Digital human memory video generation method, system and device, medium and product

Abstract

The invention discloses a digital human memory video generation method, system and device, a medium and a product, and the method comprises the steps: selecting a preset basic digital human model according to the picture data of a target person; according to user behavior data and the picture data, reconstructing the basic digital human model to obtain a digital human static model; screening target video data from the user behavior data, and extracting human body motion track information in the target video data; and adding the human body motion track information to the digital human static model to generate a digital human memory video. By adopting the embodiment of the invention, the unique digital human image memory video can be generated based on the digital asset data of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronics, and in particular to a digital person recall video generation method, system, device, medium and product. BACKGROUND

[0002] Currently, some software automatically extracts metadata such as tags, faces, shooting time and location based on user-uploaded pictures, generates thumbnails of various materials and sorts them according to the time axis, and finally synthesizes a story video with user's personal characteristics.

[0003] However, the content of the traditional recall video is limited, and it only plays pictures in the form of a slideshow, which has a weak visual effect. SUMMARY

[0004] Embodiments of the present application aim to provide a digital person recall video generation method, system, device, medium and product, which can generate a unique digital person recall video based on user's digital asset data.

[0005] In a first aspect, the embodiments of the present application provide a digital person recall video generation method, comprising: selecting a preset basic digital person model according to picture data of a target person; reconstructing the basic digital person model according to user behavior data and the picture data to obtain a digital person static model; screening target video data from the user behavior data, and extracting human motion trajectory information in the target video data; adding the human motion trajectory information to the digital person static model to generate a digital person recall video.

[0006] As an improvement of the above-mentioned scheme, the step of selecting a preset basic digital person model according to picture data of a target person comprises: obtaining all picture materials uploaded by a user, performing target person recognition on the picture materials to obtain target picture materials; performing quality scoring on the target picture materials, and screening effective picture data from the target picture materials according to the quality scoring result; identifying the basic information of a target person according to the effective picture data; the basic information includes gender and age; selecting a preset basic digital person model according to the basic information.

[0007] As an improvement of the above-mentioned scheme, the step of reconstructing the basic digital person model according to user behavior data and the picture data to obtain a digital person static model comprises: According to the user behavior data and the picture data, the face shape style preference and the facial feature style preference of the user are analyzed; According to the face shape style preference, a target face shape is selected from preset face shapes; User posture data is obtained from the user behavior data, and a vertical reference axis where a face center point is located is calculated according to the user posture data and the target face shape; According to the height of the target face shape, a first horizontal axis, a second horizontal axis and a third horizontal axis of the face are calculated; the first horizontal axis is an eye horizontal reference line, the second horizontal axis is a horizontal reference axis where the face center point is located, and the third horizontal axis is a mouth horizontal reference line; According to the vertical reference axis, the first horizontal axis, the second horizontal axis and the third horizontal axis, the facial feature shape and position are obtained based on the facial feature style preference; According to the target face shape and the facial feature shape and position, the face features of the basic digital human model are reconstructed to obtain a digital human static model.

[0008] As an improvement of the above scheme, the face shape style preference and the facial feature style preference of the user are analyzed according to the user behavior data and the picture data, including: According to the user behavior data and the picture data, game data, storage data, sharing data and publishing data are obtained, and the face shape style preference of the user is extracted therefrom; According to the user behavior data and the picture data, emotion description data and use habit data are obtained, and the facial feature style preference of the user is extracted in combination with the face shape style preference.

[0009] As an improvement of the above scheme, the user posture data is obtained from the user behavior data, and the vertical reference axis where the face center point is located is calculated according to the user posture data and the target face shape, including: User horizontal deflection angle, user vertical deflection angle and user eye direction are obtained from the user behavior data to obtain user posture data; According to the user eye direction, a straight line distance between the user's eyes and the device is calculated; According to the target face shape, a face center point is obtained, a vertical axis line is drawn at the face center point to obtain an initial vertical reference axis; According to the target face shape, the user horizontal deflection angle, the user vertical deflection angle and the straight line distance, a deflection distance of the initial vertical reference axis is calculated to obtain a vertical reference axis line.

[0010] As an improvement of the above scheme, the first horizontal axis, the second horizontal axis and the third horizontal axis of the face are calculated according to the height of the target face shape, including: Based on the height of the target face shape, three equidistant horizontal axes are drawn on the face, namely the initial first horizontal axis, the initial second horizontal axis, and the initial third horizontal axis; The initial second horizontal axis is adjusted to the center point of the face to obtain the second horizontal axis; From the user behavior data, user social data is obtained to determine the user's facial feature ratio preferences; Based on the height of the target face shape and the facial feature proportion preference, the distances between the initial first horizontal axis and the initial third horizontal axis and the second horizontal axis are adjusted to obtain the first horizontal axis and the third horizontal axis.

[0011] As an improvement to the above solution, the step of obtaining the shape and position of facial features based on the vertical reference axis, the first horizontal axis, the second horizontal axis, and the third horizontal axis, and based on the facial feature style preference, includes: The eye is located based on the vertical reference axis and the first horizontal axis to obtain the eye position. The nose is located based on the second horizontal axis to obtain its position; The mouth is positioned based on the vertical reference axis and the third horizontal axis to obtain the mouth position; The positions of the facial features are obtained based on the positions of the eyes, nose, and mouth. The facial features are adjusted according to the facial feature style preference and the position of the facial features; the facial feature shape includes the size, angle and shape of the facial features.

[0012] As an improvement to the above solution, the step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on user behavior data and the image data, user height data, clothing browsing data, and clothing order data are obtained, and the torso height of the basic digital human model is calculated. Based on the image data, calculate the torso width of the basic digital human model; Calculate the ratio of the torso height to the torso width, and select a target body shape from the preset body shapes; Based on the target body shape, calculate the limb dimensions of the basic digital human model; Based on the torso height, torso width, and limb dimensions, the body features of the basic digital human model are reconstructed to obtain a static digital human model.

[0013] As an improvement to the above solution, the step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on the image data, color data and light and shadow transformation data are obtained, and the user's clothing style preference and clothing color compatibility are analyzed. Based on the clothing style preference and the clothing color compatibility, the clothing features of the basic digital human model are reconstructed to obtain a static digital human model.

[0014] As an improvement to the above solution, the step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on user behavior data and the image data, extract the user's background preference style; Based on the background preference style, the background features of the basic digital human model are reconstructed to obtain a static digital human model.

[0015] As an improvement to the above solution, the step of filtering target video data from the user behavior data and extracting human motion trajectory information from the target video data includes: Filter audio playback data or music video playback data from the user behavior data, and use the music video corresponding to the audio playback data or the music video playback data as the target video data; A neural motion feature learning algorithm is used to track the motion of the human body in the target video data and obtain the human body motion trajectory information.

