An AI-based method for emotion recognition and analysis of facial expression data

Abstract

This invention discloses an emotion recognition and analysis method based on artificial intelligence facial expression data, belonging to the field of image recognition technology. The method involves acquiring the facial contour of a target person and constructing a facial model. Multiple traction points are selected corresponding to the facial model, and these points are combined to obtain a combined point cloud of multiple corresponding emotions. A connection relationship is established between the combined point cloud of multiple corresponding emotions and the facial model. Video image data of facial expressions of the target person within a preset time period is acquired, and this video image data is then overlaid onto the facial model to obtain an overlaid model. Based on the facial model in the overlaid model, the combined point cloud of the corresponding emotions is obtained through the connection relationship. This invention can concretize and clarify emotion recognition data, and can filter out emotion types that do not correspond to the target person during the dynamic analysis of video image data of facial expressions of the target person, ensuring the accuracy of emotion type recognition.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and more specifically to an emotion recognition and analysis method based on artificial intelligence facial expression data. Background Technology

[0002] With continuous technological advancements and growing application demands, the application scenarios for emotion recognition and analysis methods based on facial expression data are constantly expanding. Besides human-computer interaction, social media analysis, and mental health, this technology can also be applied to entertainment, security monitoring, and many other fields. Currently, due to the complexity and diversity of human emotions, as well as the ambiguity and uncertainty of facial expression data, the accuracy of emotion recognition is affected, making clear identification of facial expression data difficult. Summary of the Invention

[0003] The purpose of this invention is to provide an emotion recognition and analysis method based on artificial intelligence facial expression data to address the shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for emotion recognition and analysis based on artificial intelligence facial expression data, comprising the following steps:

[0005] Collect the facial contours of the target person and construct a facial model. Select multiple traction points for the facial model and combine them according to the emotions corresponding to the traction points to obtain a combined point cloud of multiple corresponding emotions.

[0006] Establish a connection between multiple point clouds corresponding to different emotions and facial models;

[0007] Collect facial expression video images of the target person within a preset time period, and then fit the facial expression video images data with the facial model to obtain the fitted model;

[0008] Based on the facial model in the fitting model, the corresponding emotional combination point cloud is obtained through the connection relationship, and the emotional type of the target person is determined.

[0009] In a preferred embodiment, the steps of acquiring the facial contour of the target person and constructing a facial model, selecting multiple traction points corresponding to the facial model, and combining the traction points according to the emotions corresponding to the emotions to obtain a combined point cloud of multiple corresponding emotions include:

[0010] A facial model is constructed based on the facial contour images of the target person obtained from multiple angles.

[0011] Based on facial muscle composition information, multiple traction points are selected on the facial model, where the traction points are feature points of the corresponding facial emotional muscles.

[0012] Multiple trigger points corresponding to different emotion types are combined to obtain multiple combined point clouds.

[0013] In a preferred embodiment, the step of establishing the connection between multiple corresponding emotion combination point clouds and facial models includes:

[0014] Multiple emotional facial models are obtained by copying the facial model. The number of emotional facial models is the same as the number of combined point clouds. The combined point clouds are then bound to the emotional facial models in a one-to-one correspondence.

[0015] Based on the emotional facial model, motion trajectory information of multiple corresponding emotion combination point clouds is established. The motion trajectory information includes the movement trajectory of multiple moving points in the combination point cloud according to the emotion type and the positional relationship between the movement trajectories of multiple moving points in the combination point cloud.

[0016] Multiple facial emotion point sets are obtained by combining multiple combined point clouds with corresponding motion trajectory information on the emotional facial model;

[0017] Establish a connection between multiple facial emotion point clusters and the facial model.

[0018] In a preferred embodiment, the step of establishing motion trajectory information of multiple corresponding emotion combination point clouds based on the emotion facial model includes:

[0019] On the emotional facial model, the corresponding combined point cloud is used to determine the traction position information of multiple traction points in the combined point cloud according to the emotion type. The traction position information includes the final traction position of multiple traction points and the positional relationship between multiple traction points when traction reaches the final traction position.

