AI insole modeling system based on multi-modal biomechanics and 3D printing method

Abstract

The invention discloses an AI insole modeling system based on multi-modal biomechanics and a 3D printing method, and relates to the technical field of biomedical engineering, and the AI insole modeling system specifically comprises an external parameter calibration module, an image processing module, a three-dimensional reconstruction module, a measurement and analysis module, a foot database establishment module, an insole database modeling module and an insole 3D printing module. Based on the biomechanical principle and the digital medical technology, a full-link digital service system of three-dimensional scanning, AI modeling and 3D printing is constructed, firstly, based on a foot scanning and analysis system of multi-view three-dimensional reconstruction, through the processing flow of camera calibration, image distortion removal, semantic segmentation, multi-view three-dimensional reconstruction and measurement, a full-link digital service system of three-dimensional scanning, AI modeling and 3D printing is constructed; and finally, key parameters of the foot length, the foot width, the hallux valgus angle and the flatfoot degree are output, high-precision foot modeling and measurement are achieved, accurate customization of the orthopedic insoles is achieved, and the device is suitable for the fields of medical diagnosis, orthopedic shoe customization and sports science and has the characteristics of high precision, automation and modularization.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, specifically to an AI insole modeling system and 3D printing method based on multimodal biomechanics. Background Technology

[0002] Insoles, as an important footwear accessory, have a history dating back to ancient times. Initially, they were mainly used to increase the comfort and extend the lifespan of shoes. Over time and with technological advancements, the design and function of insoles have been significantly improved and developed. Currently, orthotic insoles, as an important assistive device for improving foot biomechanical problems, have their core value in achieving a balance between support and comfort through scientific design. Orthotic insoles have now evolved from simple medical assistive tools into lifestyle products that integrate health, fashion, and personalization. Currently, the public generally lacks awareness of foot health problems and often ignores early symptoms. Typical symptoms include flat feet leading to collapsed arches, high arches causing insufficient cushioning, pronation and varus causing gait abnormalities, and hallux valgus causing joint deformities. Failure to intervene in these problems in time can lead to more serious complications and even affect walking function. Summary of the Invention

[0003] This invention provides an AI insole modeling system and 3D printing method based on multimodal biomechanics, which can effectively address the common lack of awareness of foot health issues among the public mentioned in the background, often ignoring early symptoms. Typical symptoms include collapsed arches caused by flat feet, insufficient cushioning caused by high arches, gait abnormalities caused by pronation and supination, and joint deformities caused by hallux valgus. Failure to intervene in these problems in time can lead to more serious complications and even affect walking function.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an AI insole modeling system based on multimodal biomechanics, which, through a processing flow of camera calibration, image distortion correction, semantic segmentation, multi-view 3D reconstruction and measurement, ultimately outputs key parameter information such as foot length, foot width, hallux valgus angle and flat foot degree, to achieve high-precision foot modeling and measurement, and combines 3D printing technology to achieve precise customization of orthopedic insoles; Specifically, it includes an external parameter calibration module, an image processing module, a 3D reconstruction module, a measurement and analysis module, a foot database establishment module, an insole database modeling module, and an insole 3D printing module.

[0005] According to the above technical solution, the external parameter calibration module uses a high-precision calibration box as a spatial reference object, and calculates the camera's external parameters through multi-view image analysis to achieve spatial alignment of the multi-camera system. The camera's external parameters include position and attitude. The intrinsic and extrinsic parameters of the camera are calculated through marker point detection and feature matching. The intrinsic parameters of the camera specifically include focal length, principal point, and distortion coefficient, while the extrinsic parameters of the camera specifically include rotation matrix and translation vector.

[0006] According to the above technical solution, the image processing module specifically includes a distortion correction unit and a semantic segmentation unit; The distortion correction unit is used to eliminate lens distortion. Based on the Brown-Conrady distortion model, it uses OpenCV's `cv2.undistort()` function to correct radial and tangential distortion, and outputs a corrected distortion-free image, providing high-quality image data for subsequent segmentation and reconstruction. The semantic segmentation unit achieves accurate separation of the foot from the background. Specifically, it uses the PointRend model for pixel-level segmentation. The training data is manually labeled foot masks, the input is a distortion-free RGB image, and the output is a binary mask image, providing a clear foot outline for 3D reconstruction.