[0016] As an improvement to the above solution, the step of adding the human motion trajectory information to the digital human static model to generate a digital human memory video includes: The fusion tool interface is invoked to add the human motion trajectory information to the digital human static model, and the digital human static model is rendered to generate a digital human memory video.

[0017] Secondly, embodiments of the present invention provide a digital human memory video generation system, comprising: The basic model selection module is used to select a preset basic digital human model based on the image data of the target person; The basic model reconstruction module is used to reconstruct the basic digital human model based on user behavior data and the image data to obtain a static digital human model. The trajectory information extraction module is used to filter target video data from the user behavior data and extract human motion trajectory information from the target video data. The memory video generation module is used to add the human motion trajectory information to the digital human static model to generate a digital human memory video.

[0018] Thirdly, embodiments of the present invention provide a digital human memory video generation device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the digital human memory video generation method as described above.

[0019] Fourthly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the digital human memory video generation method as described above.

[0020] Fifthly, embodiments of the present invention provide a computer program product, the computer program product including a computer program or computer instructions, which, when executed by a processor, performs the digital human memory video generation method as described above.

[0021] Compared with existing technologies, the present invention discloses a method, system, device, medium, and product for generating digital human memory videos. This involves selecting a preset basic digital human model based on image data of a target person; reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model; filtering target video data from the user behavior data and extracting human motion trajectory information from the target video data; and adding the human motion trajectory information to the static digital human model to generate a digital human memory video. Using embodiments of the present invention, it is possible to generate unique digital human memory videos based on a user's digital asset data. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the steps of a digital human memory video generation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a preset basic digital human model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the deflection of an initial vertical reference axis provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the distance between horizontal axes provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a digital human memory video generation system provided in an embodiment of the present invention. Detailed Implementation

[0023] 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, and 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.

[0024] In the description and claims, it should be understood that the terms "first," "second," etc., used in the description and claims are only for the purpose of distinguishing the description of the same technical features, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated, nor necessarily the order of description or chronological order. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature.

[0025] Traditional memory video generation solutions mostly cluster identical faces and then play them in chronological order as a slideshow to create a memory video. Such memory videos simply display user images and offer no novelty or fresh experience for the user.

[0026] Based on the above considerations, embodiments of the present invention provide a method for generating digital human memory videos. Please refer to [link to relevant documentation]. Figure 1 In this embodiment, the digital human memory video generation method is specifically executed through steps S1 to S4: S1. Select a preset basic digital human model based on the image data of the target person; S2. Based on user behavior data and the image data, the basic digital human model is reconstructed to obtain a static digital human model; S3. Filter target video data from the user behavior data and extract human motion trajectory information from the target video data; S4. Add the human motion trajectory information to the digital human static model to generate a digital human memory video.

[0027] It should be noted that the target personnel can be user-specified individuals or automatically selected based on their frequency of appearance. The image data of the target personnel is the core basis for selecting the basic digital human model.

[0028] Preferably, several basic digital human models are pre-set, each corresponding to a different gender and age range, possessing the basic morphological structure of a digital human, with specific features obtained through subsequent optimization.

[0029] For example, the user behavior data includes user access data, user social data, user storage data, user usage habit data, etc., that can be obtained during the execution of the embodiments of the present invention. For instance, user behavior data can be used to create user profiles and generate personalized digital human memory videos based on user identity and user interests.

[0030] In this embodiment of the invention, user behavior data and target user image data are used together as the basis for adjusting the basic digital human model. This ensures that the digital human form fits the target person and meets the user's associated needs, thus achieving personalization and targeting of the static form of the digital human.

[0031] It should also be noted that the target video data is video material from user behavior data from which human movement trajectories can be extracted. In this embodiment of the invention, human movement trajectory information is further extracted from the target video data to reflect human movement path and posture change information, providing dynamic movement capabilities for static digital human models.

[0032] The embodiments of this invention can be coupled and applied to various scenarios. For example, in cloud storage products, personalized digital human memory videos can be customized for users based on user-uploaded images, storage habits, search history, and interaction records; similarly, the embodiments of this invention can also be used in social platforms or dedicated video generation platforms.

[0033] In the above solution, the generated digital human memory video can not only match the characteristics of a specific target person, but also carry the memory scenes associated with the user's past behavior, providing the user with a unique digital human image memory video.

[0034] As a preferred implementation, step S1 involves selecting a preset basic digital human model based on the image data of the target person, including: Obtain all image materials uploaded by users, perform target person identification on the image materials, and obtain target image materials; The target image materials are scored for quality, and valid image data is selected from the target image materials based on the quality score results; Based on the valid image data, identify the basic information of the target person; the basic information includes gender and age. Based on the aforementioned basic information, a preset basic digital human model is selected.

[0035] In some preferred embodiments, the image materials include image files directly uploaded by the user and image frames of people appearing in video files uploaded by the user.

[0036] Different image materials may vary in size, compression level, clarity, and completeness. Therefore, in this embodiment of the invention, the target image materials, including the target person, are further quality-scored to select valid image data and avoid recognition errors caused by low-quality images.

[0037] Further, preferably, the step of scoring the target image material for quality and filtering valid image data from the target image material based on the quality score results includes: Based on the target image material, a pixel histogram is generated to obtain the median frequency and median brightness distribution of the histogram. The target image material is subjected to frequency domain transformation to obtain the median of the frequency distribution; The target image material is scored based on the histogram median frequency, the luminance median, and the frequency median. Select target image materials with a quality score greater than the preset score threshold as valid image data.

[0038] It should be noted that the frequency domain conversion of the target image material can be performed by Fourier transform to convert the image from the spatial domain of pixel arrangement to the frequency domain, reflecting the dominant frequency level of the texture in the image.

[0039] It should also be noted that when selecting valid images, the preset scoring threshold can be a dynamic scoring threshold, such as sorting the quality scores and using the nth quality score as the dynamic scoring threshold.

[0040] Preferably, when the number of target image materials is less than n, the client's local storage is accessed to supplement the target image materials.

[0041] For example, the quality score of the target image material Represented as: ; in, The width of the target image material. The height of the target image material, The median frequency of the histogram. The median of the brightness distribution. The compression ratio of the target image material. This represents the median of the frequency distribution.

[0042] It should be noted that several basic digital human models are preset in the embodiments of the present invention, so as to automatically select the corresponding basic digital human model based on the basic information of the target person.

[0043] For example, please see Figure 2From left to right, these are four basic digital human models preset in this embodiment of the invention. When the target person is identified as male and less than 30 years old, the first basic digital human model is used; when the target person is identified as male and more than 30 years old, the second basic digital human model is used; when the target person is identified as female and less than 30 years old, the third basic digital human model is used; and when the target person is identified as female and more than 30 years old, the fourth basic digital human model is used.