[0020] The path from the original position to the final position of multiple traction points in the combined point cloud is determined as the movement trajectory of the multiple traction points;

[0021] The motion trajectory information is based on the movement trajectory of multiple points in the combined point cloud of emotion type and the positional relationship between the movement trajectories of multiple points in the combined point cloud.

[0022] In a preferred embodiment, the step of establishing the connection between multiple facial emotion point clouds and the facial model includes:

[0023] Configure corresponding ports for each of the multiple traction points in the facial model;

[0024] For each of the multiple traction points in a cloud of facial emotion points, a corresponding sub-port is configured, and a positional attachment relationship is established between the sub-port and the movement trajectory of the corresponding traction point.

[0025] By linearly connecting the sub-ports with established attachment relationships to the ports of the corresponding traction points in the facial model, the connection relationship between multiple facial emotion point clusters and the facial model is obtained.

[0026] In a preferred embodiment, the step of acquiring facial expression video image data of the target person within a preset time period, and then fitting the facial expression video image data to a facial model to obtain a fitted model includes:

[0027] Collect facial expression video image data of the target person within a preset time period. The facial expression video image data includes the facial contour and the traction points in the facial contour.

[0028] The facial contours in the facial expression video image data are superimposed on the facial model. The traction points in the facial contours in the facial expression video image data are adjusted to match the corresponding traction points in the facial model to obtain the fitting model.

[0029] In a preferred embodiment, the step of determining the emotion type of the target person by obtaining a combined point cloud of corresponding emotions based on the facial model in the fitting model through the connection relationship includes:

[0030] Bind the traction points in the facial contour of the fitted facial expression video image data to the corresponding traction points in the facial model;

[0031] The traction points in the facial model move along with the traction points in the facial contour of the bound facial emotional expression video image data, and the movement speed of the traction points in the facial contour of the facial emotional expression video image data is obtained.

[0032] According to the movement speed, multiple traction points in multiple facial emotion point clouds are dynamically moved. Based on the connection relationship, multiple traction points in multiple dynamically moving facial emotion point clouds are connected to traction points in the facial model, respectively, to obtain the recognition index of the facial model corresponding to multiple facial emotion point clouds.

[0033] A combined point cloud is determined for the facial emotion point cloud set that meets the preset conditions, and the emotion type corresponding to the target person is obtained based on the combined point cloud.

[0034] In a preferred embodiment, the step of obtaining the recognition index corresponding to the facial model and the set of multiple facial emotion points includes:

[0035] Obtain the length and angle of the connection between multiple traction points in the dynamically moving facial emotion point cloud and the traction point line in the facial model.

[0036] The recognition index of multiple facial emotion point clusters corresponding to the facial model is obtained. The formula for calculating the recognition index includes:

[0037] Where θ is the recognition index, LY n LD is the original length of the line connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n Let μY be the current length of the line connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n Let μD be the original angle connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n α is the current angle connecting the nth traction point in a single facial emotion point cloud to the traction point line in the corresponding facial model, where n is the nth traction point among k traction points in a single facial emotion point cloud, and α and β are constants greater than zero.

[0038] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0039] This invention connects multiple traction points in a dynamically moving set of facial emotion points to traction points in a facial model using lines. It also collects the lengths and angles of these line connections, calculating a recognition index in real time. This process concretizes and clarifies emotion recognition data. Then, it determines a combined point cloud of facial emotion points that meets preset recognition index conditions. Based on this combined point cloud, it obtains the emotion type corresponding to the target person. This allows for the dynamic analysis of facial expression video image data of the target person, enabling the filtering out of emotion types that do not correspond to the target person. This efficient approach ensures accurate emotion type recognition. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0041] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example 1, please refer to Figure 1 As shown in this embodiment, an emotion recognition and analysis method based on artificial intelligence facial expression data includes the following steps:

[0044] S1. Collect the facial contour of the target person and construct a facial model. Select multiple traction points corresponding to the facial model and combine them according to the emotions corresponding to the traction points to obtain a combination point cloud of multiple corresponding emotions.

[0045] S2. Establish the connection between multiple point clouds corresponding to different emotions and the facial model;

[0046] S3. Collect facial expression video image data of the target person within a preset time period, and fit the facial expression video image data with the facial model to obtain the fitted model.

