used for 3D target detection.

[0005] This invention can effectively achieve multi-frame 4D millimeter-wave radar point cloud aggregation across all scenarios, especially applicable to dynamic regions. Its key feature lies in combining optical flow estimation and radial velocity observation from 4D millimeter-wave radar to achieve full velocity recovery, resulting in more accurate multi-frame alignment and thus improving the performance of 3D target detection.

[0006] The technical solution adopted in this invention is as follows:

[0007] I. An Image-Guided Multi-Frame 4D Millimeter-Wave Radar Point Cloud 3D Target Detection Method

[0008] 1) Preprocess two consecutive frames of images to obtain a preprocessed image pair;

[0009] 2) The optical flow of the preprocessed image pairs is estimated using the first deep learning neural network to obtain the dense optical flow relationship between two consecutive frames;

[0010] 3) Project the 4D millimeter-wave radar points onto the image plane using sensor extrinsic parameters and camera intrinsic parameters, and then combine the dense optical flow relationship between two consecutive frames to obtain the optical flow corresponding to each 4D millimeter-wave radar point.

[0011] 4) Using the 4D millimeter-wave radar inter-frame pose, the position of the 4D millimeter-wave radar point, the relative radial velocity, and the data frame time interval as input, the relative radial velocity of each 4D millimeter-wave radar point is compensated to the absolute radial velocity.

[0012] 5) Using the inter-frame pose of the 4D millimeter-wave radar, the position and corresponding optical flow of the 4D millimeter-wave radar point, the absolute radial velocity, and the data frame time interval as input, the full velocity of each 4D millimeter-wave radar point is obtained by solving the geometric relationship.

[0013] 6) The position of the 4D millimeter-wave radar point in the historical frame is compensated point by point according to the full velocity of the 4D millimeter-wave radar point, and then the positions of the 4D millimeter-wave radar point cloud in multiple frames are superimposed to obtain the aggregated 4D millimeter-wave radar point cloud.

[0014] 7) Using aggregated 4D millimeter-wave radar point clouds as input, 3D target detection is performed using a second deep learning neural network to obtain target detection results.

[0015] Specifically, 1) refers to:

[0016] After normalizing and standardizing two consecutive frames of images, a preprocessed image pair is obtained.

[0017] In step 2), the first deep learning neural network is UniMatch or RAFT.

[0018] In step 4), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is taken as the current frame, and the absolute radial velocity of each 4D millimeter-wave radar point in the current frame is obtained through compensation. The specific steps are as follows:

[0019] 4.1) The self-movement velocity of the radar is obtained based on the inter-frame pose and data frame time interval of the 4D millimeter-wave radar;

[0020] 4.2) The self-motion velocity of the radar in the current frame is superimposed onto each 4D millimeter-wave radar point in the current frame, and the self-motion velocity of the radar is decomposed into mutually orthogonal radial and tangential components. The radial component is used to compensate for the relative radial velocity of the observed 4D millimeter-wave radar point to obtain the absolute radial velocity of each 4D millimeter-wave radar point in the current frame.

[0021] In step 5), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is taken as the current frame, and the full velocity of each radar point in the next moment is calculated. The specific steps are as follows:

[0022] 5.1) Determine the spatial constraint plane where each 4D millimeter-wave radar point is located at the next moment based on the position, absolute radial velocity, and data frame time interval of each 4D millimeter-wave radar point in the current frame;

[0023] 5.2) Determine the spatial constraint line where the current 4D millimeter-wave radar point is located at the next moment based on the optical flow corresponding to the current 4D millimeter-wave radar point and the inter-frame pose of the 4D millimeter-wave radar.

[0024] 5.3) The intersection of the spatial constraint plane and the spatial constraint line is taken as the position of the current 4D millimeter-wave radar point in the next moment. The velocity vector of the current 4D millimeter-wave radar point is obtained by combining the data frame time interval and taken as the total velocity.

[0025] 5.4) Repeat steps 5.1)-5.3) to iterate and calculate the position and full velocity of the remaining 4D millimeter-wave radar points in the current frame at the next moment.