[0044] Understandably, the above-mentioned basic digital human model's form and selection logic are only examples. In actual use, the number of basic digital human models is not limited, but it is required that the overall model can cover all basic information.

[0045] The above scheme uses a dual screening mechanism of target person identification and image quality scoring to retain high-quality, highly relevant, and effective image data from image materials, providing reliable support for basic information identification and thus ensuring the matching accuracy of the basic digital human model.

[0046] In some preferred embodiments, the basic digital human model is reconstructed, that is, the features of the basic digital human are reconstructed. Preferably, the features include facial features, body shape features, clothing features, and background features.

[0047] It should be noted that the basic digital human model possesses a complete human body structure, but lacks personalized details, only providing the basic skeleton that constitutes a digital human. By reconstructing specific features, the basic digital human model can be customized.

[0048] More preferably, the features also include hairstyle features and accessory features.

[0049] In a preferred embodiment, when reconstructing facial features, step S2 involves reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model, which is performed through steps S211-S216: S211. Based on the user behavior data and the image data, analyze and obtain the user's face shape style preference and facial feature style preference; S212. Select a target face shape from the preset face shapes according to the face shape style preference; S213. Obtain user posture data from user behavior data, and calculate the vertical reference axis where the center point of the face is located based on the user posture data and the target face shape. S214. Calculate the first horizontal axis, second horizontal axis, and third horizontal axis of the face based on the height of the target face shape; wherein, the first horizontal axis is the horizontal reference line of the eyes, the second horizontal axis is the horizontal reference axis of the center point of the face, and the third horizontal reference line of the mouth. S215. Based on the vertical reference axis, the first horizontal axis, the second horizontal axis, and the third horizontal axis, and based on the facial feature style preference, obtain the facial feature shape and position; S216. Based on the target face shape and the shape and position of the facial features, the facial features of the basic digital human model are reconstructed to obtain a static digital human model.

[0050] It should be noted that the user behavior data and the image data can reflect the target person's own facial features and the user's preference features. In this embodiment of the invention, the two are fused and matched to obtain the user's style preference for face shape and style preference for facial features.

[0051] In some preferred embodiments, when determining facial style preferences, user habits are extracted based on user behavior data. Combining these habits with preset facial shapes, a matching score between each facial shape and the target person is calculated, ultimately yielding the facial style preference.

[0052] In this embodiment of the invention, unlike the traditional cross-positioning method, facial features are located using a rotatable vertical reference axis and three horizontal axes, which can adapt to the proportions and shapes of facial features under different angles and face shapes.

[0053] Further, preferably, step S211, analyzing the user's face shape style preference and facial feature style preference based on the user behavior data and the image data, includes: Based on user behavior data and the image data, game data, storage data, sharing data, and publishing data are obtained, from which user facial style preferences are extracted; Based on user behavior data and the image data, emotion description data and usage habit data are obtained. Combined with the facial style preference, the user's facial feature style preference is extracted.

[0054] It should be noted that the game data includes data related to behaviors such as storing, searching, browsing, and sharing games. In this embodiment of the invention, user style preferences are determined based on game style. For example, for a certain game, its style includes MOBA, role-playing, and action, where the MOBA style index is 1.0, the role-playing sub-style index is 0.4, and the action style index is 0.3. Further, the user's style preferences are obtained based on the game's style.

[0055] The stored data includes the number of categorized documents and categorized document tags when the user stores them. The shared data includes the content shared by the user and the frequency of user sharing actions. The published data includes the dynamic content and dynamic interactive information published by the user. By storing data, sharing data and publishing data, the user's usage habits can be obtained, reflecting the user's preference for face shape and facial feature style, and finally outputting face shape style preference and facial feature style preference.

[0056] In some preferred embodiments, several face shapes are preset in the embodiments of the present invention, and the most suitable face shape is calculated based on user behavior data and image data to obtain the face shape preference style.

[0057] For example, each face shape is numbered, and the face shape matching coefficient is expressed as follows: ; in, Adjust the coefficients for the game style. This refers to the game behavior style bias calculated based on game data. Adjust the coefficient for storage style. These are document category classification preference values ​​calculated based on stored data. Adjust the coefficient for the content sharing preference style. The sharing function usage preference value is calculated based on the sharing data. Adjustment coefficient for posting dynamic messages. This is the frequency index of publishing dynamic messages, calculated based on the published data.

[0058] Furthermore, in some preferred embodiments, the game behavior style preference is represented as: ; in, The number of games the user has searched for. Let i be the game style of the i-th searched game, where... , For users searching for game-related content Search count for all games For the game style in user space The number of related documents downloaded For the user's last search for game style The difference from the current time.

[0059] It's understandable that some users will store game strategy images and screenshots, and then use the search function to retrieve and read them when needed. Game data can reflect a user's style preferences to some extent.

[0060] The stored data includes information about documents stored by the user. In some preferred embodiments, document category scoring is performed based on the user's document classification pattern. Furthermore, the document category classification preference value is represented as: ; in, This represents the category score for the i-th document. Total number of user document categories.

[0061] Sharing data includes user sharing behavior, namely, the functions of sharing files and generating shared links. In some preferred embodiments, the sharing function is represented using preference values: ; in, Number of times a user shares Let be the total number of clicks after the i-th user shares the content. The total duration of the sharing action for the i-th user.

[0062] In some preferred embodiments, the frequency index of posting dynamic messages is expressed as: ; in, The total number of posts published by users. The average time interval for users to post updates. The average number of citations for users is dynamically calculated. This represents the average number of comments received by users in their activity feeds.

[0063] When extracting a user's facial feature style preferences, emotional description data is extracted, and the shape of the facial features can be obtained through this data. For example, the eye rotation angle and the upward angle of the corners of the mouth can be adjusted using emotional description data. Combined with usage habit data, such as gaming habits and lifestyle habits, a more refined understanding of facial feature style preferences can be obtained.

[0064] In the above scheme, game data, stored data, shared data, and published data are filtered and extracted from user behavior data and image data. The frequently occurring facial features in these data are analyzed, and these features are combined to obtain the user's facial style preference. Furthermore, emotional description data and usage habit data are extracted from user behavior data and image data. Combined with facial style preference, the facial feature tendencies that match the facial shape are analyzed, and finally, the user's facial feature style preference is extracted.

