Artificial intelligence assisted make-up and make-up robot collaborative make-up method

Abstract

The invention discloses an artificial intelligence assisted makeup and makeup robot collaborative makeup method, and belongs to the technical field of artificial intelligence and robots. The method comprises the following steps: collecting user face data through a multi-modal sensor, extracting multi-scale features based on improved DenseNet and HRNet networks, and generating a three-dimensional face model; a high-fidelity virtual makeup effect is generated by using a conditional generative adversarial network in combination with user requirements, and real-time comparison and display are performed through an augmented reality interface; the makeup difference is quantified through multi-dimensional feature point mapping, multi-color space analysis and texture comparison, and correction suggestions are dynamically generated; operation is executed through the seven-degree-of-freedom redundant mechanical arm, a hybrid force control strategy and a visual monitoring system are integrated, and safe and accurate execution is ensured. The method solves the problems of uncontrollable traditional makeup effect, low execution precision and lack of dynamic feedback, has the characteristics of high adaptability, real-time response and zero-damage safety, and can be widely applied to the fields of family beauty, film and television makeup and special crowd nursing.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method for collaborative makeup using artificial intelligence-assisted makeup and a makeup robot. Background Art

[0002] Traditional makeup techniques rely heavily on manual labor, requiring users to apply makeup manually in front of a mirror or rely on professional makeup artists. This has the following significant drawbacks: Uncontrollable effects: Non-professional users find it difficult to master makeup techniques accurately, and often encounter problems such as improper color matching and asymmetrical lines; Time-consuming and inefficient: Complex makeup requires repeated revisions, which takes up to several hours. This is especially unfriendly to high-frequency makeup needs (such as daily commuting). Lack of dynamic guidance: Existing makeup assistance tools (such as tutorial videos) cannot analyze users' facial features in real time and make it difficult to provide personalized correction suggestions; Technical limitations: Early automated makeup equipment relied on fixed programs and could not adapt to diverse facial features and scene requirements. It also lacked safety and precision, which could easily lead to cosmetic waste or skin damage.

[0003] To address the above issues, existing technologies attempt to introduce computer vision and robotics technologies, but the following bottlenecks still exist: Facial feature analysis lacks accuracy, making it difficult to capture fine-grained skin texture and three-dimensional contours; The makeup effect generation algorithm has low fidelity and is prone to losing user identity features during style transfer. The robot execution system lacks a dynamic feedback mechanism and cannot correct operational deviations in real time; Multimodal interaction capabilities are weak, and user demand analysis and execution strategy optimization are disconnected. Summary of the Invention

[0004] In response to the above-mentioned deficiencies in the prior art, the present invention provides a method for collaborative makeup using artificial intelligence-assisted makeup and a makeup robot.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: A method for collaborative makeup application using artificial intelligence-assisted makeup and a makeup robot, comprising the following steps: S1. Collect user facial data through multimodal sensors, extract multi-scale facial features based on the improved DenseNet deep convolutional neural network, and generate a 3D facial reconstruction model; S2. Combine user needs and scenario information, use conditional generative adversarial networks to generate personalized makeup plans, and display the target effect and real-time comparison through a multimodal interactive interface; S3, based on a real-time feedback mechanism, quantifies the difference between the current makeup effect and the target effect through multi-dimensional feature point mapping, color space analysis and texture comparison, and generates dynamic correction suggestions; S4. Perform makeup operations through a multi-degree-of-freedom robotic arm, combine force feedback control with visual monitoring to achieve safe and accurate cosmetic application, and continuously optimize the makeup process based on user feedback.

[0006] Furthermore, the specific method of extracting facial features in S1 is: S11, uses multi-stage densely connected blocks to extract facial features and integrates multi-scale semantic information through feature pyramid networks; S12. Locate facial key points based on the high-resolution representation network and generate standard feature point coordinates and confidence evaluation results; S13. Use the pixel-level skin analysis network to output skin quality classification results, including oil distribution, pore density, and pigmentation area heat map; S14. Generate a parametric facial model based on single-image 3D reconstruction technology to preview makeup effects from multiple angles.

