Method and device for generating sole cementing trajectories based on dynamic time warping

Abstract

The present invention discloses a method and device for generating a sole coating trajectory based on dynamic time warping. The method includes: taking the original reference point cloud, the original point cloud to be registered, and the teaching trajectory t0 as input parameters; preprocessing the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud p f0 to be registered; performing point cloud ICP rigid body transformation g r0 on the reference point cloud p r to obtain the rigidly registered point cloud p r1 . At the same time, the teaching trajectory t0 is subjected to a synchronous rigid body transformation once to obtain the rigidly registered trajectory t1; obtaining the reference image i r0 and the image i f0 to be registered, and the rigidly registered trajectory t1 is synchronously transformed into the teaching trajectory image i t0 ; performing dynamic time warping processing on the reference image i r0 to obtain the softly registered image i r1 , and the teaching trajectory image i t0 is synchronously transformed to obtain the softly registered trajectory image i t1 ; restoring the depth direction value during teaching, so as to finally obtain the reproduced trajectory t2, and the coating robot performs coating according to the reproduced trajectory t2. The advantages of the present invention are: it helps to improve the operation efficiency and accuracy of the coating operation robot and can meet the requirements of the industrial site.

Description

Technical Field

[0001] The present invention relates to the field of robot glue application in the shoe manufacturing process, and more particularly to a method and device for generating a sole glue application trajectory based on dynamic time warping. Background Art

[0002] At present, the non-standard nature of sole and upper products in the shoe manufacturing industry is a common phenomenon. Even shoes of the same model and the same size may have inconsistent sizes and shapes, which poses a great challenge to the automation of shoe production operations. Currently, the method of manual glue application is commonly used in factories, where the human eye is used to judge and compensate for the key features of the shoes and perform the glue application operation. In order to achieve industrial automation and retain the experienced glue application techniques of workers, the best way is "teaching + reproduction": by manually teaching a shoe and inputting the optimal glue application trajectory and technique to the robot. When the next shoe (which may be the same or different size) reaches the robot work station, the installation error and flexible deviation between the two shoes are calculated through "machine vision", and the compensated trajectory is output to the robot, thereby obtaining the optimal trajectory and technique for the new shoe and implementing it through the robot. In this process, how to design the "machine vision" algorithm, that is, the eyes of the robot, to realize the recognition and compensation of the errors between the teaching and reproduction of the two shoes is particularly important.

[0003] Currently, there is no relevant point cloud or image registration technology in the shoe manufacturing industry. However, in other fields, such as medical images, satellite images, geological image registration, etc., it has good applications. Briefly speaking, the goal of non-rigid registration of point clouds (images) is to find a spatial geometric transformation so that the corresponding points in the two point clouds (images) are aligned with each other for direct comparison or fusion. In order to evaluate the similarity between images and increase the accuracy of the final registration, an evaluation index of similarity measure is required, which usually includes the mean square error sum considering image gray levels, correlation coefficient, correlation rate, and mutual information, etc.; in addition, in order to determine the transformation method between images, a spatial transformation model is needed for constraint. The spatial transformation model for non-rigid registration is usually divided into two categories: the model based on basis function expansion and the physical model. The theoretical basis of the method based on function expansion is the theory of function interpolation and approximation, while the basis of the method based on physical model is the continuous medium mechanics; also, in order to adjust the spatial deformation parameters and quickly find the optimal value of the similarity measure, an optimization method is needed, such as the commonly used steepest gradient descent method, quasi-Newton method, evolutionary algorithm, etc.