[0065] Preferably, step S213, obtaining user posture data from user behavior data, and calculating the vertical reference axis of the facial center point based on the user posture data and the target face shape, includes: User posture data is obtained by acquiring the user's horizontal yaw angle, vertical yaw angle, and eye direction from user behavior data. Calculate the straight-line distance between the user's eyes and the device based on the user's eye direction; Based on the target face shape, the center point of the face is obtained, and a vertical axis is drawn on the center point of the face to obtain the initial vertical reference axis. Based on the target face shape, the user's horizontal deflection angle, the user's vertical deflection angle, and the straight-line distance, the deflection distance of the initial vertical reference axis is calculated to obtain the vertical reference axis.

[0066] It should be noted that the user posture data is obtained through the phone's gyroscope and camera data. Specifically, the phone's gyroscope is used to obtain the user's horizontal and vertical deflection angles, and the camera data is used to read the user's eye direction. The distance sensor reading in that direction is then read to obtain the straight-line distance between the user's eyes and the device. When any value cannot be read, the user's most recent historical value is used; if the historical value is empty, the average value of all users is taken.

[0067] In this embodiment of the invention, after obtaining the user's horizontal deflection angle and the user's vertical deflection angle, they are normalized and then calculated.

[0068] For example, the user's horizontal deflection angle Vertical deflection angle with the user The normalization transformation is expressed as: ; ; in, This indicates the user's horizontal deflection angle. Vertical deflection angle with the user The average value. Using the normalized user horizontal deflection angle. Vertical deflection angle with the user Perform subsequent calculations.

[0069] It should be noted that, please refer to Figure 3 The facial center point is initially determined based on the length and width of the target face shape. A perpendicular line is drawn from the facial center point to obtain the initial vertical reference axis. Then, based on the user's horizontal deflection angle, the user's vertical deflection angle, and the straight-line distance, the initial vertical reference axis is deflected. During this deflection process, the facial center point will move with the vertical reference axis to ensure that the facial center point is always at the center of the vertical reference axis.

[0070] In some preferred embodiments, the deflection distance through the initial vertical reference axis includes both horizontal and vertical deflection distances.

[0071] Preferably, to avoid large-scale deflection affecting image size, the horizontal deflection distance is calculated based on the user's horizontal deflection angle, the user's vertical deflection angle, and the width of the target face shape.

[0072] For example, the horizontal deflection distance is expressed as: ; in, The preset horizontal deflection angle threshold, This means selecting the one with the smallest absolute value between x and y. The width of the target face shape.

[0073] Preferably, the vertical deflection distance is calculated based on the horizontal deflection distance, the user's horizontal deflection angle, the user's vertical deflection angle, the height of the target face shape, and the straight-line distance between the user's eyes and the device.

[0074] For example, the vertical deflection distance is expressed as: ; in, The preset vertical deflection angle threshold, For the height of the target face shape, The straight-line distance between the user's eyes and the device.

[0075] In the above scheme, by extracting horizontal deflection, vertical deflection, eye direction and straight-line distance, the angle, posture and spatial position of the user's head are covered, avoiding axis deviation caused by a single parameter, so that the vertical reference axis is more in line with the user's actual posture. Furthermore, the initial vertical reference axis is determined based on the target face shape, and then combined with posture parameter adjustment to ensure that the vertical reference axis is aligned with the center point of the face and adapts to head deflection, forming a close relationship with facial features.

[0076] Preferably, step S214, calculating the first horizontal axis, second horizontal axis, and third horizontal axis of the face based on the height of the target face shape, includes: Based on the height of the target face shape, three equidistant horizontal axes are drawn on the face, namely the initial first horizontal axis, the initial second horizontal axis, and the initial third horizontal axis; The initial second horizontal axis is adjusted to the center point of the face to obtain the second horizontal axis; From the user behavior data, user social data is obtained to determine the user's facial feature ratio preferences; Based on the height of the target face shape and the facial feature proportion preference, the distances between the initial first horizontal axis and the initial third horizontal axis and the second horizontal axis are adjusted to obtain the first horizontal axis and the third horizontal axis.

[0077] It should be noted that, by default, the three horizontal axes are equally spaced, meaning the distance between any two adjacent horizontal axes is [missing information]. In this embodiment of the invention, to make the layout of the user's facial features more interactive with the user's historical behavior, adjustments are made to each horizontal axis.

[0078] It should also be noted that when the vertical reference axis is deflected, the center point of the face may not be on the initial second horizontal axis. Therefore, in this embodiment of the invention, the initial second horizontal axis is offset and adjusted to ensure that the first horizontal axis and the second horizontal axis can be used as reference lines for the eyes and mouth, respectively.

[0079] Please see Figure 4 By adjusting the initial distance between the first and second horizontal axes The first horizontal axis is obtained by adjusting the distance between the initial third horizontal axis and the second horizontal axis. The third horizontal axis is obtained.

[0080] In some preferred embodiments, the distance between the initial first horizontal axis and the second horizontal axis is calculated using the eye vertical adjustment ratio coefficient and the height of the target face shape, wherein the eye vertical adjustment ratio coefficient is based on the user's facial style preferences analyzed from social data.

[0081] For example, the initial distance between the first and second horizontal axes is expressed as: ; in, For the height of the target face shape, This is the preset maximum threshold for the vertical eye adjustment ratio. This is the vertical accommodation ratio coefficient for the eye.

[0082] In some preferred embodiments, the distance between the initial third horizontal axis and the second horizontal axis is calculated using a mouth vertical adjustment ratio coefficient and the height of the target face shape, wherein the mouth vertical adjustment ratio coefficient is obtained based on an analysis of the degree of fit with the target face shape.

[0083] For example, the distance between the initial third horizontal axis and the second horizontal axis is expressed as: ; in, For the height of the target face shape, The preset maximum threshold for the vertical adjustment ratio of the mouthpiece. This is the vertical adjustment ratio coefficient for the mouthpiece. .

[0084] In the above scheme, an initial axis is established starting from the height of the target face shape, and then the second horizontal axis is anchored at the center point of the face to ensure that the horizontal axis system is deeply bound to the core structure of the face, providing a basic reference for the positioning of facial features that conforms to the proportions of the face shape; facial feature proportion preferences are extracted through social data and used to adjust the distance between the upper and lower horizontal axes, so that the horizontal axis not only conforms to the objective facial structure, but also carries the user's subjective preferences, avoiding generic design and enhancing the exclusivity of facial features.

[0085] Preferably, step S215, obtaining the shape and position of facial features based on the vertical reference axis, the first horizontal axis, the second horizontal axis, and the third horizontal axis, and based on the facial feature style preference, includes: The eye is located based on the vertical reference axis and the first horizontal axis to obtain the eye position. The nose is located based on the second horizontal axis to obtain its position; The mouth is positioned based on the vertical reference axis and the third horizontal axis to obtain the mouth position; The positions of the facial features are obtained based on the positions of the eyes, nose, and mouth. The facial features are adjusted according to the facial feature style preference and the position of the facial features; the facial feature shape includes the size, angle and shape of the facial features.