[0047] S4. Based on the facial model in the fitting model, obtain the combined point cloud of the corresponding emotion through the connection relationship to determine the emotion type of the target person;

[0048] As described in steps S1-S4 above, with the continuous advancement of technology and the increasing demand for applications, the application scenarios of emotion recognition and analysis methods based on facial expression data are constantly expanding. Besides human-computer interaction, social media analysis, and mental health, this technology can also be applied to entertainment, security monitoring, and many other fields. Currently, due to the complexity and diversity of human emotions, as well as the ambiguity and uncertainty of facial expression data, the accuracy of emotion recognition is affected, making clear identification of facial expression data difficult. However, in this application, multiple traction points in a dynamically moving facial emotion point cloud are connected to traction points in a facial model by lines. The length and angle of the line connections between these traction points and the traction points in the facial model are collected in real time, and the recognition index is calculated. This process concretizes and clarifies the emotion recognition data. Then, a combined point cloud of facial emotion point clouds corresponding to the recognition index that meets preset conditions is determined. Based on the combined point cloud, the emotion type corresponding to the target person is obtained. This allows for the filtering out of emotion types that do not correspond to the target person during the dynamic analysis of facial expression video image data, enabling efficient identification and ensuring the accuracy of emotion type recognition.

[0049] In one embodiment, step S1, which involves acquiring the facial contour of the target person and constructing a facial model, selecting multiple traction points corresponding to the facial model, and combining the traction points according to the emotions they correspond to to obtain a combined point cloud of multiple corresponding emotions, includes:

[0050] S11. Obtain facial contour images of the target person from multiple angles, and construct a facial model based on the facial contour images of the target person.

[0051] S12. Select multiple traction points on the facial model based on facial muscle composition information, where the traction points are feature points of the corresponding facial emotional muscles.

[0052] S13. Combine multiple trigger points corresponding to different emotion types to obtain multiple combined point clouds;

[0053] As described in steps S11-S13 above, before performing emotion recognition on the target person, it is necessary to collect facial contour information of the target person. The collection method is as follows: use a high-quality camera to capture facial images, ensuring that the images are clear, without blur or noise; capture facial contour images of the target person from multiple angles; convert color images to grayscale images; use filters to remove noise from the images to improve image quality; use contrast enhancement, sharpening and other techniques to make the facial contours clearer; use computer vision technology, such as the shape predictor in the dlib library, to detect facial key points, such as eyes, nose, and mouth; match the detected feature points with feature points in a predefined facial model to provide a foundation for subsequent 3D reconstruction; use a set of 2D facial images and corresponding feature point information to construct a 3D facial model using a 3D reconstruction algorithm (such as 3DMM); then, in order to understand the target person's emotions, it is necessary to analyze the muscles on their face that affect emotions to obtain multiple points of muscle movement. The traction points are feature points corresponding to facial muscles. These feature points are the points where the muscles move the facial contour when emotions are expressed. For example, when smiling happily, the highest point of the muscles around the cheekbone and the muscles around the mouth change. These muscle points are marked as feature points and then used as traction points. This operation relies on understanding the composition of the human face. Therefore, multiple traction points are selected on the facial model based on the composition of facial muscles. Then, these multiple traction points are combined according to the emotion type to obtain multiple combined point clouds. For example, the muscles involved in a happy smile mainly include the risorius, cheek, orbicularis oris, and zygomaticus major muscles. Therefore, it is necessary to combine the traction points corresponding to the risorius, cheek, orbicularis oris, and zygomaticus major muscles for the corresponding emotion type to obtain a combined point cloud. Multiple emotion types have corresponding muscle traction points, so they can be combined separately according to the emotion type, which facilitates the preparation work for emotion recognition of the target person.

[0054] In one embodiment, step S2, which establishes the connection between multiple corresponding emotion combination point clouds and facial models, includes:

[0055] S21. Copy the facial model to obtain multiple emotional facial models. The number of emotional facial models is the same as the number of combined point clouds. Bind the combined point clouds to the emotional facial models one-to-one.

[0056] S22. Based on the emotional facial model, establish motion trajectory information for multiple corresponding emotion combination point clouds. The motion trajectory information includes the movement trajectory of multiple moving points in the combination point cloud according to the emotion type and the positional relationship between the movement trajectories of multiple moving points in the combination point cloud.