[0004] For example, the image registration technology based on the Demons algorithm proposed by Thirion regards all the pixel points of the reference image or the pixel points on the contour as Demons points, and regards the image to be registered as a deformable grid. The Demons force on each integer point grid changes the contour of the reference image along the direction of the gray gradient in the image to be registered, driving the reference image to deform towards the image to be registered, so as to achieve the matching between the two images. This algorithm is a registration method based on the optical flow theory, and its premise is that the gray level remains unchanged during the movement of the image. Finally, in order to achieve the convergence of the displacement field, the displacement field must be continuously optimized until convergence. However, this method only uses the gradient information of the image to be registered to drive, and may change the topological structure of the image when the information is insufficient, resulting in misregistration; in addition, when the deformation between the two images is large, this method usually cannot complete the registration or converges very slowly. And this is also the main problem in the registration of sole images. Due to the non-standard nature of shoes, the error between the taught and reproduced shoes is usually large, resulting in too low computational efficiency of the algorithm and not meeting the requirements of the industrial site. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing flexible registration method for shoe images and the method for generating glue application trajectories are prone to misregistration, and the computational efficiency is too low to meet the requirements of the industrial site.

[0006] The present invention solves the above technical problems through the following technical means: a method for generating a sole glue application trajectory based on dynamic time warping, the method comprising:

[0007] Step 1: Take the taught sole point cloud as the original reference point cloud, the sole point cloud to be matched as the original point cloud to be registered, and the original reference point cloud, the original point cloud to be registered, and the taught trajectory t0 as input parameters;

[0008] Step 2: Preprocess the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud to be registered p f0 ;

[0009] Step 3: Based on the point cloud to be registered p f0 Perform point cloud ICP rigid body transformation g r0 on the reference point cloud p r to obtain the rigidly registered point cloud p r1 , and at the same time, perform a synchronous rigid body transformation on the taught trajectory t0 to obtain the rigidly registered trajectory t1;

[0010] Step 4: Convert the sole spatial point cloud into a depth map to obtain the reference image i r0 and the image to be registered i f0 , and the rigidly registered trajectory t1 is synchronously converted into the taught trajectory image it0 ;

[0011] Step Five: Based on the image i to be registered f0 perform dynamic time warping on the reference image i r0 to obtain the soft-matched image i r1 , and perform synchronous transformation on the taught trajectory image i t0 to obtain the soft-matched trajectory image i t1 ;

[0012] Step Six: Find the correspondence between the data of the soft-matched trajectory image after flexible registration and the corresponding points of the taught trajectory t0, and restore the depth direction value during teaching, so as to finally obtain the reproduced trajectory t2, and the gluing robot performs gluing according to the reproduced trajectory t2.

[0013] The present invention uses the point cloud information and depth image information of the sole, with complete information and not easily causing mis-matching. And based on the conversion from point cloud to depth map and dynamic time warping processing, the complex process of image flexible registration is simplified to simple stretching and compression operations of row and column values, realizing the migration of the taught shoe type manipulation trajectory to the reproduced shoe type, completing trajectory reproduction, which helps to improve the operation efficiency and accuracy of the gluing operation robot and can meet the requirements of the industrial site.

[0014] Further, the second step includes:

[0015] Separate the taught sole and the sole to be registered from the original data through a series of preprocessing operations such as filtering, segmentation, and sampling, so as to obtain the reference point cloud p r0 and the point cloud p to be registered f0 .

[0016] Further, the third step includes:

[0017] Based on the point cloud p to be registered f0 perform the point cloud ICP rigid body transformation g on the reference point cloud p r0 , adjust the position and attitude of the reference point cloud p r to the state closest to that of the point cloud p to be registered r0 , and obtain the rigidly matched point cloud p f0 r1 . At the same time, perform a synchronous rigid body transformation g r1 on the taught trajectory t0 r , to obtain the rigidly matched trajectory t1, thereby compensating for the installation deviation between the two soles and completing the rough registration of their positions and attitudes.

[0018] Further, the fourth step includes:

[0019] For the rigidly matched point cloud p r1 and the point cloud p to be registered f0Perform a point cloud to depth map conversion respectively to convert the sole spatial point cloud into a depth map, and obtain the reference image i r0 and the image i to be registered f0 , while the just-matched trajectory t1 performs a point cloud to depth map conversion synchronously to be converted into the teaching trajectory image i t0 , and complete the dynamic time warping flexible transformation at the image level.