[0086] In some preferred embodiments, when locating the eyes, the vertical midpoint of the eye coincides with the first horizontal axis, and the vertical center line of both eyes coincides with the vertical reference axis. When locating the nose, the midpoint of the nose coincides with the center point of the face. When locating the mouth, the vertical midpoint of the mouth coincides with the third horizontal axis, and the vertical center line of the mouth coincides with the vertical reference axis.

[0087] As a preferred embodiment, the preset face shape includes the default shape of the facial features. When adjusting the shape of the facial features, the present invention adjusts the shape based on the default shape corresponding to the face shape, and prioritizes adjusting the shape of the eyes, nose and mouth.

[0088] Preferably, the eye shape includes the inner width of the eyes, the outer width of the eyes, and the range of eye rotation. Further, in some preferred embodiments, the inner and outer widths of the eyes are adjusted based on facial feature style preferences, such as considering the user's gaming style preferences and content sharing style preferences. The range of eye rotation is primarily adjusted based on the user's emotional description data.

[0089] For example, the range of eye rotation is expressed as: ; in, The number of words with negative emotional value in the user's search terms. The number of words with positive emotional value among the user's search terms. This refers to the number of neutral words in user search terms that do not have a clear emotional value bias. This represents the percentage of readable files out of all files in user space. This represents the percentage of scenario-type files among all files in the user space.

[0090] Preferably, the nose shape includes nose length and nose width. Further, in some preferred embodiments, the nose shape is primarily adjusted based on the user's interest in the fitness content.

[0091] For example, the length of the nose is expressed as: ; in, The average duration of a single image view for a user. The median duration of a single image view by a user is 80%. The percentage of images downloaded after browsing them. This represents the distance between the first and second horizontal axes.

[0092] For example, the width of the nose is represented as: ; ; in, This is the base coefficient for nose width. As a potential fitness index for users, The number of user photos whose environment is identified as a fitness venue. This refers to the number of user documents that contain fitness content. This refers to the number of times a user uses their fitness peripherals and cloud-based fitness functions. The sum of users' fitness intensity is represented by ms, which is the recommended intensity coefficient. The sum of user fitness frequencies is denoted by mf, which is the recommendation frequency coefficient.

[0093] Preferably, the mouth shape includes the mouth width and the upward angle of the corners of the mouth. The upward angle of the corners of the mouth is mainly adjusted based on the user's emotional description data.

[0094] In some preferred embodiments, the mouth width is calculated based on the shape of the nose.

[0095] For example, the width of the mouth is represented as: ; in, The length of the nose. The width of the nose The distance between the third horizontal axis and the second horizontal axis. This represents the offset distance of the mouth position relative to the third horizontal axis.

[0096] For example, the angle of the corners of the mouth turning up is represented as: ; in, The number of words with negative emotional value in the user's search terms. The number of words with positive emotional value among the user's search terms. This refers to the number of neutral words in user search terms that do not have a clear emotional value bias. The total number of files in user space. This represents the number of known, unencrypted, and readable files in user space. This represents the sentiment bias coefficient for readable documents. This indicates a default sentiment neutrality index. For user age, The percentage of entertainment files in all of the user's files. The percentage of entertainment files per user. This represents the percentage of individual user images within the overall image. This represents the percentage of individual images per user on average.

[0097] In the above scheme, the eyes, nose and mouth are positioned by combining the vertical reference axis and three horizontal axes, so that they correspond to their own reference lines. This ensures that the position of each facial feature conforms to the logic of facial structure, forming a harmonious and unified facial feature layout, and avoiding spatial misalignment or proportional imbalance. The axis ensures the standardization of the position of the facial features, and the shape is adjusted according to the user's facial feature style preference to meet the user's personalized needs for facial feature shape.

[0098] As a preferred implementation, when reconstructing body features, step S2 involves reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model, which is executed through steps S221-S225: S221. Based on user behavior data and the image data, obtain user height data, clothing browsing data and clothing order data, and calculate the torso height of the basic digital human model; S222. Calculate the torso width of the basic digital human model based on the image data; S223. Calculate the ratio of the torso height to the torso width, and select a target body shape from the preset body shapes; S224. Calculate the limb dimensions of the basic digital human model based on the target body shape; S225. Based on the torso height, torso width, and limb dimensions, the body features of the basic digital human model are reconstructed to obtain a static digital human model.

[0099] In some preferred embodiments, if the target person is a user and the user has already filled in their height data, then the filled-in height data is used directly; otherwise, the height of the target person is inferred based on image data, clothing browsing data, and clothing order data, in order to further calculate the torso height of the basic digital human model.

[0100] For example, the estimation of a target person's height can be expressed as: ; in, The default height parameter is used. This indicates whether the user's height data has been entered in their basic information; 1 indicates yes, 0 indicates no. The height data in the user's basic information. This indicates whether the user clicked on a clothing advertisement and placed an order; a value of 1 indicates yes, and a value of 0 indicates no. The average size of clothing ordered by the user. This indicates whether the user clicked on the shoe advertisement and placed an order; a value of 1 indicates yes, and a value of 0 indicates no. The average size of shoes ordered by the user.

[0101] Preferably, the torso height of the basic digital human model is represented as: .

[0102] As a preferred implementation, step S222, calculating the torso width of the basic digital human model based on the image data, includes: Based on the image data, the user's image perspective preference and lifestyle image preference are obtained; Calculate the initial torso width of the basic digital human model based on the image perspective preference; Calculate the user's body shape confidence based on the aforementioned lifestyle image preferences; Based on the body shape information, the initial torso width is adjusted to obtain the torso width.

[0103] For example, the initial torso width is represented as: ; in, This represents the number of images taken from the user's camera at a specific angle. Let i be the total viewing time of the user for the i-th image. The average total time users spend browsing images. This indicates the total number of photos a user has. This represents the number of images in the user's photo safe and secret album.

[0104] It should be noted that considering the impact of the image's perspective on the user's body shape perspective can reflect the user's expectations for body width. From this, we can obtain the user's expected torso width for the static digital human model. For example, thinner users prefer to face the camera directly to make themselves look fuller, while heavier users prefer to stand sideways to make themselves appear thinner.

[0105] For example, body shape confidence is represented as: ; in, This refers to the number of images in the image data that contain the user's facial features and can be publicly viewed by friends. The frequency of browsing weight loss images, along with images containing fitness and beauty information, is adjusted based on the user's currently stored images. The number of image categories stored for the user.

[0106] Preferably, the initial torso width and the body shape confidence are multiplied to calculate the torso width.

