Multi-keyframe redirection method, multi-keyframe redirection model construction method and system

Abstract

The present invention provides a multi-keyframe redirection method, a multi-keyframe redirection model construction method and a system, the method comprising: obtaining a source keyframe sequence of a source object; based on a multi-keyframe redirection model, redirecting each source keyframe in the source keyframe sequence to a corresponding character animation keyframe, and obtaining a character animation video sequence corresponding to the source object, wherein the multi-keyframe redirection model comprises a global keyframe conversion model and a keyframe residual model. The present invention effectively avoids the semantic deviation between a real actor and an animation character in a single keyframe redirection process, and improves the accuracy of the keyframe redirection result.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a multi-keyframe redirection method, a multi-keyframe redirection model construction method and a system. Background Art

[0002] Retargeting technology plays a vital role in virtual humans, animation production and game production. The main goal of this technology is to achieve smooth dynamic effects and natural deformation of animated characters in animations or virtual scenes, thereby enhancing visual realism and user experience.

[0003] In existing redirection technologies, Tpose (or Apose) is usually used as a keyframe to solve a transformation matrix, so that the corresponding animation character posture can be obtained for any input real actor posture. However, due to the difference in bones between real actors and animation characters, redirection through this transformation matrix leads to a large deviation in the action semantics between real actors and animation characters, and the action semantics of real actors cannot be accurately expressed through animation characters.

[0004] Therefore, there is an urgent need for a multi-keyframe redirection method, a multi-keyframe redirection model construction method and a system to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a multi-keyframe redirection method, a multi-keyframe redirection model construction method and a system.

[0006] The present invention provides a multi-keyframe redirection method, comprising:

[0007] Get the source key frame sequence of the source object;

[0008] Based on a multi-keyframe redirection model, each source keyframe in the source keyframe sequence is redirected to a corresponding character animation keyframe to obtain a character animation video sequence corresponding to the source object;

[0009] Wherein, the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model;

[0010] The global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames;

[0011] The key frame residual model is constructed based on the distance difference between the sample source key frame and other sample source key frames of each frame, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame.

[0012] According to a multi-keyframe redirection method provided by the present invention, the global keyframe conversion model is obtained by the following steps:

[0013] The global key frame conversion models corresponding to different sample animation video scene types are obtained according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames.

[0014] According to a multi-keyframe redirection method provided by the present invention, the keyframe residual model is obtained by the following steps:

[0015] Obtaining a sample standard animation key frame corresponding to the sample source key frame through the global key frame conversion model;

[0016] Constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames;

[0017] Calculate the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame;

[0018] Based on the radial basis function, the key frame residual model is constructed according to the mapping relationship between the key frame residual and the deviation value.

[0019] According to a multi-keyframe redirection method provided by the present invention, the global keyframe conversion model corresponding to different sample animation video scene types is obtained according to the sample source keyframes and the sample animation keyframes corresponding to each of the sample source keyframes, including:

[0020] Acquire a first sample standard posture and a second sample standard posture, wherein the first sample standard posture is a standard posture corresponding to the sample source key frame, and the second sample standard posture is a standard posture corresponding to the sample animation key frame;

[0021] Based on the sample animation video scene type, the corresponding global key frame conversion model is calculated according to the first sample standard posture and the second sample standard posture.

[0022] According to a multi-keyframe redirection method provided by the present invention, the keyframe residual corresponding to the sample source keyframe of each frame is constructed according to the distance difference between the sample source keyframe of each frame and other sample source keyframes, including:

[0023] Based on the distance function, the distance difference between the sample source key frame of each frame and other sample source key frames is calculated;

[0024] According to the distance difference, a key frame residual corresponding to the sample source key frame of each frame is constructed.

[0025] According to a multi-keyframe redirection method provided by the present invention, before obtaining the global keyframe conversion model corresponding to different sample animation video scene types according to the sample source keyframes and the sample animation keyframes corresponding to each of the sample source keyframes, the method further includes:

[0026] Based on the farthest point sampling algorithm, the frame sequence is sampled to obtain the corresponding key frames in the frame sequence, wherein the frame sequence includes a sample source frame sequence and / or a sample animation frame sequence.