[0007] Furthermore, the specific method of generating the makeup plan in S2 is: S21. Analyze makeup needs by fusing user voice, text, and gesture inputs through a cross-modal Transformer. S22, Generate high-fidelity virtual makeup effects based on the improved StyleGAN architecture and achieve style transfer through adaptive instance normalization; S23, constructing a regionalized expert network set to generate specialized makeup step sequences for the eyes, lips, and contour areas respectively; S24. Decompose the makeup plan into structured parameterized operation instructions, including tool selection, action trajectory and force parameters.

[0008] Furthermore, the specific method of the real-time feedback mechanism in S3 is: S31. Calculate the feature point position deviation using the weighted Euclidean distance. The formula is:

[0009] Where, and are the corresponding feature points of the current effect and the target effect respectively, The importance weight of feature points is dynamically adjusted through the attention mechanism; S32, calculating the regionalized histogram similarity in the multi-color space and quantifying the color difference by EMD; S33, extracting texture features based on Gabor filter and gray level co-occurrence matrix, and calculating the matching degree between the current texture and the target texture; S34. Generate a comprehensive proximity score through the hierarchical scoring aggregation module. The formula is:

[0010] Where, is the comprehensive proximity score, For the Ratings for each dimension, is the nonlinear adjustment factor, is the importance weight of the feature point, is the total number of dimensions.

[0011] Furthermore, the specific manner in which the robot performs the steps in S4 is as follows: S41, using redundant degrees of freedom robotic arms to plan the motion trajectory of makeup tools, and realizing path optimization based on the rapid exploration random tree algorithm; S42, using a six-dimensional force sensor and a hybrid position / force control strategy to ensure that the contact force is below a preset safety threshold; S43, integrated electromagnetic and pneumatic composite clamping system, adapts to different tools such as eyebrow pencils and powder puffs, and realizes automatic tool switching based on RFID and visual recognition; S44. Monitor the user's head movements and facial expressions in real time, and avoid sudden movements through the dynamic trajectory adjustment module.

[0012] Furthermore, it also includes: S5. Optimize and modify strategies based on the reinforcement learning framework, update the knowledge base based on user feedback, and form a closed-loop iterative optimization system.

[0013] Furthermore, the specific method of closed-loop optimization in S5 is: S51, capturing user responses through facial expression analysis and operation duration recording, and generating reward signals; S52. Use incremental learning and knowledge distillation technology to update the makeup expert system to avoid catastrophic forgetting; S53. Deploy the optimized strategy to the cloud server to achieve multi-device collaborative learning and template sharing.

[0014] The present invention has the following beneficial effects: 1. High-precision adaptation: The HRNet network achieves facial key point positioning error ≤ 0.1 pixel, supports complex contour and weak texture area analysis, and adapts to most facial features; 2. Dynamic Real-Time Response: A multi-dimensional feedback mechanism (color / texture / shape) combined with RRT* path planning reduces trajectory adjustment latency and significantly improves the success rate of sudden maneuver avoidance. 3. Safety and zero damage: Hybrid force control strategy and composite clamping system achieve zero skin damage record during multiple operations; 4. Closed-loop self-optimization: The federated learning framework supports multi-device collaboration, shortens the knowledge base iteration cycle, and improves makeup satisfaction; 5. Efficiency breakthrough: Automated execution reduces the time required for a single makeup application and improves the efficiency of care for special groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the artificial intelligence-assisted makeup and makeup robot collaborative makeup method of the present invention. DETAILED DESCRIPTION

[0016] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0017] A method for collaborative makeup application between artificial intelligence-assisted makeup and a makeup robot, such as Figure 1 As shown, the following steps are included: S1. Collect user facial data through multimodal sensors, extract multi-scale facial features based on the improved DenseNet deep convolutional neural network, and generate a 3D facial reconstruction model; This system adopts a layered and modular artificial intelligence architecture, which includes three core layers: perception layer, cognitive layer and interaction layer. It realizes makeup assistance function through multimodal data fusion and deep learning technology. It uses an improved DenseNet deep convolutional neural network to achieve high-precision facial feature extraction and analysis. The input layer receives images, and the multi-stage feature extraction network includes multiple convolutional layers and densely connected blocks. The feature reuse mechanism is implemented in each densely connected block to improve feature expression capabilities. The feature pyramid network (FPN) integrates multi-scale features and retains semantic information at different levels. The attention enhancement module adaptively adjusts the feature channel weights.