[0107] In some preferred embodiments, several body types are preset, with different limb proportions for each type, such as apple-shaped, H-shaped, and pear-shaped. By calculating the ratio of the torso height to the torso width, the target body type of the basic digital human model can be obtained.

[0108] Furthermore, based on the target body shape, the limb dimensions of the basic digital human model are calculated, including the upper limb length, upper limb width, lower limb length, and lower limb width.

[0109] More preferably, the limbs are further subdivided, such as dividing the upper limbs into upper arms and forearms, and the lower limbs into thighs and calves, and adjusting each part accordingly. This adjustment process takes into account the target person's body shape characteristics and the user's interests.

[0110] The above solution calculates torso parameters based on user height, clothing data, and image data to ensure that body shape features are related to the user's actual characteristics and reflect their preferences, thus avoiding generic body shape design. The logic of selecting the target body shape by the torso height-to-width ratio and then calculating the limb dimensions ensures that the torso and limbs are proportionally matched, avoids the disconnect between different parts of the body shape, and improves the natural coordination of the digital human body shape.

[0111] As a preferred implementation, when reconstructing the clothing features, step S2 involves reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model, which is executed through steps S231-S232: S231. Based on the image data, obtain color data and light and shadow transformation data, and analyze the user's clothing style preference and clothing color compatibility. S232. Based on the clothing style preference and the clothing color compatibility, the clothing features of the basic digital human model are reconstructed to obtain a static digital human model.

[0112] To ensure the digital human is more holistic, after determining the facial features, torso, and other components, in this embodiment of the invention, the clothing style is further adjusted by combining the color and light and shadow transformation coefficients in the source images.

[0113] In some preferred embodiments, a style coefficient is introduced to map clothing. For example, the style coefficient gradually increases in the order of home, sports, leisure, workplace, evening, and punk, and the user's clothing style preference is obtained based on the style coefficient.

[0114] In some preferred embodiments, the color matching suitability of the clothing of the person in the image data is calculated using color blocks in a preset color scheme table. For example, when the identified color cannot match a color block in the color scheme table, the suitability is 1; otherwise, it is 0.5.

[0115] Preferably, by analyzing the clothing style component tags of user images, the color saturation vector length of the top three images with the highest image quality scores, the DPI / 100 value in the image resolution, the frequency of user adjustments to classification criteria tags, and color matching suitability, a digital person with a distinctive clothing style that better suits the user's personality traits can be generated.

[0116] By combining color saturation vectors and image resolution, it's possible to deduce which colors and image qualities best suit a user's personal style, thus improving the accuracy of clothing choices for digital humans. More preferably, by tracking the frequency with which users adjust classification criteria tags, it's possible to understand their adaptability and preferences regarding fashion trends, resulting in a digital human with overall harmonious and aesthetically pleasing clothing style.

[0117] In the above scheme, by combining clothing style preferences and color compatibility for reconstruction, the digital human's clothing can not only conform to the user's preferred style, but also have a harmonious color matching, ultimately resulting in a static model of a digital human with personalized clothing.

[0118] As a preferred implementation, when reconstructing the background features, step S2 involves reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model, which is executed through steps S241-S242: S241. Extract the user's background preference style based on the user behavior data and the image data; S242. Based on the background preference style, reconstruct the background features of the basic digital human model to obtain a static digital human model.

[0119] It should be noted that several background image templates are pre-set for selection. For example, the background image templates include outdoor scenery, outdoor scenes, underwater scenes, karaoke rooms, libraries, and vehicles. When reconstructing background features, a matching target background is selected from the background image templates.

[0120] For example, the initial background value of the background image template is represented as: ; in, The width of the background image template. The height of the background image template. For target image The grayscale value at position i,j.

[0121] Furthermore, the user's behavioral context value is represented as: ; in, Background image templates The average value, This represents the maximum number of download tasks a user can perform when there is no limit to the number of concurrent downloads. For source images quantity, Let f be the aperture number of the i-th image. Let be the focal length of the i-th source image. (x / skin) is the clothing coefficient. This is a preset correction factor for the dress code.

[0122] Next, the user's behavioral background value is compared with the initial background value of each background image template, and the selected background image is chosen. and The closest background image template is used to reconstruct the background features of the basic digital human model.

[0123] In the above solution, preferences are extracted through user behavior data and image data to ensure that the reconstructed background not only conforms to the user's subjective preferences but also relates to the actual scene of the target person, thereby enhancing the exclusivity and relevance of the background. Furthermore, the background features match the style of the digital human, making the static model of the digital human form a unified and harmonious visual whole.

[0124] As a preferred implementation, step S3, filtering target video data from the user behavior data and extracting human motion trajectory information from the target video data, includes: Filter audio playback data or music video playback data from the user behavior data, and use the music video corresponding to the audio playback data or the music video playback data as the target video data; A neural motion feature learning algorithm is used to track the motion of the human body in the target video data and obtain the human body motion trajectory information.

[0125] It should be noted that the audio playback data refers to records related to audio file playback in the user behavior data, such as the number of times played, playback duration, and favorited audio files; the music video playback data refers to records related to music video playback in the user behavior data.

[0126] Preferably, audio playback data or music video playback data is filtered based on the playback frequency and download frequency of all audio files and music video files. It should be noted that the music videos corresponding to the audio playback data refer to music videos such as music videos and dance videos.

[0127] In this embodiment of the invention, the human body is tracked based on marked points and unmarked points using a neural motion feature learning algorithm to obtain human motion trajectory information.

[0128] In the above solution, by filtering audio or music videos played by the user, the target video data is ensured to be strongly correlated with the user's interests. A neuromotor feature learning algorithm is used to ensure the accuracy and completeness of human motion trajectory information, so that the digital human can accurately reproduce the dance movements in the video and improve the realism of the final video.

[0129] As a preferred implementation, step S4, adding the human motion trajectory information to the digital human static model to generate a digital human memory video, includes: The fusion tool interface is invoked to add the human motion trajectory information to the digital human static model, and the digital human static model is rendered to generate a digital human memory video.

[0130] In some preferred embodiments, the fusion tool interface is the Blender software capability interface, which can fuse human motion trajectory information and digital human information to generate animated digital human videos.

[0131] For example, the video content of the digital human memory video is a video of a digital human with a unique user profile performing a dance to music that the user once liked.

[0132] In the above scheme, the rendering process transforms the abstract data of static model and trajectory information into concrete dynamic images, allowing the digital human to make coherent movements according to the human motion trajectory information in the target video data. The rendered video fully preserves the personalized features of the digital human static model and the authenticity of the human motion trajectory, taking into account both personalization and dynamism.