[0027] According to a multi-key frame redirection method provided by the present invention, the farthest point sampling algorithm is used to sample a frame sequence to obtain corresponding key frames in the frame sequence, including:

[0028] Step 1, randomly sampling in the frame sequence, and adding the randomly sampled frames as sampling frames to a sampling pool;

[0029] Step 2, obtaining a matching frame from the remaining frames of the frame sequence, and adding the matching frame to the sampling pool, wherein the matching frame is a frame having the largest minimum distance to the sampling frames in the sampling pool;

[0030] Step 3, repeating step 2, after the number of sampled frames in the sampling pool meets the preset number of frames, taking the sampled frames in the sampling pool as the key frames.

[0031] According to a multi-keyframe redirection method provided by the present invention, the method further includes:

[0032] It is determined whether there is a frame semantics missing situation between the key frames. If so, a key frame is added between the key frames with the frame semantics missing situation according to the frame semantics information of the key frames with the frame semantics missing situation.

[0033] According to a multi-keyframe redirection method provided by the present invention, after obtaining the source keyframe sequence of the source object, the method further includes:

[0034] Receive a first input, wherein the first input includes an operation of determining a character animation video scene type, wherein the character animation video scene type is an animation video scene type corresponding to a current redirection task;

[0035] In response to the first input, the multi-keyframe retargeting model required for the current retargeting task is obtained according to the character animation video scene type.

[0036] The present invention also provides a method for constructing a multi-keyframe redirection model, the method for constructing a multi-keyframe redirection model in any of the multi-keyframe redirection methods described above, comprising:

[0037] Constructing a global keyframe conversion model according to sample source keyframes and sample animation keyframes corresponding to each of the sample source keyframes;

[0038] According to the global key frame conversion model, obtaining the sample standard animation key frame corresponding to the sample source key frame;

[0039] Constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames;

[0040] Calculate the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame;

[0041] Based on the radial basis function, a key frame residual model is constructed according to the mapping relationship between the key frame residual and the deviation value;

[0042] The multi-key frame redirection model is obtained through the global key frame conversion model and the key frame residual model.

[0043] The present invention also provides a multi-keyframe redirection system, comprising:

[0044] A source key frame acquisition module, used to acquire a source key frame sequence of a source object;

[0045] A multi-keyframe redirection module, used for redirecting each source keyframe in the source keyframe sequence to a corresponding character animation keyframe based on a multi-keyframe redirection model, so as to obtain a character animation video sequence corresponding to the source object;

[0046] Wherein, the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model;

[0047] The global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames;

[0048] The key frame residual model is constructed based on the distance difference between the sample source key frame and other sample source key frames of each frame, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame.

[0049] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any one of the multi-keyframe redirection methods described above, or the multi-keyframe redirection model construction method described above.

[0050] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements any of the multi-keyframe redirection methods described above, or the multi-keyframe redirection model construction method described above.

[0051] The present invention also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned multi-keyframe redirection methods or the above-mentioned multi-keyframe redirection model construction method.

[0052] The multi-keyframe redirection method, multi-keyframe redirection model construction method and system provided by the present invention redirect each source keyframe in a source keyframe sequence to a corresponding character animation keyframe through a multi-keyframe redirection model, and obtain a character animation video sequence corresponding to the source object, thereby effectively avoiding the semantic deviation between the source object and the animation character and improving the accuracy of the keyframe redirection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A schematic diagram of a flow chart of a multi-keyframe redirection method provided by the present invention;

[0055] Figure 2 A schematic diagram of the process of constructing a multi-keyframe redirection model provided by the present invention;

[0056] Figure 3 A schematic diagram of the structure of a multi-keyframe redirection system provided by the present invention;