[0026] In step 7), the second deep learning neural network is PointPillar, SECOND, Voxel-RCNN, or Point-RCNN.

[0027] In step 6), for each 4D millimeter-wave radar point at time t-1 Its three-dimensional position vector is After a time interval Δt, it moves to... satisfy in, The full velocity of the 4D millimeter-wave radar point is determined based on the inter-frame pose of the radar point. Transform to the radar coordinate system of the latest frame.

[0028] II. A computer device

[0029] The device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method.

[0030] III. A computer-readable storage medium

[0031] The medium stores a computer program that, when executed by a processor, implements the steps of the method.

[0032] IV. A computer program product

[0033] The product includes a computer program / instructions that, when executed by a processor, implement the steps of the method.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] Compared with existing multi-frame point cloud aggregation methods, this invention combines optical flow estimation and radial velocity observation of 4D millimeter-wave radar to recover the full velocity of radar points. By aggregating multi-frame point clouds through point-by-point compensation, it can simultaneously process dynamic and static targets in the scene. The resulting aggregated point cloud has better accuracy and consistency, thus achieving better 3D target detection performance. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method of the present invention.

[0037] Figure 2 This is a schematic diagram of the spatial constraint line and the spatial constraint plane from a top-down perspective in the embodiment.

[0038] Figure 3 The following are the visual detection results of different methods in the embodiments, where (a) is the detection result of rigid body transformation, (b) is the detection result of radial compensation, and (c) is the detection result of the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] like Figure 1 As shown, the present invention includes the following steps:

[0041] 1) After preprocessing the two consecutive frames of images to be matched, the preprocessed image pairs are obtained;

[0042] 1) Specifically:

[0043] First, the two three-channel RGB images to be matched are normalized by dividing all pixel values ​​by 255. Then, the images are standardized channel by channel, with the mean values ​​for the three channels being [0.485, 0.456, 0.406] and the standard deviations being [0.229, 0.224, 0.225]. The preprocessed image pair is obtained after these steps.

[0044] 2) Optical flow estimation is performed on the preprocessed image pairs using a deep learning neural network to obtain the dense optical flow relationship between two consecutive frames;

[0045] 2) Specifically:

[0046] The dense optical flow F between the two images is estimated using a first deep learning neural network, which is UniMatch.

[0047] 3) Project the 4D millimeter-wave radar points onto the image plane using sensor extrinsic parameters and camera intrinsic parameters, and then combine the dense optical flow relationship between two consecutive frames to obtain the optical flow corresponding to each 4D millimeter-wave radar point.

[0048] 3) Specifically:

[0049] The 4D millimeter-wave radar point cloud is represented as {r 1 ,r 2 ,…,r n The three-dimensional coordinates of the radar point are represented as r. n =[x n ,y n ,z n ] T The radar-camera extrinsic parameter matrix is ​​represented as M. 4×4 The camera intrinsic parameter matrix is ​​represented as K 4×4 Dense optical flow is denoted as F. Radar points are expanded to homogeneous coordinates. The projection point p of the radar point on the pixel plane can be obtained through matrix transformation. n , K and M are the camera intrinsic parameter matrix and the radar-camera extrinsic parameter matrix, respectively. Since the projection point coordinates are floating-point numbers, the corresponding optical flow f can be obtained through bilinear interpolation. n =Interpolate(F,p n Interpolate() represents the bilinear interpolation operation.

[0050] 4) Using the inter-frame pose of the 4D millimeter-wave radar (on the vehicle), the position (i.e., three-dimensional coordinates) of the 4D millimeter-wave radar point, the relative radial velocity, and the data frame time interval as input, the relative radial velocity of each radar point is compensated to the absolute radial velocity.

[0051] In step 4), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is used as the current frame, and the absolute radial velocity of each radar point in the current frame is obtained through compensation. The specific steps are as follows:

[0052] 4.1) Obtain the self-movement velocity of the radar (on the vehicle) based on the inter-frame pose of the 4D millimeter-wave radar (on the vehicle) and the data frame time interval;

[0053] Specifically:

[0054] Use E 4×4 R represents the radar coordinate system at time t-1. t-1 radar coordinate system R at time t t The inter-frame pose, during which the self-motion displacement vector of the radar (on its platform) can be measured in R. t-1 The following is represented as [0,0,0,1] T From this, the self-movement speed of the radar (on its vehicle) can be obtained. Where Δt is the time interval.