[0007] According to the above technical solution, the three-dimensional reconstruction module specifically includes a Dust3R submodule, a normal estimation submodule, and a three-dimensional reconstruction and rendering submodule; The Dust3R submodule generates initial point cloud data of the foot based on multi-view images. It selects images from 7 key viewpoints and uses an exhaustive pairing method to generate 21 ordered image pairs, covering all possible viewpoint combinations. The image is input to the model and then processed by a ViT-Large encoder with shared weights and a Transformer decoder with a ViT-Base architecture. The output is a point map and a corresponding confidence map through a cross-attention mechanism. Each point map is a W×H×3 tensor, where each pixel position (i,j) corresponds to a 3D point coordinate (X,Y,Z). The confidence map output by the network is used to filter low confidence points and improve the quality of the point cloud. The normal estimation submodule adopts a lightweight CNN architecture, combined with a loop update mechanism, to predict the normals of the foot image and generate the normal vector corresponding to each pixel, providing important geometric information for subsequent high-precision 3D reconstruction. The encoder is a feature extraction network built on EfficientNet B5, the decoder is a feature extraction network that is progressively upsampled and fused with features at different scales to generate an initial normal map, hidden states and context features, and ConvGRU units are used to iteratively optimize the normal map. The 3D reconstruction and rendering submodule is based on the distortion-reduced image, initial point cloud, mask image, and normal map. The initial point cloud initializes the position and distribution of Gaussian surface elements, and the mask is used to distinguish the foreground and background regions to avoid noise interference. In the optimization stage, the color, depth, and normal map are calculated through differentiable rendering and compared with the input normal map to optimize the surface geometry. After optimization, the depth and normal data from multi-view rendering are used to reconstruct a high-quality 3D mesh using TSDF. The entire process makes full use of the complementary information of all input data to ensure the accuracy and stability of the output.

[0008] According to the above technical solution, the measurement and analysis module is based on the reconstructed three-dimensional mesh model and automatically calculates the key morphological parameters of the foot according to the FIND parametric model. The specific parameters include foot length, foot width, hallux valgus and flat feet. Specifically, foot length is determined by the Euclidean distance from the toe to the heel, foot width is calculated by the distance between the two widest points of the foot cross-section, hallux valgus is assessed by detecting key points of the first and second metatarsals and calculating the HVA joint angle, and flat feet are assessed by calculating the arch index based on the output of pressure plate data.

[0009] According to the above technical solution, the foot database establishment module mainly collects millions of foot biomechanical data, specifically including plantar pressure distribution maps, 3D foot scans, and gait analysis videos; Label the abnormality type and severity level, and establish a standardized data format. The abnormality types include foot arch collapse, high arch, and inversion / exversion. The specific data formats include pressure value matrix and STL format 3D model.

[0010] According to the above technical solution, the insole database modeling module is based on the foot clinical database to establish a three-dimensional insole model database, including insoles with three different arch heights corresponding to flat feet, corresponding to mild flat feet, moderate flat feet and severe flat feet respectively. A high arch insole is made for high arch feet to meet the needs of most people. Three different degrees of insole tilt and height models are made according to the inversion and supination angles of the heel. Based on the input AI diagnostic results, the database backend will automatically upload the best matching basic insole model. The diagnostic results include specific foot size, flat feet or high arches. Customers can also manually select the model data they need, and finally output a standardized STL model file.

[0011] According to the above technical solution, the insole 3D printing module uses 3D printed TPU elastic material and realizes the honeycomb structure design through high-precision dual-station 3D printing technology; This honeycomb structure design provides dynamic support force transmission and enables multi-directional energy rebound. Its porosity is precisely controlled within the range of 25%-30%, and the printing speed is controlled between 6.5-8 cubic millimeters per second depending on the material's hardness to ensure the best printing results.

[0012] According to the above technical solution, the AI ​​insole modeling system and 3D printing method based on multimodal biomechanics includes the following steps: Step S1: Collect personalized correction data; Step S2: Obtain the 3D model of the insole; Step S3: 3D printing technology is used to print the insoles.