[0018] In this embodiment, the specific method of facial feature extraction is: S11, uses multi-stage densely connected blocks to extract facial features and integrates multi-scale semantic information through feature pyramid networks; In this embodiment, facial features include: 1) Facial structure parameters: eye shape classification (monolid / double / double), nose shape index, lip contour parameters, and facial geometric proportions. 2) Skin analysis parameters: T-zone oiliness index, pore density distribution, pigmentation area identification, and skin tone evenness score. 3) 3D contour parameters: facial concavity and convexity depth map, bone structure feature points, and light and shadow distribution model.

[0019] S12. Locate facial key points based on the high-resolution representation network and generate standard feature point coordinates and confidence evaluation results; In this embodiment, a parallel multi-resolution branch architecture is implemented based on a high-resolution representation network (HRNet) to implement precise facial feature point positioning technology, maintain high-resolution representation, and implement an inter-resolution feature fusion module to achieve cross-scale information exchange, output standard facial feature points, and a feature point confidence assessment network to automatically identify occluded or uncertain areas.

[0020] S13. Use the pixel-level skin analysis network to output skin quality classification results, including oil distribution, pore density, and pigmentation area heat map; A pixel-level segmentation and analysis network for skin characteristics, a convolutional network architecture dedicated to skin texture analysis, outputs multiple skin condition classifications, such as normal, dry, oily, pigmentation, redness, enlarged pores, fine lines, etc., and uses a pigment distribution mapper to generate a skin tone uniformity heat map.

[0021] S14. Generate a parametric facial model based on single-image 3D reconstruction technology to preview makeup effects from multiple angles.

[0022] S2. Combine user needs and scenario information, use conditional generative adversarial networks to generate personalized makeup plans, and display the target effect and real-time comparison through a multimodal interactive interface; In this embodiment, the specific method of generating a makeup plan is as follows: S21. Analyze makeup needs by fusing user voice, text, and gesture inputs through a cross-modal Transformer. In this embodiment, the cross-modal representation learning module includes: 1) Speech encoder: Based on the Conformer architecture, it extracts speech features.

[0023] 2) Text Encoder: Based on pre-trained language architecture, such as BERT, it extracts text semantic features.

[0024] 3) Image encoder: Based on the Vision Transformer architecture, it extracts visual features.

[0025] 4) Posture encoder: Extracts motion features based on spatiotemporal graph convolutional networks.

[0026] 5) Shared multimodal representation space: multidimensional vector space.

[0027] 6) Contrastive learning framework: NCE loss is used to bring related modal representations closer and push irrelevant modalities further away.

[0028] S22, Generate high-fidelity virtual makeup effects based on the improved StyleGAN architecture and achieve style transfer through adaptive instance normalization; In this embodiment, the improved StyleGAN architecture mainly includes: 1) Generator Network: Mapping network: multi-layer MLP, potential code is converted into the intermediate latent space W; Synthesis network: progressive synthesis, supporting high-resolution output, such as 1024×1024; Normalization layer,controls makeup features through style code.

[0029] 2) Discriminator network: Multi-scale discriminator architecture to simultaneously evaluate global and local realism; Identity preservation loss function ensures that facial identity features remain unchanged; Style consistency assessment module to ensure that the makeup style meets expectations; 3) Training Optimizer: WGAN-GP training strategy based on gradient penalty term; Differential privacy protection mechanism to prevent identity information leakage; S23, constructing a regionalized expert network set to generate specialized makeup step sequences for the eyes, lips, and contour areas respectively; Multiple regional expert networks are used to target different parts of the face, such as eyes, eyebrows, lips, cheeks, and contours. Each expert network adopts a U-Net architecture with skip connections, and shares an encoder and a regional-specific decoder. At the same time, it is equipped with a local attention mechanism to enhance the detail representation within the region and uses the regional boundary fusion module to achieve a seamless transition effect.