[0057] Figure 4 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] In the retargeting technology, firstly, special motion capture equipment, such as sensors and cameras, is used to record the actor's body movements, postures, expressions and other information, and convert them into digital data, so as to obtain the action sequence of the real actor's performance. Then, the captured action sequence is applied to the specified virtual character or animated character to ensure the naturalness and realism of the action on the character. In the field of virtual humans, retargeting technology can be used to create more realistic characters and make their movements more vivid and natural. By interpolating or deforming between key frames, the smooth action transition and morphological changes of the characters can be achieved, making them look more real. In animation production, retargeting technology can help animators create smooth animation effects more effectively, reduce the workload of manually drawing or adjusting frames, and ensure the smoothness and realism of animation transitions. In game production, retargeting technology can improve the dynamic performance of game characters and objects. By implementing the retargeting algorithm in the game engine, the character's movements can be made more vivid, and real-time animation deformation effects can be achieved at runtime, which improves the interactivity and entertainment of the game.

[0060] In current redirection technology, Tpose or Apose is usually used as a key frame. For example, if the Tpose of a real actor and the Tpose of an animated character are known, a transformation model T can be solved (the transformation model corresponds to the scene type. As a global transformation model, the posture of the real actor can be redirected to the animated character of the specified scene type). For any input posture of the real actor, the corresponding posture of the animated character can be solved through this transformation model T.

[0061] However, existing redirection technologies have great limitations. This is mainly because there are inconsistencies between the bone lengths and bone templates of real actors and animated characters. If redirection is performed only through a global transformation model T, there will be semantic inconsistencies in the postures between the real actors and animated characters. For example, when a real actor puts his hands together, the two hands of the animated character may not touch each other due to bone length issues. Or, when a real actor walks, the landing point is always at the same height, but after mapping, one foot of the animated character will be higher than the other, resulting in a relatively large deviation in the action semantics between the real actor and the animated character.

[0062] In order to solve the problem of inconsistent action semantics between real actors and animated characters caused by the current single standard frame, the present invention adopts multiple standard frames, that is, a global transformation model T is dispersed into multiple local areas, so as to model each key frame separately, and then realize the redirection of multiple key frames through the trained model.

[0063] Figure 1 A schematic diagram of the process flow of the multi-keyframe redirection method provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides a multi-keyframe redirection method, comprising:

[0064] Step 101: Obtain a source key frame sequence of a source object.

[0065] Acquiring the source key frame sequence generated by a real person (i.e., the source object, which is described in terms of a person image in the present invention) is a key step in the redirection technology, which involves extracting important data for redirection from the data collected by motion capture. Specifically, firstly, the motion sequence of a real person is captured by motion capture acquisition equipment, such as a sensor suit, a camera array, etc., and converted into digital motion data. Since the acquired motion capture data may contain a large amount of information, it is necessary to go through data processing steps to extract the source key frame sequence, including data cleaning, noise reduction, posture recognition and other steps, to ensure that the extracted data is accurate, complete and meets the requirements.

[0066] Furthermore, in the processed motion data, it is necessary to identify key frames, i.e., the most representative and important frames in the motion sequence. These key frames usually correspond to turning points, peaks or other important motion stages in the motion, and are crucial for redirection. In redirection technology, determining key frames can be accomplished by analyzing and processing the motion sequence. When determining key frames, key frames can be selected by analyzing the importance of the motion, such as the start, end or turning point of the motion, to highlight the important parts of the motion.

[0067] Furthermore, source key frame sequences are extracted from the processed data. These source key frame sequences will be used as references in the redirection process. They contain important motion information generated by real people during the motion capture process and are the basis for redirection to virtual characters or animated characters, ensuring that the movements of the redirected animated characters are consistent with those of the real people.

[0068] Step 102, based on a multi-keyframe redirection model, redirect each source keyframe in the source keyframe sequence to a corresponding character animation keyframe, to obtain a character animation video sequence corresponding to the source object;

[0069] Wherein, the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model;

[0070] The global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames;

[0071] The key frame residual model is constructed based on the distance difference between the sample source key frame and other sample source key frames of each frame, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame.