[0133] The digital human memory video generation method provided by the present invention can generate digital human memory videos that not only fit the characteristics of specific target personnel, but also carry memory scenes related to the user's past behavior, providing users with unique digital human image memory videos.

[0134] This invention provides a digital human memory video generation system. Please refer to [link / reference]. Figure 5 The digital human memory video generation system includes a basic model selection module 11, a basic model reconstruction module 12, a trajectory information extraction module 13, and a memory video generation module 14, wherein: The basic model selection module 11 is used to select a preset basic digital human model based on the image data of the target person; The basic model reconstruction module 12 is used to reconstruct the basic digital human model based on user behavior data and the image data to obtain a static digital human model. The trajectory information extraction module 13 is used to filter target video data from the user behavior data and extract human motion trajectory information from the target video data. The memory video generation module 14 is used to add the human motion trajectory information to the digital human static model to generate a digital human memory video.

[0135] In a preferred embodiment, the basic model selection module 11 is specifically used for: Obtain all image materials uploaded by users, perform target person identification on the image materials, and obtain target image materials; The target image materials are scored for quality, and valid image data is selected from the target image materials based on the quality score results; Based on the valid image data, identify the basic information of the target person; the basic information includes gender and age. Based on the aforementioned basic information, a preset basic digital human model is selected.

[0136] In a preferred embodiment, the basic model reconstruction module 12 reconstructs facial features, and the basic model reconstruction module 12 includes: The facial preference analysis unit is used to analyze and obtain the user's face shape style preference and facial feature style preference based on the user behavior data and the image data; The target face shape selection unit is used to select a target face shape from preset face shapes based on the face shape style preference. The vertical reference axis calculation unit is used to obtain user posture data from user behavior data and calculate the vertical reference axis where the center point of the face is located based on the user posture data and the target face shape. The horizontal axis calculation unit is used to calculate the first horizontal axis, the second horizontal axis, and the third horizontal axis of the face based on the height of the target face shape; wherein, the first horizontal axis is the horizontal reference line of the eyes, the second horizontal axis is the horizontal reference axis of the center point of the face, and the third horizontal reference line of the mouth; The facial feature determination unit is used to obtain the shape and position of facial features based on the vertical reference axis, the first horizontal axis, the second horizontal axis and the third horizontal axis, and the facial feature style preference. The facial feature reconstruction unit is used to reconstruct the facial features of the basic digital human model based on the target face shape and the shape and position of the facial features to obtain a static digital human model.

[0137] Further, preferably, the facial preference analysis unit is specifically used for: Based on user behavior data and the image data, game data, storage data, sharing data, and publishing data are obtained, from which user facial style preferences are extracted; Based on user behavior data and the image data, emotion description data and usage habit data are obtained. Combined with the facial style preference, the user's facial feature style preference is extracted.

[0138] Preferably, the vertical reference axis calculation unit is specifically used for: User posture data is obtained by acquiring the user's horizontal yaw angle, vertical yaw angle, and eye direction from user behavior data. Calculate the straight-line distance between the user's eyes and the device based on the user's eye direction; Based on the target face shape, the center point of the face is obtained, and a vertical axis is drawn on the center point of the face to obtain the initial vertical reference axis. Based on the target face shape, the user's horizontal deflection angle, the user's vertical deflection angle, and the straight-line distance, the deflection distance of the initial vertical reference axis is calculated to obtain the vertical reference axis.

[0139] Preferably, the horizontal axis calculation unit is specifically used for: Based on the height of the target face shape, three equidistant horizontal axes are drawn on the face, namely the initial first horizontal axis, the initial second horizontal axis, and the initial third horizontal axis; The initial second horizontal axis is adjusted to the center point of the face to obtain the second horizontal axis; From the user behavior data, user social data is obtained to determine the user's facial feature ratio preferences; Based on the height of the target face shape and the facial feature proportion preference, the distances between the initial first horizontal axis and the initial third horizontal axis and the second horizontal axis are adjusted to obtain the first horizontal axis and the third horizontal axis.

[0140] Preferably, the facial features determination unit is specifically used for: The eye is located based on the vertical reference axis and the first horizontal axis to obtain the eye position. The nose is located based on the second horizontal axis to obtain its position; The mouth is positioned based on the vertical reference axis and the third horizontal axis to obtain the mouth position; The positions of the facial features are obtained based on the positions of the eyes, nose, and mouth. The facial features are adjusted according to the facial feature style preference and the position of the facial features; the facial feature shape includes the size, angle and shape of the facial features.

[0141] In a preferred embodiment, the basic model reconstruction module 12 reconstructs body shape features, and the basic model reconstruction module 12 further includes: The torso height calculation unit is used to calculate the torso height of the basic digital human model based on user behavior data and the image data, which are obtained from user height data, clothing browsing data and clothing order data. The torso width calculation unit is used to calculate the torso width of the basic digital human model based on the image data. The target body shape selection unit is used to calculate the ratio of the torso height to the torso width and select the target body shape from the preset body shapes. The limb size calculation unit is used to calculate the limb size of the basic digital human model based on the target body shape. The body shape feature reconstruction unit is used to reconstruct the body shape features of the basic digital human model based on the torso height, torso width and limb dimensions to obtain a static digital human model.

[0142] In a preferred embodiment, the basic model reconstruction module 12 reconstructs the clothing features, and the basic model reconstruction module 12 further includes: The clothing preference analysis unit is used to obtain color data and light and shadow transformation data based on the image data, and to analyze the user's clothing style preference and clothing color suitability. The clothing feature reconstruction unit is used to reconstruct the clothing features of the basic digital human model based on the clothing style preference and the clothing color adaptation, so as to obtain a static digital human model.

[0143] In a preferred embodiment, the basic model reconstruction module 12 reconstructs the background features, and the basic model reconstruction module 12 further includes: The background preference analysis unit is used to extract the user's background preference style based on user behavior data and the image data; The background feature reconstruction unit is used to reconstruct the background features of the basic digital human model according to the background preference style to obtain a static digital human model.

[0144] In a preferred embodiment, the trajectory information extraction module 13 is specifically used for: Filter audio playback data or music video playback data from the user behavior data, and use the music video corresponding to the audio playback data or the music video playback data as the target video data; A neural motion feature learning algorithm is used to track the motion of the human body in the target video data and obtain the human body motion trajectory information.

[0145] In a preferred embodiment, the memory video generation module 14 is specifically used for: The fusion tool interface is invoked to add the human motion trajectory information to the digital human static model, and the digital human static model is rendered to generate a digital human memory video.

[0146] The digital human memory video generation system provided by this invention can generate digital human memory videos that not only match the characteristics of specific target individuals, but also carry memory scenes related to the user's past behaviors, providing users with unique digital human image memory videos.