[0055] 4.2) The self-motion velocity of the radar in the current frame is superimposed onto each radar point in the current frame, and the self-motion velocity of the radar is decomposed into mutually orthogonal radial and tangential components. The radial component is used to compensate for the relative radial velocity of the observed radar point to obtain the absolute radial velocity of each radar point in the current frame.

[0056] Specifically:

[0057] use Represents the three-dimensional position vector of the radar point. Represents the relative radial velocity observed by the radar; the radial direction vector of the radar point. It can be represented as Among them, R i The Euclidean distance from the radar point to the origin of the coordinate system. These are the components of the radar point position on the x, y, and z axes, representing the self-motion velocity. Decomposing to the radial direction yields the components. Then, by superimposing the radial component of the self-moving velocity onto the relative radial velocity observed by the radar, it can be compensated to obtain the absolute radial velocity. Represented as

[0058] 5) Using the inter-frame pose of the 4D millimeter-wave radar (on the vehicle), the position (i.e., three-dimensional coordinates) of the 4D millimeter-wave radar point, the corresponding optical flow, absolute radial velocity, and data frame time interval as input, the full velocity of each 4D millimeter-wave radar point is obtained by solving through geometric relationships.

[0059] In step 5), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is taken as the current frame, and the full velocity of each 4D millimeter-wave radar point in the next moment is calculated. The specific steps are as follows:

[0060] 5.1) Determine the spatial constraint plane where each 4D millimeter-wave radar point is located at the next moment based on the position (i.e., three-dimensional coordinates), absolute radial velocity, and data frame time interval of each 4D millimeter-wave radar point in the current frame;

[0061] Specifically:

[0062] For each 4D millimeter-wave radar point at time t-1 Its three-dimensional position vector is known. Absolute radial velocity Therefore, the radial component of the radar point displacement from time t-1 to t is determined and can be expressed as: Where Δt represents the time interval. Displacement The radial component. From the definition of radial, we can obtain... and The relationship is From the definition of displacement, we can obtain that in Represents the radar point at time t The three-dimensional position vector. In summary, the relation can be obtained. This relation is a point-normal expression in a spatial plane, which describes a constraint: the radar point at time t It must lie on this plane. The plane can be represented as:

[0063]

[0064] 5.2) Determine the spatial constraint line where the current 4D millimeter-wave radar point is located at the next moment based on the optical flow corresponding to the current 4D millimeter-wave radar point and the inter-frame pose of the 4D millimeter-wave radar (the vehicle in which it is located);

[0065] Specifically:

[0066] For each 4D millimeter-wave radar point at time t-1 The corresponding image projection points can be obtained from the method described in 3). and optical flow f i From the definition of optical flow, we can obtain that That is, the radar point at time t. The image projection points. This can be derived from the imaging principle of a camera. The image projection point at time t On the straight line defined by the optical center of the camera. To be consistent with 5.1), at time t-1, the radar coordinate system R...t-1 The spatially constrained straight line is represented in the diagram. The radar-camera extrinsic parameter matrix M is known. 4×4 Radar coordinate system R at time t-1 t-1 radar coordinate system R at time t t Inter-frame pose E 4×4 and camera intrinsic parameter matrix K 4×4 Points on the spatially constrained line can be represented as λ∈R, where Representing the projection points respectively The horizontal and vertical coordinates on the pixel plane. Two points l are arbitrarily selected on the spatially constrained line. i (λ=λ0) and l i (λ=λ1) and denoted as and The spatially constrained line can be represented in parametric form as follows:

[0067]

[0068] 5.3) The intersection of the spatial constraint plane and the spatial constraint line is taken as the position (three-dimensional coordinates) of the current 4D millimeter-wave radar point at the next moment. The velocity vector of the current 4D millimeter-wave radar point is obtained by combining the data frame time interval and used as the total velocity.