[0072] In the present invention, a multi-keyframe redirection model is used to redirect the actions of a real person to a virtual character or an animated character. The model uses multiple sample source keyframes and their corresponding sample animation keyframes as training data, and is constructed through a global keyframe conversion model and keyframe residuals. It should be noted that in the present invention, according to the actual redirection task requirements, the multi-keyframe redirection model can be constructed based on sample keyframes of different animation video scene types, so that the source keyframes can generate animation videos of corresponding scene styles through the multi-keyframe redirection models corresponding to different animation video scene types.

[0073] In the present invention, the sample source keyframe is a keyframe sequence extracted from the motion capture data of a real person, and the sample animation keyframe is a keyframe in the animation sequence of the corresponding virtual character or animation character. The global keyframe conversion model is a conversion model for converting source keyframes into animation keyframes, which describes the overall conversion relationship between source keyframes and animation keyframes. The keyframe residual is used to describe the residual information between the sample source keyframe and other sample source keyframes. In the present invention, the keyframe residual is converted into the deviation of the animation keyframe through the mapping relationship between the deviation value between the sample animation keyframe and the corresponding sample standard animation keyframe and the keyframe residual, and then applied to the animation keyframe. During the training process, the multi-keyframe redirection model will try to minimize the difference between the sum of the sample standard animation keyframe and the residual and the sample animation keyframe to ensure that the difference between the real character action and the character animation after redirection is minimized.

[0074] Furthermore, after the model training is completed, the multi-keyframe redirection model can redirect each frame in the source keyframe sequence to the corresponding character animation keyframe, including converting the source keyframe using the global keyframe conversion model and fine-tuning according to the keyframe residuals to make the redirected action more accurate.

[0075] The multi-keyframe redirection method provided by the present invention redirects each source keyframe in a source keyframe sequence to a corresponding character animation keyframe through a multi-keyframe redirection model, and obtains a character animation video sequence corresponding to the source object, thereby effectively avoiding the semantic deviation between the source object and the animation character and improving the accuracy of the keyframe redirection result.

[0076] Based on the above embodiment, after obtaining the source key frame sequence of the source object, the method further includes:

[0077] Receive a first input, wherein the first input includes an operation of determining a character animation video scene type, wherein the character animation video scene type is an animation video scene type corresponding to a current redirection task;

[0078] In response to the first input, the multi-keyframe retargeting model required for the current retargeting task is obtained according to the character animation video scene type.

[0079] In the present invention, for different redirection task requirements, the request operation of the animation video scene type initiated by the user end is first obtained, including different animation styles (such as cartoon, realism, science fiction, etc.), different action types (such as running, jumping, dialogue, etc.), different character features (such as humans, animals, robots, etc.), etc. For each scene type, a different multi-keyframe redirection model needs to be used. After determining the model required for this redirection task, the model needs to be configured and prepared to ensure that it can be applied to the current redirection task, including setting the input and output format of the model, loading the weights and parameters of the model, and performing necessary initialization and preparation.

[0080] Based on the above embodiment, the global key frame conversion model is obtained by the following steps:

[0081] The global key frame conversion models corresponding to different sample animation video scene types are obtained according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames.

[0082] Based on the above embodiment, the key frame residual model is obtained by the following steps:

[0083] Obtaining a sample standard animation key frame corresponding to the sample source key frame through the global key frame conversion model;

[0084] Constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames;

[0085] Calculate the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame;

[0086] Based on the radial basis function, the key frame residual model is constructed according to the mapping relationship between the key frame residual and the deviation value.

[0087] In the present invention, in order to solve the defects caused by the single standard frame in the current redirection technology, a transformation relationship T is obtained for each frame by selecting sample key frames of real actors (i.e., sample source key frames) and animation character key frames with the same semantics as these sample source key frames (i.e., sample animation key frames). new . Further, based on each key frame, this transformation relationship T new Modeled in the form of standard keyframes and keyframe residuals.