[0030] S24. Decompose the makeup plan into structured parameterized operation instructions, including tool selection, action trajectory and force parameters.

[0031] In this embodiment, the makeup step decision is: 1) Decision tree mapping: trigger corresponding rule branches based on facial features.

[0032] 2) Regional Prioritization: Determine the focus based on the degree of defect in each facial area.

[0033] 3) Process dependency graph construction: Create a DAG (directed acyclic graph) to ensure the logical order of makeup steps Makeup plan: 1) Build a structured, parameterized makeup plan data model that supports multimodal dynamic generation and adjustment. Each step contains 16-24 adjustable parameters, allowing for fine-grained customization. Application paths are parameterized using Bezier curves to ensure smooth transitions.

[0034] 2) Establish a multi-level makeup template library and implement intelligent template selection based on multiple factors.

[0035] S3, based on a real-time feedback mechanism, quantifies the difference between the current makeup effect and the target effect through multi-dimensional feature point mapping, color space analysis and texture comparison, and generates dynamic correction suggestions; The makeup effect is mapped to the computer screen in real time and compared with the baseline effect. The scores of various parts of the face, such as facial features, are quantified, and modification suggestions are given in real time. The matching degree between the makeup effect of each facial area and the target effect is accurately quantified through computer vision technology. First, the facial area is finely segmented into multiple key areas, and the regional weights can be dynamically adjusted according to the occasion and style. Then, the shape accuracy, color reproduction and texture expression of different areas are calculated, and the score of the area is calculated according to different weights. Finally, the comprehensive score is calculated based on the regional weights, which includes the following steps: S31. Calculate the feature point position deviation using the weighted Euclidean distance. The formula is:

[0036] Where, and are the corresponding feature points of the current effect and the target effect respectively, The importance weight of feature points is dynamically adjusted through the attention mechanism; Global alignment using Procrustes analysis eliminates pose differences, and weighted Euclidean distance is used to calculate the positional deviation of corresponding feature points. Feature points are grouped into makeup-related regions such as eyebrows, eyes, lips, and contours, and a shape similarity score is output for each region.

[0037] S32, calculating the regionalized histogram similarity in the multi-color space and quantifying the color difference by EMD; In this embodiment, the system performs precise comparisons across multiple color spaces to quantify the color differences between the current makeup effect and the target effect. Based on the regionalized color comparison module, regionalized comparison analysis is performed in multiple color spaces, such as RGB, HSV, Lab, and YCbCr. After region segmentation, local histogram comparisons are performed to calculate the EMD metric distribution similarity. The specific calculation method is:

[0038] Where, are the color histograms of the rth region of the current effect and the target effect, is the regional importance weight.

[0039] Based on the above similarity calculation, we can get the hue, saturation, brightness and three-dimensional independent scores. Taking the color coordination between regions into consideration, we can evaluate the overall color balance.

[0040] S33, extracting texture features based on Gabor filter and gray level co-occurrence matrix, and calculating the matching degree between the current texture and the target texture; In this embodiment, multi-scale and multi-directional texture analysis based on Gabor filter banks is performed, and local binary patterns are used for feature extraction. The gray-level co-occurrence matrix is used to analyze texture uniformity and roughness, and the texture similarity score is calculated. The specific calculation method is:

[0041] in is the distance metric of the m-th texture feature, and They are the texture feature directions of the current effect and the target effect respectively.

[0042] S34. Generate a comprehensive proximity score through the hierarchical scoring aggregation module. The formula is:

[0043] Where, is the comprehensive proximity score, For the Ratings for each dimension, is the nonlinear adjustment factor, is the importance weight of the feature point, is the total number of dimensions.

[0044] S4. Execute makeup operations using a multi-degree-of-freedom robotic arm, combining force feedback control with visual monitoring to achieve safe and accurate cosmetic application. The makeup process is continuously optimized based on user feedback. The robot performs the following operations: S41, using redundant degrees of freedom robotic arms to plan the motion trajectory of makeup tools, and realizing path optimization based on the rapid exploration random tree algorithm; In this embodiment, a multi-axis redundant robotic arm structure is used to achieve precise execution of human-like makeup movements, such as: 1) Base rotation axis (first axis): ±175° rotation range, 0.005° positioning accuracy.