[0020] Furthermore, the fifth step includes:

[0021] Based on the image i to be registered f0 , perform dtw processing on the rows and columns of the reference image i r0 respectively, so as to obtain the flexible matching image i f0 that is most similar to the image i to be registered r1 . Correspondingly, perform dtw processing on the rows and columns of the teaching trajectory image i t0 synchronously, and obtain the flexible matching trajectory image i t1 after flexible matching.

[0022] The present invention also provides a sole glue application trajectory generation device based on dynamic time warping, and the device includes:

[0023] An input module for using the taught sole point cloud as the original reference point cloud, the sole point cloud to be matched as the original point cloud to be registered, and the original reference point cloud, the original point cloud to be registered, and the taught trajectory t0 as input parameters;

[0024] A preprocessing module for preprocessing the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud p to be registered f0 ;

[0025] A rigid body transformation module for performing point cloud ICP rigid body transformation g f0 on the reference point cloud p r0 based on the point cloud p to be registered r , and obtaining the rigidly matched point cloud p r1 . At the same time, the taught trajectory t0 performs a synchronous rigid body transformation to obtain the rigidly matched trajectory t1;

[0026] A depth map conversion module for converting the sole spatial point cloud into a depth map to obtain the reference image i r0 and the image i to be registered f0 , and the rigidly matched trajectory t1 is synchronously converted into the teaching trajectory image i t0 ;

[0027] A flexible matching module for performing dynamic time warping processing on the reference image i f0 based on the image i to be registered r0 to obtain the flexible matching image i r1 , the teaching trajectory image it0 Perform synchronous transformation to obtain the flexible registration trajectory image i t1 ;

[0028] The glue application generation module is used to find the corresponding relationship between the flexible registration trajectory image data after flexible registration and the corresponding points of the teaching trajectory t0, and restore the depth direction value during teaching, so as to finally obtain the reproduction trajectory t2, and the glue application robot performs glue application according to the reproduction trajectory t2.

[0029] Furthermore, the preprocessing module is also used for:

[0030] Separate the teaching sole and the sole to be registered from the original data through a series of preprocessing operations such as filtering, segmentation, and sampling, so as to obtain the reference point cloud p r0 and the point cloud p to be registered f0 .

[0031] Furthermore, the rigid body transformation module is also used for:

[0032] Based on the point cloud p to be registered f0 Perform point cloud ICP rigid body transformation g on the reference point cloud p r0 , adjust the position and attitude of the reference point cloud p r to the state closest to the point cloud p to be registered r0 , and obtain the rigidly registered point cloud p f0 . At the same time, the teaching trajectory t0 is subjected to a synchronous rigid body transformation g r1 , and the rigidly registered trajectory t1 is obtained, so as to compensate for the installation deviation between the two soles and complete the rough registration of their positions and attitudes. r

[0033] Furthermore, the depth map conversion module is also used for:

[0034] Perform a point cloud to depth map conversion on the rigidly registered point cloud p r1 and the point cloud p to be registered f0 respectively, convert the sole space point cloud into a depth map, and obtain the reference image i r0 and the image i to be registered f0 . The rigidly registered trajectory t1 is synchronously subjected to a point cloud to depth map conversion and converted into a teaching trajectory image i t0 , completing the dynamic time warping flexible transformation at the image level.

[0035] Furthermore, the flexible registration module is also used for:

[0036] Based on the image i to be registered f0 , perform dtw processing on the rows and columns of the reference image i r0 respectively, so as to obtain the flexible registration image i f0 most similar to the image i to be registered r1, correspondingly, synchronously perform DTW processing on the rows and columns of the teaching trajectory image i t0 to obtain the flexible matching trajectory image i after flexible matching t1 .