[0088] Specifically, in the present invention, when training a multi-keyframe redirection model, sample keyframes under any type of animation video scene are used for illustration. First, a data set of sample animation video scene types needs to be collected, including sample source keyframes and their corresponding sample animation keyframes. Then, for each sample, features need to be extracted from the sample source keyframes and sample animation keyframes, and these features may include information such as the position and posture of key points. Next, a representative sample source keyframe and its corresponding sample animation keyframe are selected from the sample. In the present invention, computer vision technology (such as tracking algorithms) can be used to align the sample source keyframes and sample animation keyframes, or they can be selected through human experience to ensure that their spatial positions and postures are as similar as possible, and then an optimization algorithm (such as minimizing the mean square error) is used to calculate the conversion model or conversion function from the sample source keyframe to the sample animation keyframe, that is, the global keyframe conversion model.

[0089] In the present invention, for each frame, a global keyframe conversion model is used to map the coordinate points of the sample source keyframe to the space of the sample animation keyframe, thereby obtaining the sample standard keyframe. Furthermore, the radial basis function (RBF) is used to model the keyframe residual term. The reason why the radial basis function is selected in the present invention is mainly that the radial basis function can make the data points with very small deviations pass approximately when modeling, that is, inputting the source keyframe of the first frame will obtain the corresponding animation keyframe of the first frame, and inputting the source keyframe of the second frame will obtain the corresponding animation keyframe of the second frame, and there will be no deviation in the position of the keyframe. At the same time, the radial basis function has a good local modeling characteristic, and can be modeled separately in different local areas. It can be understood that a residual is dispersed into multiple local areas, and the modeling of each local area has very weak influence on each other.

[0090] In the present invention, for each frame of the sample source key frame, the distance difference between it and other sample source key frames is calculated, wherein the distance difference can use the Euclidean distance or other distance metrics to calculate the distance between two key frames. Further, for each frame, the distance difference is used as the input of the radial basis function, and the sample animation key frame residual corresponding to the sample source key frame of each frame is calculated by weighted summing the distance difference. In the present invention, for the key frame residual corresponding to each frame, a kernel function (such as a Gaussian kernel function) is first determined, and the hyperparameters in the function can be adjusted according to actual needs (fixed after adjustment), and the distance function in the kernel function can use the Euclidean distance, and the distance function is used to calculate the distance difference between the current sample source key frame and each other sample source key frame, and then a model parameter to be solved is used as a weight coefficient. The distance difference is weighted summed to obtain the animation key frame residual corresponding to the current frame.

[0091] Furthermore, after a plurality of sample keyframes are given, the deviation value between the sample animation keyframe of each frame and the corresponding sample standard keyframe is calculated, and then a mapping relationship between the deviation value of each frame and the corresponding keyframe residual is established, so that a keyframe residual model is constructed based on the radial basis function and the established mapping relationship. In the application scenario of multi-keyframe redirection, the current frame is input to the multi-keyframe redirection model. The model first converts the source keyframe into a character standard keyframe according to the global keyframe conversion model. At the same time, based on the radial basis function, the model parameters and distance differences corresponding to the frame are weightedly summed to obtain the character keyframe residual. Finally, the character standard keyframe and the character keyframe residual are added to obtain the animation keyframe corresponding to the source keyframe, so that the keyframe redirection result is more accurate.

[0092] On the basis of the above embodiment, the step of obtaining the global key frame conversion model corresponding to different sample animation video scene types according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames includes:

[0093] Acquire a first sample standard posture and a second sample standard posture, wherein the first sample standard posture is a standard posture corresponding to the sample source key frame, and the second sample standard posture is a standard posture corresponding to the sample animation key frame;

[0094] Based on the sample animation video scene type, the corresponding global key frame conversion model is calculated according to the first sample standard posture and the second sample standard posture.

[0095] In the present invention, Tpose and Apose are commonly used in animation production and character modeling. Tpose refers to a standardized pose in which the character's arms are straightened and form a capital letter "T" shape with the body, which is used as the starting pose for character modeling, making the modeling process easier and ensuring consistency during the animation process. Apose is a character's arms straightened and slightly downward in the shape of a capital letter "A", which is used to express specific actions or emotions.