[0013] According to the above technical solution, step S1 is to use 24 high-precision matrix cameras to establish a high-precision three-dimensional reconstructed foot model. The scanning process does not require contact with the skin and the whole foot scan can be completed within 3 seconds. Simultaneously, it can collect foot pressure data and combine it with AI point cloud processing algorithms to generate a digital twin model of the foot with millimeter-level precision, providing a precise data foundation for subsequent personalized correction; Step S2 involves using a large AI model trained on a database of millions of clinical cases to automatically identify areas of abnormal plantar pressure distribution through deep learning algorithms, and to accurately locate biomechanical problems such as arch collapse and inversion / exversion. Based on this, the built-in model automatically outputs a 3D insole model file that adapts to individual gait characteristics, eliminating the need for manual secondary modeling; This file contains topological data accurate to 0.5mm resolution, which can be directly imported into a 3D printing system to ensure that every support surface can perfectly fit the user's unique foot anatomy. Step S3 involves using 3D printed TPU elastic material and high-precision dual-station 3D printing technology to achieve a honeycomb structure design. This structure can provide dynamic support force transmission and achieve multi-directional energy rebound, with its porosity precisely controlled within the range of 25%-30%.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: Based on biomechanical principles and digital medical technology, this invention constructs a full-chain digital service system encompassing 3D scanning, AI modeling, and 3D printing. It first uses a foot scanning and analysis system based on multi-view 3D reconstruction. Through processing steps such as camera calibration, image distortion correction, semantic segmentation, multi-view 3D reconstruction, and measurement, it ultimately outputs key parameters such as foot length, foot width, hallux valgus angle, and degree of flat feet, achieving high-precision foot modeling and measurement. Furthermore, by acquiring three-dimensional foot data through high-precision scanning, using deep learning algorithms to analyze potential foot health risks and perform adaptive 3D insole modeling, and finally using 3D printing technology to achieve precise customization of orthopedic insoles, it is applicable to the fields of medical diagnosis, orthopedic shoe customization, and sports science, and features high precision, automation, and modularity. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0016] In the attached diagram: Figure 1 This is a structural block diagram of the modeling system of the present invention; Figure 2 This is a flowchart of the 3D printing method of the present invention; Figure 3 This is a schematic diagram illustrating the acquisition of the three-dimensional model of the insole of this invention; Figure 4 This is a schematic diagram of the 3D printing of the insole of this invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] Example: Figure 1 As shown, the present invention provides a technical solution, an AI insole modeling system based on multimodal biomechanics, which, through a processing flow of camera calibration, image distortion correction, semantic segmentation, multi-view 3D reconstruction and measurement, finally outputs key parameter information of foot length, foot width, hallux valgus angle and flat foot degree, to achieve high-precision foot modeling and measurement, and combines 3D printing technology to achieve precise customization of orthopedic insoles; Specifically, it includes an external parameter calibration module, an image processing module, a 3D reconstruction module, a measurement and analysis module, a foot database establishment module, an insole database modeling module, and an insole 3D printing module.

[0019] Based on the above technical solution, the external parameter calibration module uses a high-precision calibration box as a spatial reference object, and calculates the camera's external parameters through multi-view image analysis to achieve spatial alignment of the multi-camera system. The camera's external parameters include position and attitude. The intrinsic and extrinsic parameters of the camera are calculated through marker point detection and feature matching. The intrinsic parameters of the camera specifically include focal length, principal point and distortion coefficient, while the extrinsic parameters of the camera specifically include rotation matrix and translation vector. The RANSAC EPnP solver is used to optimize multi-camera pose and improve calibration accuracy. It mainly outputs the transformation matrix (R|t) from the camera coordinate system to the world coordinate system, providing a geometric constraint basis for 3D reconstruction.

[0020] Based on the above technical solution, the image processing module specifically includes a distortion correction unit and a semantic segmentation unit; The distortion correction unit is used to eliminate lens distortion and ensure the accuracy of image geometry. Based on the Brown-Conrady distortion model, it uses OpenCV's `cv2.undistort()` function to correct radial and tangential distortion, and outputs a distort-free image after correction, providing high-quality image data for subsequent segmentation and reconstruction. The semantic segmentation unit achieves accurate separation of the foot from the background. Specifically, it uses the PointRend model for pixel-level segmentation. The training data is manually labeled foot masks, the input is a distortion-free RGB image, and the output is a binary mask image, where the foot region is 1 and the background is 0, providing a clear foot contour for 3D reconstruction.