[0057] S23. On the emotional facial model, multiple combined point clouds are combined with corresponding motion trajectory information to construct multiple facial emotion point cloud sets;

[0058] S24. Establish the connection between multiple facial emotion point clusters and the facial model.

[0059] As described in steps S21-S23 above, since each emotion type needs to be displayed separately, multiple emotion face models need to be copied to obtain multiple emotion face models. The number of emotion face models is the same as the number of combined point clouds. The combined point clouds are bound one-to-one with the emotion face models. The emotion face models can display the combined point clouds of the corresponding emotion type, which facilitates subsequent emotion recognition and confirmation. Motion trajectory information of multiple corresponding emotion combined point clouds is established. This motion trajectory information includes the movement trajectory of the moving points in the combined point cloud according to the corresponding emotion type and the positional relationship between the movement trajectories of multiple moving points in the combined point cloud. This allows us to obtain the positional relationship of multiple combined point clouds in the combined point cloud under the corresponding emotion type. By analyzing the motion trajectory of moving points and the positional relationship between the movement trajectories of multiple moving points in the combined point cloud, we can understand the changes in facial muscles of the target person under different emotional types, which facilitates the subsequent identification of the target person's emotions. Then, we construct information by linking multiple combined point clouds with their corresponding motion trajectory information. This construction involves building a three-dimensional trajectory of the combined point cloud on the facial model according to the motion trajectory information, resulting in multiple facial emotion point clouds. After that, we establish the connection relationship between multiple facial emotion point clouds and the facial model, which facilitates the correspondence between the subsequently collected facial expression video image data and multiple facial emotion point clouds, enabling better subsequent expression matching and recognition, and improving the efficiency and accuracy of expression recognition.

[0060] In one embodiment, step S22, which establishes motion trajectory information of multiple corresponding emotion combination point clouds based on the emotion facial model, includes:

[0061] S221. On the emotional face model, determine the traction position information of multiple traction points of the corresponding combined point cloud according to the emotion type. The traction position information includes the final traction position of multiple traction points and the positional relationship between multiple traction points when traction reaches the final traction position.

[0062] S222. Determine the path from the original position to the final position of multiple traction points in the combined point cloud as the movement trajectory of the multiple traction points;

[0063] S223. The motion trajectory information is based on the movement trajectory of multiple points in the combined point cloud of emotion type and the positional relationship between the movement trajectories of multiple points in the combined point cloud.

[0064] As described in steps S221-S223 above, the position of the traction points changes when the target person has facial expressions. Therefore, based on the combined point cloud, the final traction position of multiple traction points in the combined point cloud is determined according to the emotion type, as well as the positional relationship between the multiple traction points at the final traction position. After determining the final traction position of the traction points in the combined point cloud according to the emotion type, it can be displayed on the corresponding emotional facial model. Then, according to the traction position information and the original positions of multiple traction points on the corresponding emotional facial model, the paths for moving the original positions of multiple traction points to the final traction positions are determined, and these paths are... As the movement trajectory of multiple traction points, the movement trajectory of the traction points in the combined point cloud according to the emotion type, as well as the positional relationship between the movement trajectories of multiple traction points in the combined point cloud, are used as motion trajectory information. This facilitates the subsequent dynamic construction on the corresponding emotional facial model using motion trajectory information. This results in multiple traction points moving dynamically from their original positions according to the motion trajectory information, thereby obtaining the subsequent facial emotion point cloud. Multiple combined point clouds can be used to construct multiple facial emotion point clouds, which can separately construct multiple emotion types, facilitating the subsequent connection between the emotions of the actual target person and the rapid emotion recognition.

[0065] In one embodiment, step S24, which establishes the connection between multiple facial emotion point clouds and the facial model, includes:

[0066] S241. Configure the corresponding ports for the multiple traction points in the facial model;

[0067] S242. For each of the multiple traction points corresponding to the multiple facial emotion point clouds, a corresponding sub-port is configured, and the position attachment relationship between the sub-port and the movement trajectory of the corresponding traction point is established.

[0068] S243. Linearly connect the sub-ports with the established attachment relationships to the ports of the corresponding traction points in the facial model to obtain the connection relationship between multiple facial emotion point clusters and the facial model.