[0037] The advantages of the present invention are as follows: The present invention adopts the point cloud information and depth image information of the sole, with complete information and not easily causing false matching. And based on the conversion from point cloud to depth map and dynamic time warping processing, the complex process of image flexible registration is simplified to simple stretching and compression operations of row and column values, realizing the migration of the teaching shoe type manipulation trajectory to the reproduced shoe type and completing trajectory reproduction, which helps to improve the operation efficiency and accuracy of the glue application robot and can meet the requirements of the industrial site. Brief Description of the Drawings

[0038] Figure 1 is a flowchart of the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention;

[0039] Figure 2 is the original reference shoe and point cloud, the original shoe to be registered and point cloud in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention, where a is the original reference shoe, b is the original registered shoe, c is the point cloud of the original reference shoe, and d is the point cloud of the original registered shoe;

[0040] Figure 3 is a schematic diagram of the preprocessing result in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention, where a is the preprocessing result of the reference point cloud and b is the preprocessing result of the point cloud to be registered;

[0041] Figure 4 is a schematic diagram of the rigid matching trajectory in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention;

[0042] Figure 5 is a schematic diagram after depth conversion in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention, where a is the reference image and b is the image to be registered;

[0043] Figure 6 is a schematic diagram of the registration result and comparison in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention, where a is the registration result i r1 and the registration error display i r1 -i f0 , and b is the trajectory image after flexible matching;

[0044] Figure 7 is a spatial comparison of the reproduced trajectory t2 and the teaching trajectory t0 in the sole glue application trajectory generation method based on dynamic time warping disclosed in the embodiment of the present invention. Detailed Implementation Manner

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] Embodiment 1

[0047] As Figure 1 shown, a method for generating a sole rubber coating trajectory based on dynamic time warping, the method comprising:

[0048] S1: As Figure 2 shown, in order to achieve the flexible registration transformation from the teaching trajectory of the teaching sole to the reproduction sole, the teaching sole point cloud is used as the original reference point cloud, the sole point cloud to be matched is used as the original point cloud to be registered, and the original reference point cloud p′ r0 , the original point cloud p′ f0 to be registered, and the teaching trajectory t0 are used as input parameters;

[0049] S2: Due to the factors of the 3D camera itself, the extracted sole point cloud will be affected by environmental interference, that is, there are many environmental background points and interference points. Therefore, it is first necessary to separate the target (sole) from the original data through a series of preprocessing operations such as filtering, segmentation, and sampling, so as to obtain the reference point cloud p r0 and the point cloud p f0 to be registered; the preprocessing result is as Figure 3 shown. Filtering, segmentation, and sampling belong to conventional techniques in image processing and will not be elaborated here.

[0050] S3: In order to compensate for the installation deviation between the two soles, it is necessary to perform a rough registration of their positions and postures. Therefore, based on the point cloud p f0 to be registered, a point cloud ICP rigid body transformation g r0 is performed on the reference point cloud p r , and the position and posture of the reference point cloud p r0 are adjusted to the state closest to the point cloud p f0 to be registered, obtaining the rigidly registered point cloud p r1 . At the same time, the teaching trajectory t0 is subjected to a synchronous rigid body transformation g r , obtaining the rigidly registered trajectory t1, thereby compensating for the installation deviation between the two soles and completing the rough registration of their positions and postures; the rigidly registered trajectory is as Figure 4 shown. Rigid body transformation belongs to the prior art, and the rigid body transformation method of the prior art is used to compensate for the installation deviation between the two soles.

[0051] S4: Perform a point cloud to depth map conversion on the just-matched point cloud p r1 and the point cloud p to be registered f0 respectively, to convert the sole space point cloud into a depth map, obtaining the reference image i r0 and the image i to be registered f0 . At the same time, perform a point cloud to depth map conversion on the just-matched trajectory t1 to convert it into a teaching trajectory image i t0 , and complete the dynamic time warping flexible transformation at the image level; the images after the depth map conversion are as Figure 5 shown. Converting the space point cloud into a depth map belongs to the prior art.