[0021] Based on the above technical solution, the 3D reconstruction module specifically includes the Dust3R submodule, the normal estimation submodule, and the 3D reconstruction and rendering submodule. Among them, the Dust3R submodule is for multi-view 3D reconstruction. The Dust3R submodule generates initial point cloud data of the foot based on multi-view images. It selects images from 7 key viewpoints and uses an exhaustive pairing method to generate 21 ordered image pairs, covering all possible viewpoint combinations. The image is input to the model and then processed by a ViT-Large encoder with shared weights and a Transformer decoder with a ViT-Base architecture. The output is a point map and a corresponding confidence map through a cross-attention mechanism. Each point map is a W×H×3 tensor, where each pixel position (i,j) corresponds to a 3D point coordinate (X,Y,Z). The confidence map output by the network is used to filter low confidence points and improve the quality of the point cloud. The normal estimation submodule adopts a lightweight CNN architecture, combined with a loop update mechanism, to predict the normals of the foot image and generate the normal vector corresponding to each pixel, providing important geometric information for subsequent high-precision 3D reconstruction. The encoder is a feature extraction network built on EfficientNet B5, the decoder is a feature extraction network that is progressively upsampled and fused with features at different scales to generate an initial normal map, hidden states and context features, and ConvGRU units are used to iteratively optimize the normal map. The 3D reconstruction and rendering submodule is based on the distortion-free image, initial point cloud, mask map, and normal map. The initial point cloud initializes the position and distribution of Gaussian surface elements, while the mask is used to distinguish the foreground and background regions to avoid noise interference. In the optimization stage, color, depth, and normal maps are calculated through differentiable rendering and compared with the input normal map to optimize the surface geometry. During the reconstruction process, a multi-task loss function is used, including photometric loss, normal prior loss, depth-normal consistency loss, opacity loss, and mask loss, which together guide the parameter optimization of Gaussian surface elements; After optimization, the depth and normal data from multi-view rendering are used to reconstruct a high-quality 3D mesh using TSDF. The entire process makes full use of the complementary information of all input data to ensure the accuracy and stability of the output.

[0022] Based on the above technical solution, the measurement and analysis module is based on the reconstructed three-dimensional mesh model and automatically calculates the key morphological parameters of the foot according to the FIND parametric model. The specific parameters include foot length, foot width, hallux valgus and flat feet. Specifically, foot length is determined by the Euclidean distance from the toe to the heel, foot width is calculated by the distance between the two widest points of the foot cross-section, hallux valgus is assessed by detecting key points of the first and second metatarsals and calculating the HVA joint angle, and flat feet are assessed by calculating the arch index based on the output of pressure plate data.

[0023] Based on the above technical solution, the foot database establishment module mainly collects millions of foot biomechanical data, specifically including plantar pressure distribution maps, 3D foot scans, and gait analysis videos; Label the abnormality type and severity level, and establish a standardized data format. The abnormality types include foot arch collapse, high arch, and inversion / exversion. The specific data formats include pressure value matrix and STL format 3D model.

[0024] Based on the above technical solution, the insole database modeling module is based on the foot clinical database to establish a three-dimensional insole model database, including insoles with three different arch heights corresponding to flat feet, corresponding to mild flat feet, moderate flat feet and severe flat feet respectively. A high arch insole is made for high arch feet to meet the needs of most people. For pronation and supination, three different degrees of insole tilt height models are made according to the heel pronation and supination angles, with tilt heights of 5 degrees, 10 degrees and 15 degrees respectively. Hallux valgus correction mainly involves raising the insole portion of the big toe. At the same time, a database of insole models has been created for customers with both flat feet and pronation, containing 9 different insole models. Mild flat feet correspond to three degrees of pronation, moderate flat feet correspond to three different degrees of pronation, severe flat feet correspond to three different degrees of pronation, hallux valgus corresponds to three different degrees of pronation, and hallux valgus corresponds to three different degrees of flat feet. Comfortable insoles are available to meet the needs of the general public. Currently, the database supports 27 different insole types for single foot insoles and 54 types for double foot insoles, covering sizes from 30 to 45. The database backend contains a total of 864 insole models, each with different levels of support height and range. Based on the input AI diagnostic results, the database backend automatically uploads and matches the best basic insole model. The diagnostic results include specific foot shape and size, whether it is flat foot or high arch. For special needs, customers can also manually select the model data they need, and finally output a standardized STL model file.

[0025] Based on the above technical solution, the insole 3D printing module uses 3D printed TPU elastic material and realizes the honeycomb structure design through high-precision dual-station 3D printing technology. This honeycomb structure design provides dynamic support force transmission and multi-directional energy rebound. Its porosity is precisely controlled within the range of 25%-30% to balance air permeability and durability. The printing speed is controlled between 6.5-8 cubic millimeters per second depending on the hardness of the material to ensure the best printing effect.

[0026] like Figure 2 As shown, the AI ​​insole modeling system and 3D printing method based on multimodal biomechanics includes the following steps: Step S1: Collect personalized correction data; Step S2: Obtain the 3D model of the insole; Step S3: 3D printing technology is used to print the insoles.