[0045] 2) Shoulder pitch axis (second axis): ±110° range of motion, load capacity 15kg.

[0046] 3) Elbow rotation axis (third axis): ±170° rotation range, repeatability accuracy ±0.02mm.

[0047] 4) Forearm rotation axis (fourth axis): ±120° rotation range, maximum angular velocity 180° / s.

[0048] 5) Wrist pitch axis (fifth axis): ±110° range of motion, dynamic tracking error <0.1°.

[0049] 6) Wrist rotation axis (sixth axis): ±170° rotation range, response time <15ms.

[0050] 7) End effector rotation axis (seventh axis): 360° continuous rotation capability, torque control accuracy 0.01Nm.

[0051] Considering multiple optimization objectives such as efficiency, stability, and safety, this embodiment uses rapid exploration of random numbers for path planning, taking into account the time-optimal trajectory under dynamic constraints. This solution can detect and avoid obstacles, and provide visual feedback for trajectory fine-tuning.

[0052] S42, using a six-dimensional force sensor and a hybrid position / force control strategy to ensure that the contact force is below a preset safety threshold; In this embodiment, a safe contact system based on force sensing and control prevents damage to human skin. The design incorporates a six-dimensional force / torque sensor network, with multi-axis sensors integrated into the end effector and torque sensors built into each joint to shorten response time. Furthermore, a hybrid position / force control algorithm is employed to maintain both position accuracy and contact force safety. Dynamic force threshold adjustment allows for different contact force limits to be set based on the distribution area, and force gradient-based anomaly detection identifies sudden contact force anomalies.

[0053] S43, integrated electromagnetic and pneumatic composite clamping system, adapts to different tools such as eyebrow pencils and powder puffs, and realizes automatic tool switching based on RFID and visual recognition; The clamping system is designed according to the characteristics of different makeup tools, including: 1) Electromagnetic adsorption device: Controllable permanent magnet system; Optimized design of magnetic force distribution ensures tool positioning accuracy; Rapid demagnetization circuit to achieve rapid release of tools; 2) Pneumatic clamping mechanism: Two-finger pneumatic gripper; Force closed-loop control system to prevent over-clamping; Conformable clamping pad increases contact area and reduces local pressure; 3) Multi-mode clamping strategy: Small tools (such as eyebrow pencils): pneumatic precision clamping; Metal tools (such as eyelash curlers): use electromagnetic adsorption; Large tools (such as powder puffs): adopt pneumatic and electromagnetic composite clamping.

[0054] This embodiment introduces multimodal tool recognition technology to automatically identify makeup tools, including the introduction of an RFID chip for each makeup tool, providing a unique identification code. Computer vision is used to assist in determining the tool's status, and a weight sensor is used for verification to prevent recognition errors.

[0055] S44. Monitor the user's head movements and facial expressions in real time, and avoid sudden movements through the dynamic trajectory adjustment module.

[0056] This embodiment uses a multi-sensor fusion perception network, including a high-definition RGB camera, a depth camera, and a thermal imaging sensor. This perception network can provide facial images with depth resolution, while using the thermal imaging sensor to detect abnormal skin temperature.

[0057] S5. Optimize and modify strategies based on the reinforcement learning framework, update the knowledge base based on user feedback, and form a closed-loop iterative optimization system.

[0058] In this embodiment, the specific method of closed-loop optimization is: S51, capturing user responses through facial expression analysis and operation duration recording, and generating reward signals; In this embodiment, facial capture is used to analyze the user's response to correction suggestions. Facial expression analysis can identify emotions such as confusion or satisfaction. By monitoring the execution time of the action, user hesitation can be detected. The results of the correction can also be used to evaluate the effectiveness of the suggestions and track the success rate of the correction.