[0147] This invention also provides a digital human memory video generation device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the digital human memory video generation method described above. The working principles and beneficial effects of the two are one-to-one, so they will not be described in detail here.

[0148] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0149] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for generating digital human memory videos, characterized in that, include: Based on the image data of the target personnel, select a preset basic digital human model; Based on user behavior data and the image data, the basic digital human model is reconstructed to obtain a static digital human model; Filter target video data from the user behavior data and extract human motion trajectory information from the target video data; The human motion trajectory information is added to the digital human static model to generate a digital human memory video.

2. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of selecting a preset basic digital human model based on the image data of the target person includes: Obtain all image materials uploaded by users, identify the target person in the image materials, and obtain the target image materials; The target image materials are scored for quality, and valid image data is selected from the target image materials based on the quality score results; Based on the valid image data, identify the basic information of the target person; the basic information includes gender and age. Based on the aforementioned basic information, a preset basic digital human model is selected.

3. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on the user behavior data and the image data, the user's face shape style preference and facial feature style preference are analyzed; Based on the stated face shape style preference, select the target face shape from the preset face shapes; User posture data is obtained from user behavior data, and the vertical reference axis of the face center point is calculated based on the user posture data and the target face shape. Based on the height of the target face shape, calculate the first horizontal axis, the second horizontal axis, and the third horizontal axis of the face; wherein, the first horizontal axis is the horizontal reference line of the eyes, the second horizontal axis is the horizontal reference axis of the center point of the face, and the third horizontal reference line of the mouth; Based on the vertical reference axis, the first horizontal axis, the second horizontal axis, and the third horizontal axis, and based on the facial feature style preference, the shape and position of the facial features are obtained; Based on the target face shape and the shape and position of the facial features, the facial features of the basic digital human model are reconstructed to obtain a static digital human model.

4. The method for generating digital human memory videos as described in claim 3, characterized in that, The step of analyzing and obtaining the user's facial shape style preference and facial feature style preference based on the user behavior data and the image data includes: Based on user behavior data and the image data, game data, storage data, sharing data, and publishing data are obtained, from which user facial style preferences are extracted; Based on user behavior data and the image data, emotion description data and usage habit data are obtained. Combined with the facial style preference, the user's facial feature style preference is extracted.

5. The method for generating digital human memory videos as described in claim 3, characterized in that, The step of obtaining user posture data from user behavior data, and calculating the vertical reference axis of the facial center point based on the user posture data and the target face shape, includes: User posture data is obtained by acquiring the user's horizontal yaw angle, vertical yaw angle, and eye direction from user behavior data. Calculate the straight-line distance between the user's eyes and the device based on the user's eye direction; Based on the target face shape, the center point of the face is obtained, and a vertical axis is drawn on the center point of the face to obtain the initial vertical reference axis. Based on the target face shape, the user's horizontal deflection angle, the user's vertical deflection angle, and the straight-line distance, the deflection distance of the initial vertical reference axis is calculated to obtain the vertical reference axis.

6. The method for generating digital human memory videos as described in claim 3, characterized in that, The step of calculating the first horizontal axis, second horizontal axis, and third horizontal axis of the face based on the height of the target face shape includes: Based on the height of the target face shape, three equidistant horizontal axes are drawn on the face, namely the initial first horizontal axis, the initial second horizontal axis, and the initial third horizontal axis; The initial second horizontal axis is adjusted to the center point of the face to obtain the second horizontal axis; From the user behavior data, user social data is obtained to determine the user's facial feature ratio preferences; Based on the height of the target face shape and the facial feature proportion preference, the distances between the initial first horizontal axis and the initial third horizontal axis and the second horizontal axis are adjusted to obtain the first horizontal axis and the third horizontal axis.

7. The method for generating digital human memory videos as described in claim 3, characterized in that, The process of obtaining the shape and position of facial features based on the vertical reference axis, the first horizontal axis, the second horizontal axis, and the third horizontal axis, and based on the facial feature style preference, includes: The eye is located based on the vertical reference axis and the first horizontal axis to obtain the eye position. The nose is located based on the second horizontal axis to obtain its position; The mouth is positioned based on the vertical reference axis and the third horizontal axis to obtain the mouth position; The positions of the facial features are obtained based on the positions of the eyes, nose, and mouth. The facial features are adjusted according to the facial feature style preference and the position of the facial features; the facial feature shape includes the size, angle and shape of the facial features.

8. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on user behavior data and the image data, user height data, clothing browsing data, and clothing order data are obtained, and the torso height of the basic digital human model is calculated. Based on the image data, calculate the torso width of the basic digital human model; Calculate the ratio of the torso height to the torso width, and select a target body shape from the preset body shapes; Based on the target body shape, calculate the limb dimensions of the basic digital human model; Based on the torso height, torso width, and limb dimensions, the body features of the basic digital human model are reconstructed to obtain a static digital human model.

9. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on the image data, color data and light and shadow transformation data are obtained, and the user's clothing style preference and clothing color compatibility are analyzed. Based on the clothing style preference and the clothing color compatibility, the clothing features of the basic digital human model are reconstructed to obtain a static digital human model.

10. A method for generating digital human memory videos as described in claim 1, characterized in that, The step of reconstructing the basic digital human model based on user behavior data and the image data to obtain a static digital human model includes: Based on user behavior data and the image data, extract the user's background preference style; Based on the background preference style, the background features of the basic digital human model are reconstructed to obtain a static digital human model.

11. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of filtering target video data from the user behavior data and extracting human motion trajectory information from the target video data includes: Filter audio playback data or music video playback data from the user behavior data, and use the music video corresponding to the audio playback data or the music video playback data as the target video data; A neural motion feature learning algorithm is used to track the motion of the human body in the target video data and obtain the human body motion trajectory information.

12. The method for generating digital human memory videos as described in claim 1, characterized in that, The step of adding the human motion trajectory information to the digital human static model to generate a digital human memory video includes: The fusion tool interface is invoked to add the human motion trajectory information to the digital human static model, and the digital human static model is rendered to generate a digital human memory video.

13. A digital human memory video generation system, characterized in that, include: The basic model selection module is used to select a preset basic digital human model based on the image data of the target person; The basic model reconstruction module is used to reconstruct the basic digital human model based on user behavior data and the image data to obtain a static digital human model. The trajectory information extraction module is used to filter target video data from the user behavior data and extract human motion trajectory information from the target video data. The memory video generation module is used to add the human motion trajectory information to the digital human static model to generate a digital human memory video.

14. A digital human memory video generation device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the digital human memory video generation method as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the digital human memory video generation method as described in any one of claims 1 to 12.

16. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, perform the digital human memory video generation method as described in any one of claims 1 to 12.

Images