[0027] Based on the above technical solution, step S1 is to use 24 high-precision matrix cameras to establish a high-precision three-dimensional reconstructed foot model. The scanning process does not require contact with the skin and the whole foot scan can be completed within 3 seconds. Simultaneously, it can collect foot pressure data and combine it with AI point cloud processing algorithms to generate a digital twin model of the foot with millimeter-level precision, providing a precise data foundation for subsequent personalized correction; like Figure 3 As shown, step S2 is an AI model trained on a database of millions of clinical cases. It uses deep learning algorithms to automatically identify areas of abnormal plantar pressure distribution and accurately locate biomechanical problems such as arch collapse and inversion / exversion. Based on this, the built-in model automatically outputs a 3D insole model file that adapts to individual gait characteristics, eliminating the need for manual secondary modeling; This file contains topological data accurate to 0.5mm resolution, which can be directly imported into a 3D printing system to ensure that every support surface can perfectly fit the user's unique foot anatomy. like Figure 4 As shown, step S3 involves using 3D printed TPU elastic material and high-precision dual-station 3D printing technology to achieve the honeycomb structure design. This structure provides dynamic support force transmission and multi-directional energy rebound, with its porosity precisely controlled within the 25%-30% range to balance breathability and durability.

[0028] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI insole modeling system based on multimodal biomechanics, characterized in that: Through the processing flow of camera calibration, image distortion correction, semantic segmentation, multi-view 3D reconstruction and measurement, the key parameter information of foot length, foot width, hallux valgus angle and flat foot degree is finally output, realizing high-precision foot modeling and measurement, and combining 3D printing technology to realize the precise customization of orthopedic insoles. Specifically, it includes an external parameter calibration module, an image processing module, a 3D reconstruction module, a measurement and analysis module, a foot database establishment module, an insole database modeling module, and an insole 3D printing module.

2. The AI ​​insole modeling system based on multimodal biomechanics according to claim 1, characterized in that: The external parameter calibration module uses a high-precision calibration box as a spatial reference object and calculates the camera's external parameters through multi-view image analysis to achieve spatial alignment of the multi-camera system. The camera's external parameters include position and attitude. The intrinsic and extrinsic parameters of the camera are calculated through marker point detection and feature matching. The intrinsic parameters of the camera specifically include focal length, principal point, and distortion coefficient, while the extrinsic parameters of the camera specifically include rotation matrix and translation vector.

3. The AI ​​insole modeling system based on multimodal biomechanics according to claim 2, characterized in that: The image processing module specifically includes a distortion correction unit and a semantic segmentation unit; The distortion correction unit is used to eliminate lens distortion. Based on the Brown-Conrady distortion model, it uses OpenCV's `cv2.undistort()` function to correct radial and tangential distortion, and outputs a corrected distortion-free image, providing high-quality image data for subsequent segmentation and reconstruction. The semantic segmentation unit achieves accurate separation of the foot from the background. Specifically, it uses the PointRend model for pixel-level segmentation. The training data is manually labeled foot masks, the input is a distortion-free RGB image, and the output is a binary mask image, providing a clear foot outline for 3D reconstruction.

4. The AI ​​insole modeling system based on multimodal biomechanics according to claim 3, characterized in that: The 3D reconstruction module specifically includes the Dust3R submodule, the normal estimation submodule, and the 3D reconstruction and rendering submodule; The Dust3R submodule generates initial point cloud data of the foot based on multi-view images. It selects images from 7 key viewpoints and uses an exhaustive pairing method to generate 21 ordered image pairs, covering all possible viewpoint combinations. The image is input to the model and then processed by a ViT-Large encoder with shared weights and a Transformer decoder with a ViT-Base architecture. The output is a point map and a corresponding confidence map through a cross-attention mechanism. Each point map is a W×H×3 tensor, where each pixel position (i,j) corresponds to a 3D point coordinate (X,Y,Z). The confidence map output by the network is used to filter low confidence points and improve the quality of the point cloud. The normal estimation submodule adopts a lightweight CNN architecture, combined with a loop update mechanism, to predict the normals of the foot image and generate the normal vector corresponding to each pixel, providing important geometric information for subsequent high-precision 3D reconstruction. The encoder is a feature extraction network built on EfficientNet B5, the decoder is a feature extraction network that is progressively upsampled and fused with features at different scales to generate an initial normal map, hidden states and context features, and ConvGRU units are used to iteratively optimize the normal map. The 3D reconstruction and rendering submodule is based on the distortion-reduced image, initial point cloud, mask image, and normal map. The initial point cloud initializes the position and distribution of Gaussian surface elements, and the mask is used to distinguish the foreground and background regions to avoid noise interference. In the optimization stage, the color, depth, and normal map are calculated through differentiable rendering and compared with the input normal map to optimize the surface geometry. After optimization, the depth and normal data from multi-view rendering are used to reconstruct a high-quality 3D mesh using TSDF. The entire process makes full use of the complementary information of all input data to ensure the accuracy and stability of the output.