[0059] S52. Use incremental learning and knowledge distillation technology to update the makeup expert system to avoid catastrophic forgetting; S53. Deploy the optimized strategy to the cloud server to achieve multi-device collaborative learning and template sharing.

[0060] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0061] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0063] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0064] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for collaborative makeup application using artificial intelligence and a makeup robot, characterized in that: The steps include: S1. Collect user facial data through multimodal sensors, extract multi-scale facial features based on the improved DenseNet deep convolutional neural network, and generate a 3D facial reconstruction model; S2. Combine user needs and scenario information, use conditional generative adversarial networks to generate personalized makeup plans, and display the target effect and real-time comparison through a multimodal interactive interface; S3, based on a real-time feedback mechanism, quantifies the difference between the current makeup effect and the target effect through multi-dimensional feature point mapping, color space analysis and texture comparison, and generates dynamic correction suggestions; S4. Perform makeup operations through a multi-degree-of-freedom robotic arm, combine force feedback control with visual monitoring to achieve safe and accurate cosmetic application, and continuously optimize the makeup process based on user feedback.

2. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 1, characterized in that: The specific method of facial feature extraction in S1 is: S11, uses multi-stage densely connected blocks to extract facial features and integrates multi-scale semantic information through feature pyramid networks; S12. Locate facial key points based on the high-resolution representation network and generate standard feature point coordinates and confidence evaluation results; S13. Use the pixel-level skin analysis network to output skin quality classification results, including oil distribution, pore density, and pigmentation area heat map; S14. Generate a parametric facial model based on single-image 3D reconstruction technology to preview makeup effects from multiple angles.

3. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 1, characterized in that: The specific method of generating the makeup plan in S2 is: S21. Analyze makeup needs by fusing user voice, text, and gesture inputs through a cross-modal Transformer. S22, Generate high-fidelity virtual makeup effects based on the improved StyleGAN architecture and achieve style transfer through adaptive instance normalization; S23, constructing a regionalized expert network set to generate specialized makeup step sequences for the eyes, lips, and contour areas respectively; S24. Decompose the makeup plan into structured parameterized operation instructions, including tool selection, action trajectory and force parameters.

4. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 1, characterized in that: The specific method of the real-time feedback mechanism in S3 is: S31. Calculate the feature point position deviation using the weighted Euclidean distance. The formula is: Where, and are the corresponding feature points of the current effect and the target effect respectively, The importance weight of feature points is dynamically adjusted through the attention mechanism; S32, calculating the regionalized histogram similarity in the multi-color space and quantifying the color difference by EMD; S33, extracting texture features based on Gabor filter and gray level co-occurrence matrix, and calculating the matching degree between the current texture and the target texture; S34. Generate a comprehensive proximity score through the hierarchical scoring aggregation module. The formula is: Where, is the comprehensive proximity score, For the Ratings for each dimension, is the nonlinear adjustment factor, is the importance weight of the feature point, is the total number of dimensions.

5. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 1, characterized in that: The specific method of execution of the robot in S4 is: S41, using redundant degrees of freedom robotic arms to plan the motion trajectory of makeup tools, and realizing path optimization based on the rapid exploration random tree algorithm; S42, using a six-dimensional force sensor and a hybrid position / force control strategy to ensure that the contact force is below a preset safety threshold; S43, integrated electromagnetic and pneumatic composite clamping system, adaptable to different tools, and realize automatic tool switching based on RFID and visual recognition; S44. Monitor the user's head movements and facial expressions in real time, and avoid sudden movements through the dynamic trajectory adjustment module.

6. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 1, characterized in that: Also includes: S5. Optimize and modify strategies based on the reinforcement learning framework, update the knowledge base based on user feedback, and form a closed-loop iterative optimization system.

7. The method of artificial intelligence-assisted makeup and makeup robot collaborative makeup according to claim 6, characterized in that: The specific method of closed-loop optimization in S5 is: S51, capturing user responses through facial expression analysis and operation duration recording, and generating reward signals; S52. Use incremental learning and knowledge distillation technology to update the makeup expert system to avoid catastrophic forgetting; S53. Deploy the optimized strategy to the cloud server to achieve multi-device collaborative learning and template sharing.

Images