[0069] As described in steps S241-S243 above, the facial model is a model used to display the actual facial features of the target person. Since the facial model is constructed according to the facial contours of the target person, it can correspond to the facial traction points subsequently collected from the target person. Here, multiple traction points in the facial model are configured with corresponding ports. These ports are used to connect and traction with the corresponding traction points in the subsequent facial emotion point cloud. Then, multiple traction points in the multiple facial emotion point clouds are configured with corresponding sub-ports. Since the facial emotion point cloud is a dynamic model displaying the corresponding emotion type, and the sub-ports are configured for the traction points in the facial emotion point cloud, the sub-ports are also dynamic. This is achieved by attaching sub-ports to the movement trajectory and moving them according to the movement of the corresponding traction points. The sub-ports are then linearly connected to their corresponding ports on the facial model. This linear connection represents the connection length between the sub-ports and the ports themselves. These ports also move according to the corresponding traction points in the facial model. Currently, before collecting actual video image data of the target person's face, these ports are fixed in place, following the corresponding traction points in the facial model. Since multiple combined point clouds contain repeated traction points, the traction points in the facial model may connect to multiple sub-ports. Therefore, a port can connect to multiple sub-ports simultaneously, enabling more efficient and accurate recognition of the target person's facial emotions.

[0070] In one embodiment, step S3, which involves collecting facial expression video image data of the target person within a preset time period and then fitting the facial expression video image data to a facial model to obtain a fitted model, includes:

[0071] S31. Collect facial expression video image data of the target person within a preset time period, wherein the facial expression video image data includes the facial contour and the traction points in the facial contour.

[0072] S32. Align the facial contour in the facial expression video image data with the facial model, adjust the traction points in the facial contour in the facial expression video image data to match the corresponding traction points in the facial model, and obtain the fitted model.

[0073] As described in steps S31 and S32 above, facial expression video image data is collected within a preset time period. This preset time period is not a fixed period, but rather a point in time when the target person displays different emotional expressions. For example, if the current expression is a happy smile, the collection begins when the happy smile ends and a peaceful expression appears. The collection of facial expression video image data within the current preset time period is completed when the peaceful expression ends and the happy smile begins again. This process continues until the start of a happy smile is reached, and the collection of facial expression video image data within the next preset time period is completed after the happy smile ends. This process is repeated continuously, constantly collecting facial expression video image data of the target person. The point where different emotional expressions end is used as a data acquisition node. Facial emotional expression video image data includes the facial contour and the points of movement within the facial contour. Because there is usually a calming process in facial expressions during the expression transition, such as from a smiling expression to a calm expression and then back to a smiling expression, the starting point of the facial emotional expression video image data within the preset time period is usually calm. There are some errors, so the points of movement in the facial contour of the facial emotional expression video image data need to be adjusted until the points of movement in the facial contour of the facial emotional expression video image data match (overlap) with the points of movement in the corresponding facial model. This allows for the preparation of identifying the target person's emotions in the facial emotional expression video image data within the preset time period.

[0074] In one embodiment, step S4, which involves obtaining a combined point cloud of corresponding emotions based on the facial model in the fitting model through a connection relationship, and determining the emotion type of the target person, includes:

[0075] S41. Bind the traction points in the facial contour of the fitted facial expression video image data to the corresponding traction points in the facial model.

[0076] S42. The traction points in the facial model move along with the traction points in the facial contour of the bound facial emotion expression video image data, and the movement speed of the traction points in the facial contour of the facial emotion expression video image data is obtained.

[0077] S43. According to the movement speed, start multiple traction points in multiple facial emotion point clouds to move dynamically. Based on the traction relationship, connect the multiple traction points in the multiple moving facial emotion point clouds to the traction points in the facial model respectively to obtain the recognition index of the facial model corresponding to multiple facial emotion point clouds respectively.

[0078] S44. Determine the combined point cloud of facial emotion point sets corresponding to the recognition index that meets the preset conditions, and obtain the emotion type corresponding to the target person based on the combined point cloud.