[0069] Specifically:

[0070] As described in 5.1), the radar point at time t In a spatially constrained plane Plane i Above; as described in 5.2), radar point at time t In a spatially constrained straight line Line i Above. Therefore, the spatial constraint line is Line. i With spatial constraint plane i The intersection point is the radar point at time t on R. t-1 3D position vector in coordinate system The direction vector of a straight line in space Recorded as The following formula can be used to calculate

[0071]

[0072] The displacement of the radar point can be expressed as The full velocity of the radar point is

[0073] 5.4) Repeat steps 5.1)-5.3) to iterate and calculate the position and full velocity of the remaining 4D millimeter-wave radar points in the current frame at the next moment.

[0074] Figure 2 This is a simplified representation of the above spatial geometric relationships (top view). The rounded rectangle represents the vehicle where the radar is located, the dashed line inside the rectangle represents the imaging plane of the camera, C represents the camera coordinate system, R represents the 4D millimeter-wave radar coordinate system, and the rest of the labels in the figure are consistent with those in the text above.

[0075] 6) Based on the full velocity of the 4D millimeter-wave radar points obtained by the solution, the position of the 4D millimeter-wave radar points in the historical frames is compensated point by point, and then the positions of the 4D millimeter-wave radar point clouds of multiple frames are superimposed to obtain the aggregated 4D millimeter-wave radar point cloud.

[0076] Specifically:

[0077] For each 4D millimeter-wave radar point at time t-1 Its three-dimensional position vector is It will move to a new position after a time interval Δt. The full velocity of the 4D millimeter-wave radar point is calculated based on the pose changes between data frames, and the compensated radar point is then... Transform to the radar coordinate system of the latest frame.

[0078] By performing the above operations on historical 4D millimeter-wave radar points, multi-frame aggregation of 4D millimeter-wave radar point clouds is achieved.

[0079] 7) Using aggregated 4D millimeter-wave radar point clouds as input, 3D target detection is performed using a second deep learning neural network to obtain the target detection results. The second deep learning neural network used is PointPillar, SECOND, Voxel-RCNN, or Point-RCNN.

[0080] This invention utilizes the Unimatch deep learning neural network to extract optical flow between consecutive image frames, and combines this with radial velocity observations from 4D millimeter-wave radar to recover the full velocity of radar points. Based on this full velocity, the positions of 4D millimeter-wave radar points in historical frames are compensated point-by-point. Through this method, this invention can achieve accurate and consistent multi-frame point cloud aggregation across the entire scene (including dynamic regions), improving the performance of 3D target detection.

[0081] To evaluate the results of this invention, experiments were conducted using the publicly available autonomous driving dataset VOD, and the quantitative results are shown in Table 1. The evaluation metric was the average precision of 3D object detection, with a cross-union threshold of 0.7 for vehicles and 0.5 for pedestrians and cyclists. In the experiments, this invention was compared with two other multi-frame aggregation methods: one directly utilizes inter-frame rigid body transformation to achieve multi-frame point cloud aggregation; the other utilizes only 4D millimeter-wave radar radial velocity observations to compensate radar points radially for multi-frame aggregation. In the comparative experiments, all three methods aggregated data from 5 consecutive frames. As can be seen from Table 1, regardless of the 3D object detector used, the 3D object detection performance of this invention is significantly stronger than the other two. Furthermore, Figure 3 The document displays visualizations of some detection samples. Figure 3 of (a), Figure 3 (b) and Figure 3 Figure (c) shows the detection results of rigid body transformation, radial compensation, and the present invention, respectively, representing a qualitative analysis. The figure includes visualizations of the point cloud and the ground truth bounding box. It is evident that the point cloud aggregated by the present invention exhibits the best accuracy and consistency, with the predicted bounding box most closely resembling the ground truth bounding box. Therefore, it can be concluded that the present invention improves the performance of 3D target detection through image-guided multi-frame aggregation of 4D millimeter-wave radar point clouds.

[0082] Table 1 shows the performance comparison of 3D target detection.