[0052] S5: With the results of S4, the reference image i r0 and the image i to be registered f0 can be subjected to dynamic time warping processing, that is, based on the image i to be registered f0 , perform dtw processing on the rows and columns of the reference image i r0 respectively, so as to obtain the most similar flexible matching image i f0 to the image i to be registered. Correspondingly, perform dtw processing on the rows and columns of the teaching trajectory image i r1 simultaneously to obtain the flexible matching trajectory image i t0 after flexible matching; the registration results and comparisons are as t1 shown. The dtw processing belongs to the prior art. Figure 6 shown. The dtw processing belongs to the prior art.

[0053] S6: Finally, in order to restore the depth direction data of the trajectory, it is necessary to find the correspondence between the data of the flexible matching trajectory image after flexible registration and the corresponding points of the teaching trajectory t0, and restore the depth direction values during teaching, so as to finally obtain the reproduced trajectory t2. Thus, the reproduced trajectory for the new shoes and retaining the original teaching trajectory technique can be obtained and provided to the robot as an output. The gluing robot performs gluing according to the reproduced trajectory t2. The spatial comparison between the reproduced trajectory t2 and the teaching trajectory t0 is as Figure 7 shown.

[0054] It should be noted that the main improvement of the present invention lies in the entire method process. Some image processing technologies involved in the method adopt the prior art, such as dtw processing, preprocessing, point cloud to depth map conversion, rigid body transformation, etc. all adopt the prior art.

[0055] Through the above technical solutions, the present invention adopts the point cloud information and depth image information of the sole, with complete information and not easily causing mis-matching. And based on the conversion from point cloud to depth map and dynamic time warping processing, the complex process of image flexible registration is simplified to simple stretching and compression operations of row and column values, realizing the migration of the taught shoe type technique trajectory to the reproduced shoe type, completing the trajectory reproduction, which helps to improve the operation efficiency and accuracy of the glue application robot and can meet the requirements of the industrial site.

[0056] Embodiment 2

[0057] Based on Embodiment 1, Embodiment 2 of the present invention further provides a sole glue application trajectory generation device based on dynamic time warping, and the device includes:

[0058] An input module, configured to use the taught sole point cloud as the original reference point cloud, the sole point cloud to be matched as the original point cloud to be registered, and the original reference point cloud, the original point cloud to be registered, and the taught trajectory t0 as input parameters;

[0059] A preprocessing module, configured to preprocess the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud to be registered p f0 ;

[0060] A rigid body transformation module, configured to perform point cloud ICP rigid body transformation g f0 on the reference point cloud p r0 based on the point cloud to be registered p r to obtain the rigidly registered point cloud p r1 , and at the same time, perform a synchronous rigid body transformation on the taught trajectory t0 to obtain the rigidly registered trajectory t1;

[0061] A depth map conversion module, configured to convert the sole spatial point cloud into a depth map to obtain the reference image i r0 and the image to be registered i f0 , and synchronously convert the rigidly registered trajectory t1 into the taught trajectory image i t0 ;

[0062] A flexible registration module, configured to perform dynamic time warping processing on the reference image i f0 based on the image to be registered i r0 to obtain the flexibly registered image i r1 , and perform synchronous conversion on the taught trajectory image i t0 to obtain the flexibly registered trajectory image i t1 ;

[0063] A glue application generation module, configured to find the corresponding relationship between the flexibly registered flexibly registered trajectory image data and the corresponding points of the taught trajectory t0, and restore the depth direction value during teaching, so as to finally obtain the reproduced trajectory t2, and the glue application robot performs glue application according to the reproduced trajectory t2.

[0064] Specifically, the preprocessing module is further configured to:

[0065] Separate the teaching sole and the sole to be registered from the original data through a series of preprocessing operations such as filtering, segmentation, and sampling, so as to obtain the reference point cloud p r0 and the point cloud p to be registered f0 .