5. The AI ​​insole modeling system based on multimodal biomechanics according to claim 4, characterized in that: The measurement and analysis module is based on the reconstructed three-dimensional mesh model and automatically calculates key morphological parameters of the foot according to the FIND parametric model. Specific parameters include foot length, foot width, hallux valgus, and flat feet. Specifically, foot length is determined by the Euclidean distance from the toe to the heel, foot width is calculated by the distance between the two widest points of the foot cross-section, hallux valgus is assessed by detecting key points of the first and second metatarsals and calculating the HVA joint angle, and flat feet are assessed by calculating the arch index based on the output of pressure plate data.

6. The AI ​​insole modeling system based on multimodal biomechanics according to claim 5, characterized in that: The foot database establishment module mainly collects millions of foot biomechanical data, specifically including plantar pressure distribution maps, 3D foot scans, and gait analysis videos; Label the abnormality type and severity level, and establish a standardized data format. The abnormality types include foot arch collapse, high arch, and inversion / exversion. The specific data formats include pressure value matrix and STL format 3D model.

7. The AI ​​insole modeling system based on multimodal biomechanics according to claim 6, characterized in that: The insole database modeling module is based on a foot clinical database to establish a three-dimensional insole model database, including insoles with three different arch heights corresponding to flat feet, corresponding to mild, moderate and severe flat feet respectively. A high arch insole is designed for high arch feet to meet the needs of most people. Three different insole tilt and height models are designed for pronation and supination based on the heel pronation and supination angles respectively. Based on the input AI diagnostic results, the database backend will automatically upload the best matching basic insole model. The diagnostic results include specific foot size, flat feet or high arches. Customers can also manually select the model data they need, and finally output a standardized STL model file.

8. The AI ​​insole modeling system based on multimodal biomechanics according to claim 7, characterized in that: The insole 3D printing module uses 3D printed TPU elastic material and achieves a honeycomb structure design through high-precision dual-station 3D printing technology. This honeycomb structure design provides dynamic support force transmission and enables multi-directional energy rebound. Its porosity is precisely controlled within the range of 25%-30%, and the printing speed is controlled between 6.5-8 cubic millimeters per second depending on the material's hardness to ensure the best printing results.

9. An AI insole modeling system and 3D printing method based on multimodal biomechanics, characterized in that: The printing method of the AI ​​insole modeling system based on multimodal biomechanics according to claim 7 includes the following steps: Step S1: Collect personalized correction data; Step S2: Obtain the 3D model of the insole; Step S3: 3D printing technology is used to print the insoles.

10. The AI ​​insole 3D printing method based on multimodal biomechanics according to claim 9, characterized in that: Step S1 involves using 24 high-precision matrix cameras to establish a high-precision three-dimensional reconstructed foot model. The scanning process does not require contact with the skin and can complete the full foot scan within 3 seconds. Simultaneously, it can collect foot pressure data and combine it with AI point cloud processing algorithms to generate a digital twin model of the foot with millimeter-level precision, providing a precise data foundation for subsequent personalized correction; Step S2 involves using a large AI model trained on a database of millions of clinical cases to automatically identify areas of abnormal plantar pressure distribution through deep learning algorithms, and to accurately locate biomechanical problems such as arch collapse and inversion / exversion. Based on this, the built-in model automatically outputs a 3D insole model file that adapts to individual gait characteristics, eliminating the need for manual secondary modeling; This file contains topological data accurate to 0.5mm resolution, which can be directly imported into a 3D printing system to ensure that every support surface can perfectly fit the user's unique foot anatomy. Step S3 involves using 3D printed TPU elastic material and high-precision dual-station 3D printing technology to achieve a honeycomb structure design. This structure can provide dynamic support force transmission and achieve multi-directional energy rebound, with its porosity precisely controlled within the range of 25%-30%.

Images