[0096] After determining the standard postures of sample source key frames and sample animation key frames, the present invention calculates a global key frame conversion model according to different sample animation video scene types, realizes unified representation and alignment of animation key frames, and improves the efficiency and quality of animation production.

[0097] On the basis of the above embodiment, constructing the key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames includes:

[0098] Based on the distance function, the distance difference between the sample source key frame of each frame and other sample source key frames is calculated;

[0099] According to the distance difference, a key frame residual corresponding to the sample source key frame of each frame is constructed.

[0100] Euclidean distance refers to the straight-line distance between two points in Euclidean space, while Manhattan distance is the sum of the absolute distances between two points in the coordinate system. In the present invention, it is used to measure the similarity or difference between samples. Further, based on the calculated Euclidean distance or Manhattan distance, the key frame residual corresponding to the sample source key frame of each frame can be constructed. This residual represents the difference between the current key frame and other sample source key frames, that is, the relative deviation between samples.

[0101] Specifically, in the present invention, the radial basis function It can be defined as:

[0102]

[0103] in, As the kernel function, the present invention adopts the Gaussian kernel function, that is, , is a hyperparameter (can be adjusted according to actual needs and needs to be fixed after adjustment), r is the distance function, i Indicates i Frame, the present invention uses the Euclidean distance, that is, .

[0104] Furthermore, givenn After sample keyframes, a system of equations can be established to solve the unknowns , the equation group is specifically:

[0105]

[0106]

[0107]

[0108]

[0109] Solving the above equations After that, the multi-keyframe redirection model based on radial basis function can be trained. For any actor pose A, the multi-keyframe redirection direction corresponding to his character pose C can be expressed as:

[0110]

[0111] On the basis of the above embodiment, before acquiring the global key frame conversion model corresponding to different sample animation video scene types according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames, the method further includes:

[0112] Based on the farthest point sampling algorithm, the frame sequence is sampled to obtain the corresponding key frames in the frame sequence, wherein the frame sequence includes a sample source frame sequence and / or a sample animation frame sequence.

[0113] The farthest point sampling algorithm is a method for selecting representative samples from a given data set. By calculating the shortest distance from each point to other points and then selecting the farthest point as a representative sample, it can ensure that the selected key frame has the greatest representativeness while reducing redundant information as much as possible. In the present invention, based on the farthest point sampling algorithm, the sample source frame sequence and the sample animation frame sequence are sampled to obtain the corresponding key frames in these frame sequences. These key frames are representative key time points selected from the original sequence to avoid excessive concentration of sampling near certain frames. In the present invention, the frame numbers between the source end and the character end correspond one to one, that is, the frame number is obtained by sampling at the source end, and the same frame number is used for sampling at the character end to obtain a sample source-character key frame pair; the frame number can also be obtained by sampling at the character end, and the source end uses the same frame number for sampling to obtain a sample source-character key frame pair; it can also be sampled simultaneously based on the frame number between the source end and the character end to obtain a sample source-character key frame pair.

[0114] On the basis of the above embodiment, the farthest point sampling algorithm is used to sample the frame sequence to obtain the corresponding key frame in the frame sequence, including:

[0115] Step 1, randomly sampling in the frame sequence, and adding the randomly sampled frames as sampling frames to a sampling pool;

[0116] Step 2, obtaining a matching frame from the remaining frames of the frame sequence, and adding the matching frame to the sampling pool, wherein the matching frame is a frame having the largest minimum distance to the sampling frames in the sampling pool;

[0117] Step 3, repeating step 2, after the number of sampled frames in the sampling pool meets the preset number of frames, taking the sampled frames in the sampling pool as the key frames.

[0118] In the present invention, the frame sequence is first randomly sampled, that is, a frame is randomly selected from the frame sequence as a sampling frame, and the randomly sampled frames are added to the sampling pool. The purpose of random sampling is to include a certain degree of randomness in the sampling process to cover the entire frame sequence as much as possible and avoid being too concentrated on certain specific time points.