[0083]

[0084]

[0085] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. An image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method, characterized in that, Includes the following steps: 1) Preprocess two consecutive frames of images to obtain a preprocessed image pair; 2) Optical flow estimation is performed on the preprocessed image pairs using the first deep learning neural network to obtain the dense optical flow relationship between two consecutive frames; 3) Project the 4D millimeter-wave radar points onto the image plane using sensor extrinsic parameters and camera intrinsic parameters, and then combine the dense optical flow relationship between two consecutive frames to obtain the optical flow corresponding to each 4D millimeter-wave radar point. 4) Using the 4D millimeter-wave radar inter-frame pose, the position of the 4D millimeter-wave radar point, the relative radial velocity, and the data frame time interval as input, the relative radial velocity of each 4D millimeter-wave radar point is compensated to the absolute radial velocity. 5) Using the 4D millimeter-wave radar inter-frame pose, the position and corresponding optical flow of the 4D millimeter-wave radar point, the absolute radial velocity, and the data frame time interval as input, the full velocity of each 4D millimeter-wave radar point is obtained by solving the geometric relationship. In step 5), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is taken as the current frame, and the full velocity of each radar point in the next moment is calculated. The specific steps are as follows: 5.1) Determine the spatial constraint plane where each 4D millimeter-wave radar point is located at the next moment based on the position, absolute radial velocity, and data frame time interval of each 4D millimeter-wave radar point in the current frame; 5.2) Determine the spatial constraint line where the current 4D millimeter-wave radar point is located at the next moment based on the optical flow corresponding to the current 4D millimeter-wave radar point and the inter-frame pose of the 4D millimeter-wave radar. 5.3) The intersection of the spatial constraint plane and the spatial constraint line is taken as the position of the current 4D millimeter-wave radar point in the next moment. The velocity vector of the current 4D millimeter-wave radar point is obtained by combining the data frame time interval and taken as the total velocity. 5.4) Repeat steps 5.1)-5.3) to iterate and calculate the position and full velocity of the remaining 4D millimeter-wave radar points in the current frame at the next moment; 6) The position of the 4D millimeter-wave radar point in the historical frame is compensated point by point according to the full velocity of the 4D millimeter-wave radar point, and then the positions of the 4D millimeter-wave radar point cloud in multiple frames are superimposed to obtain the aggregated 4D millimeter-wave radar point cloud. 7) Using aggregated 4D millimeter-wave radar point clouds as input, 3D target detection is performed using a second deep learning neural network to obtain target detection results.

2. The image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method according to claim 1, characterized in that, Specifically, 1) refers to: After normalizing and standardizing two consecutive frames of images, a preprocessed image pair is obtained.

3. The image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method according to claim 1, characterized in that, In step 2), the first deep learning neural network is UniMatch or RAFT.

4. The image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method according to claim 1, characterized in that, In step 4), for two consecutive frames of 4D millimeter-wave radar points, the previous frame is taken as the current frame, and the absolute radial velocity of each 4D millimeter-wave radar point in the current frame is obtained through compensation. The specific steps are as follows: 4.1) The self-movement velocity of the radar is obtained based on the inter-frame pose and data frame time interval of the 4D millimeter-wave radar; 4.2) The self-motion velocity of the radar in the current frame is superimposed onto each 4D millimeter-wave radar point in the current frame, and the self-motion velocity of the radar is decomposed into mutually orthogonal radial and tangential components. The radial component is used to compensate for the relative radial velocity of the observed 4D millimeter-wave radar point to obtain the absolute radial velocity of each 4D millimeter-wave radar point in the current frame.

5. The image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method according to claim 1, characterized in that, In step 7), the second deep learning neural network is PointPillar, SECOND, Voxel-RCNN, or Point-RCNN.

6. The image-guided multi-frame 4D millimeter-wave radar point cloud 3D target detection method according to claim 1, characterized in that, In the above 6), for Every 4D millimeter-wave radar point at any given moment Its three-dimensional position vector is After a time interval Later movement to ,satisfy ,in, The full velocity of the 4D millimeter-wave radar point is determined based on the inter-frame pose of the radar point. Transform to the radar coordinate system of the latest frame.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 6.

Images