[0079] As described in steps S41-S44 above, after the traction points in the facial contour of the facial expression video image data are aligned with the corresponding traction points in the facial model, they are bound together. This allows the traction points in the facial model to move along with the corresponding traction points in the facial contour of the facial expression video image data. The system then notifies the acquisition of the movement speed of the traction points in the facial contour of the facial expression video image data. Based on this movement speed, multiple traction points in the multiple facial emotion point sets are dynamically moved. At the same movement speed, if the length and angle of the lines connecting the dynamically moving traction points in the multiple facial emotion point sets to the traction points in the facial model are larger, it indicates that the corresponding facial emotion point set does not correspond to the actual emotion type of the target person. Based on the connection relationship, multiple traction points in the dynamically moving multiple facial emotion point sets are connected to the traction points in the facial model. This connection is similar to a marionette; the action of pulling the string is the same as the action of the puppet. Here, the line represents the multiple facial emotion point sets... Multiple traction points in the facial emotion point cloud are connected to traction points in the facial model by lines. If the movement is synchronous, the length and angle of the line connections will not change or will only change slightly within a certain range. Here, by limiting the movement by preset conditions, a combined point cloud of facial emotion point sets corresponding to the recognition index that meets the preset conditions is determined within a preset time period. The emotion type corresponding to the target person is obtained based on the combined point cloud. During the process of connecting multiple traction points in the dynamic movement of multiple facial emotion point cloud sets to traction points in the facial model by lines, and collecting the length and angle of the line connections between multiple traction points in multiple facial emotion point cloud sets and traction points in the facial model, the recognition index is calculated in real time. This can concretize and clarify the emotion recognition data. Then, a combined point cloud of facial emotion point sets corresponding to the recognition index that meets the preset conditions is determined. The emotion type corresponding to the target person is obtained based on the combined point cloud. In the process of dynamically analyzing the facial emotion expression video image data of the target person, the emotion type that does not correspond to the target person can be filtered out. This can efficiently identify the emotion type and ensure the accuracy of the emotion type recognition.

[0080] In one embodiment, step S43, which obtains the recognition index corresponding to multiple facial emotion point clouds for the facial model, includes:

[0081] S431. Obtain the length and angle of the connection between the multiple traction points in the dynamic moving multiple facial emotion point cloud and the traction point line in the facial model.

[0082] S432. Obtain the recognition index of multiple facial emotion point clusters corresponding to the facial model. The formula for calculating the recognition index includes:

[0083] Where θ is the recognition index, LY n LD is the original length of the line connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n Let μY be the current length of the line connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n Let μD be the original angle connecting the nth traction point in a single facial emotion point cloud to the corresponding traction point in the facial model. n The current angle connecting the nth traction point in a single facial emotion point cloud to the traction point in the corresponding facial model is defined as follows: n is the nth traction point among k traction points in a single facial emotion point cloud. α and β are constants greater than zero. It should be noted that the larger θ is, the lower the matching degree. The combined point cloud of facial emotion point clouds corresponding to the recognition index that meets the preset conditions is selected, and the emotion type corresponding to the combined point cloud is taken as the emotion type of the target person.