[0066] Specifically, the rigid body transformation module is further configured to:

[0067] Based on the point cloud p to be registered f0 perform point cloud ICP rigid body transformation g on the reference point cloud p r0 , adjust the position and attitude of the reference point cloud p r to the state closest to that of the point cloud p to be registered r0 , obtain the rigidly registered point cloud p f0 , and at the same time, perform a synchronous rigid body transformation g on the teaching trajectory t0 r1 , obtain the rigidly registered trajectory t1, so as to compensate for the installation deviation between the two soles and complete the rough registration of their positions and attitudes. r

[0068] Specifically, the depth map conversion module is further configured to:

[0069] Perform a point cloud to depth map conversion on the rigidly registered point cloud p r1 and the point cloud p to be registered f0 respectively, convert the sole space point cloud into a depth map, and obtain the reference image i r0 and the image i to be registered f0 , and synchronously perform a point cloud to depth map conversion on the rigidly registered trajectory t1, and convert it into the teaching trajectory image i t0 , completing the dynamic time warping flexible transformation at the image level.

[0070] Specifically, the flexible matching module is further configured to:

[0071] Based on the image i to be registered f0 , perform dtw processing on the rows and columns of the reference image i r0 respectively, so as to obtain the flexible matching image i f0 most similar to the image i to be registered r1 , and correspondingly, synchronously perform dtw processing on the rows and columns of the teaching trajectory image i t0 , and obtain the flexible matching trajectory image i t1 after flexible matching.

[0072] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a sole cementing trajectory based on dynamic time warping, characterized in that The method includes: Step 1: Using the taught sole point cloud as the original reference point cloud, the sole point cloud to be matched as the original point cloud to be registered, and the original reference point cloud, the original point cloud to be registered, and the taught trajectory t0 as input parameters; Step 2: Preprocess the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud p to be registered f0 ; Step 3: Based on the point cloud p to be registered f0 Perform point cloud ICP rigid body transformation g on the reference point cloud p r0 to obtain the rigidly registered point cloud p r , and at the same time, perform a synchronous rigid body transformation on the teaching trajectory t0 to obtain the rigidly registered trajectory t1; r1 ​ Step 4: Convert the sole space point cloud into a depth map to obtain the reference image i r0 and the image i to be registered f0 , and the just-matched trajectory t1 is synchronously converted into a teaching trajectory image i t0 ; Step 5: Based on the image i to be registered f0 Perform dynamic time warping on the reference image i r0 to obtain the soft-matched image i r1 , and perform synchronous transformation on the taught trajectory image i t0 to obtain the soft-matched trajectory image i t1 ; Step 6: Finding the correspondence between the soft registration trajectory image data after flexible registration and the corresponding points of the taught trajectory t0, and restoring the depth direction value during teaching, so as to finally obtain the reproduced trajectory t2, and the glue application robot performs glue application according to the reproduced trajectory t2.

2. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 1, wherein The said Step 2 includes: Through a series of preprocessing operations such as filtering, segmentation, and sampling, the taught sole and the sole to be registered are separated from the original data respectively, so as to obtain the reference point cloud p r0 and the point cloud p to be registered f0 .

3. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 1, wherein The said Step 3 includes: Based on the point cloud p to be registered f0 For the reference point cloud p r0 Implement the point cloud ICP rigid body transformation g r , and adjust the position and orientation of the reference point cloud p r0 to the state closest to the point cloud p to be registered f0 to obtain the rigidly registered point cloud p r1 . At the same time, perform a synchronous rigid body transformation g on the taught trajectory t0 r to obtain the rigidly registered trajectory t1, thereby compensating for the installation deviation between the two soles and completing the rough registration of their positions and orientations.

4. The method for generating a sole coating trajectory based on dynamic time warping according to claim 1, wherein The said Step 4 includes: For the just-matched point cloud p r1 and the point cloud p to be registered f0 Perform a point cloud to depth map conversion respectively, convert the sole space point cloud into a depth map, and obtain the reference image i r0 and the image i to be registered f0 , Synchronously perform a point cloud to depth map conversion on the just-matched trajectory t1, and convert it into a teaching trajectory image i t0 , Complete the dynamic time warping flexible transformation at the image level.

5. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 1, wherein The said Step 5 includes: Based on the image \(i\) to be registered f0 , perform DTW processing on the rows and columns of the reference image \(i\) r0 respectively, so as to obtain the soft-matched image \(i\) f0 that is most similar to the image \(i\) to be registered r1 . Correspondingly, perform DTW processing on the rows and columns of the teaching trajectory image \(i\) t0 synchronously, and obtain the soft-matched trajectory image \(i\) after soft matching t1 .

6. A sole gluing trajectory generation device based on dynamic time warping, characterized in that The said device includes: An input module, which is used to use the taught sole point cloud as the original reference point cloud, the sole point cloud to be matched as the original point cloud to be registered, and the original reference point cloud, the original point cloud to be registered, and the taught trajectory t0 as input parameters; A preprocessing module for preprocessing the original reference point cloud and the original point cloud to be registered to obtain the reference point cloud p r0 and the point cloud p to be registered f0 ; A rigid body transformation module, which is used to perform, based on the point cloud p to be registered f0 on the reference point cloud p r0 a point cloud ICP rigid body transformation g r to obtain the rigidly registered point cloud p r1 and at the same time perform a synchronous rigid body transformation on the teaching trajectory t0 to obtain the rigidly registered trajectory t1; A depth map conversion module for converting the sole spatial point cloud into a depth map to obtain a reference image i r0 and the image i to be registered f0 , and the just-matched trajectory t1 is synchronously converted into a teaching trajectory image i t0 ; A flexible registration module for dynamically time-warping a reference image i based on a to-be-registered image i f0 to obtain a flexibly registered image i r0 and synchronously transforming a taught trajectory image i r1 to obtain a flexibly registered trajectory image i t0 ; t1 ​ A glue application generation module, which is used to find the correspondence between the soft registration trajectory image data after flexible registration and the corresponding points of the taught trajectory t0, and restore the depth direction value during teaching, so as to finally obtain the reproduced trajectory t2, and the glue application robot performs glue application according to the reproduced trajectory t2.

7. The method for generating a sole glue spreading trajectory based on dynamic time warping according to claim 6, wherein The said preprocessing module is also used for: Through a series of preprocessing operations such as filtering, segmentation, and sampling, the taught sole and the sole to be registered are separated from the original data respectively, so as to obtain the reference point cloud p r0 and the point cloud p to be registered f0 .

8. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 6, characterized in that, The said rigid body transformation module is also used for: Based on the point cloud to be registered f0 For the reference point cloud p r0 Implement point cloud ICP rigid body transformation g r , the reference point cloud p r0 The position and posture are adjusted to the point cloud p to be registered f0 The closest state is obtained by just matching the point cloud p r1 , and at the same time teach the trajectory t0 to do a synchronous rigid body transformation g r , and obtain the rigidly matched trajectory t1, thereby compensating for the installation deviation between the two soles and completing the rough alignment of their positions and postures.

9. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 6, wherein The said depth map conversion module is also used for: For the just - matched point cloud p r1 and the point cloud p to be registered f0 perform a point - cloud - to - depth - map conversion respectively, convert the sole - space point cloud into a depth map, and obtain the reference image i r0 and the image i to be registered f0 , synchronously perform a point - cloud - to - depth - map conversion on the just - matched trajectory t1, and convert it into a teaching trajectory image i t0 , and complete the dynamic time warping flexible transformation at the image level.

10. The method for generating a sole gluing trajectory based on dynamic time warping according to claim 6, wherein The said flexible registration module is also used for: Based on the image i to be registered f0 , respectively perform dtw processing on the rows and columns of the reference image i r0 to obtain the flexible registration image i f0 that is most similar to the image i to be registered r1 . Correspondingly, synchronously perform dtw processing on the rows and columns of the teaching trajectory image i t0 to obtain the flexible registration trajectory image i t1 after flexible registration.

Images