[0084] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for emotion recognition and analysis based on facial expression data using artificial intelligence, characterized in that, Includes the following steps: Collect the facial contours of the target person and construct a facial model. Select multiple traction points for the facial model and combine them according to the emotions corresponding to the traction points to obtain a combined point cloud of multiple corresponding emotions. Establish a connection between multiple point clouds corresponding to different emotions and facial models; Collect facial expression video images of the target person within a preset time period, and then fit the facial expression video images data with the facial model to obtain the fitted model; Based on the facial model in the fitting model, the corresponding emotion combination point cloud is obtained through the connection relationship to determine the emotion type of the target person; The step of establishing the connection between multiple corresponding emotion combination point clouds and facial models includes: replicating the facial model to obtain multiple emotion facial models, wherein the number of emotion facial models is the same as the number of combination point clouds; binding the combination point clouds to the emotion facial models in a one-to-one correspondence; establishing motion trajectory information for multiple corresponding emotion combination point clouds based on the emotion facial models, wherein the motion trajectory information includes the movement trajectory of multiple moving points in the combination point cloud according to the emotion type and the positional relationship between the movement trajectories of multiple moving points in the combination point cloud; constructing multiple facial emotion point cloud sets on the emotion facial models by combining the multiple combination point clouds with the corresponding motion trajectory information; and establishing the connection between the multiple facial emotion point cloud sets and the facial model. The step of establishing motion trajectory information for multiple corresponding emotion combination point clouds based on the emotion facial model includes: determining the traction position information of multiple traction points in the corresponding combination point cloud according to the emotion type on the emotion facial model, wherein the traction position information includes the final traction position of multiple traction points and the positional relationship between multiple traction points at the final traction position; determining the path from the original position to the final traction position of multiple traction points in the combination point cloud as the movement trajectory of multiple traction points; and using the movement trajectory of multiple traction points in the combination point cloud according to the emotion type and the positional relationship of the movement trajectories between multiple traction points in the combination point cloud as motion trajectory information. The step of establishing the connection between multiple facial emotion point sets and the facial model includes: configuring corresponding ports for multiple traction points in the facial model; configuring corresponding sub-ports for multiple traction points in the multiple facial emotion point sets, and establishing a positional attachment relationship between the sub-ports and the movement trajectories of the corresponding traction points; and linearly connecting the sub-ports with the established attachment relationships to the ports of the corresponding traction points in the facial model to obtain the connection between multiple facial emotion point sets and the facial model. The step of collecting facial expression video image data of the target person within a preset time period and then fitting the facial expression video image data to a facial model to obtain a fitted model includes: collecting facial expression video image data of the target person within a preset time period, wherein the facial expression video image data includes the facial contour and the traction points in the facial contour; aligning the facial contour in the facial expression video image data with the facial model, and adjusting the traction points in the facial contour in the facial expression video image data to fit the corresponding traction points in the facial model to obtain the fitted model; The step of determining the emotion type of a target person by obtaining a combined point cloud of corresponding emotions based on the facial model in the fitting model through the connection relationship includes: binding the traction points in the facial contour of the fitted facial emotion expression video image data with the corresponding traction points in the facial model; the traction points in the facial model move following the bound traction points in the facial contour of the facial emotion expression video image data, and obtaining the movement speed of the traction points in the facial contour of the facial emotion expression video image data; starting multiple traction points in multiple facial emotion point cloud sets to move dynamically according to the movement speed, and connecting the multiple traction points in the dynamically moving multiple facial emotion point cloud sets with the traction points in the facial model based on the connection relationship to obtain the recognition index of the facial model corresponding to multiple facial emotion point cloud sets; determining the combined point cloud of the facial emotion point cloud set corresponding to the recognition index that meets the preset conditions, and obtaining the emotion type of the target person based on the combined point cloud.

2. The emotion recognition and analysis method based on artificial intelligence facial expression data according to claim 1, characterized in that: The steps of acquiring the facial contours of the target person and constructing a facial model, selecting multiple traction points corresponding to the facial model, and combining the traction points according to the emotions they correspond to to obtain a combined point cloud of multiple corresponding emotions include: A facial model is constructed based on the facial contour images of the target person obtained from multiple angles. Based on facial muscle composition information, multiple traction points are selected on the facial model, where the traction points are feature points of the corresponding facial emotional muscles. Multiple trigger points corresponding to different emotion types are combined to obtain multiple combined point clouds.

3. The emotion recognition and analysis method based on artificial intelligence facial expression data according to claim 1, characterized in that: The step of obtaining the recognition index corresponding to multiple facial emotion point sets for each facial model includes: Obtain the length and angle of the connection between multiple traction points in the dynamically moving facial emotion point cloud and the traction point line in the facial model. The recognition index of multiple facial emotion point clusters corresponding to the facial model is obtained. The formula for calculating the recognition index includes: ,in, To identify the index, For a single facial emotion point cloud, the first The original length of the line connecting each traction point to the corresponding traction point in the facial model. For a single facial emotion point cloud, the first The current length of the line connecting each traction point to the corresponding traction point in the facial model. For a single facial emotion point cloud, the first The original angle connecting each traction point to the corresponding traction point line in the facial model. For a single facial emotion point cloud, the first The current angle connecting each traction point to the corresponding traction point in the facial model. For individual facial emotion point cloud The first of the key points One key factor, among which and It is a constant greater than zero.

Images