[0119] Then, in step 2, the frames with the maximum and minimum distances from the sampled frames in the sampling pool are selected from the remaining frames of the frame sequence as matching frames. This distance can be calculated based on the similarity or distance metric between the frames. In the present invention, the frame farthest from the existing frames in the sampling pool is selected as the matching frame. Finally, step 2 is repeated until the number of sampled frames in the sampling pool reaches the preset number of frames. Once the number of sampled frames in the sampling pool reaches the preset number of frames, the sampled frames in the sampling pool are used as the final key frames. Through this process, the present invention can ensure that the selected key frames are evenly distributed in time and are representative, and can effectively capture the important features and actions in the animation sequence.

[0120] Based on the above embodiment, the method further includes:

[0121] It is determined whether there is a frame semantics missing situation between the key frames. If so, a key frame is added between the key frames with the frame semantics missing situation according to the frame semantics information of the key frames with the frame semantics missing situation.

[0122] In the present invention, both source frames and animation frames may need to define corresponding semantics in the early stage. These semantic definitions are usually to ensure the consistency and logic of the frames in terms of vision and function. For source frames, the semantic definition may involve scenes, objects, actions, etc. in the real world. For example, in virtual reality applications, source frames may refer to captured video frames of the real world, and semantic analysis and definition of scenes, objects, actions, etc. in the video are required.

[0123] For animation frames, semantic definitions may involve characters, scenes, actions, special effects, etc. For example, in animation production, it is necessary to define the character's action semantics, the semantic elements of the scene (such as trees, buildings, etc.), and possible special effects (such as explosions, smoke, etc.).

[0124] Furthermore, when defining these semantics, markup languages, metadata, annotations or specific data structures are usually used to record and describe them. These semantic definitions play an important role in subsequent data processing, analysis, rendering and interaction, and can help ensure that the final presented content meets the design and requirements.

[0125] In the present invention, it is first necessary to check the frame sequence between adjacent key frames to determine whether there is a frame semantic missing situation, wherein the frame semantic missing is the presence of animation frames in the animation sequence that are not fully expressed by the character.

[0126] For key frames with missing frame semantics, frame semantic information is needed to determine the missing frame content. Frame semantic information may include the content of the animation scene, the character's action, the position of the object and other information, which can be obtained by analyzing the frame content around the key frame.

[0127] After determining the location where the frame semantics are missing and the corresponding frame semantic information, it is necessary to add key frames at the corresponding locations. The present invention can be implemented through a variety of methods. For example, the frame with the largest difference is selected from multiple frames with deviations in performance, and the multi-key frame redirection model is trained again using the frame as the input of the model, so that the model can better redirect the source frame to the character frame.

[0128] The present invention can effectively supplement key frames by judging the missing situation of frame semantics and analyzing the frame semantic information, thereby ensuring the integrity and coherence of the key frame sequence.

[0129] Figure 2 A schematic diagram of the process of constructing a multi-keyframe redirection model provided by the present invention is shown in FIG. Figure 2 As shown, the present invention provides a method for constructing a multi-keyframe redirection model, and the method is used to construct a multi-keyframe redirection model in the multi-keyframe redirection method described in the above embodiments, including:

[0130] Step 201, constructing a global key frame conversion model according to sample source key frames and sample animation key frames corresponding to each of the sample source key frames;

[0131] Step 202, obtaining a sample standard animation key frame corresponding to the sample source key frame according to a global key frame conversion model;

[0132] Step 203, constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames;

[0133] Step 204, calculating and obtaining the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame;

[0134] Step 205, constructing a key frame residual model based on a radial basis function and a mapping relationship between the key frame residual and the deviation value;

[0135] Step 206: Obtain the multi-key frame redirection model through the global key frame conversion model and the key frame residual model.

[0136] In the present invention, in order to solve the defects caused by the single standard frame in the current redirection technology, a transformation relationship T is obtained for each frame by selecting sample key frames of real actors (i.e., sample source key frames) and animation character key frames with the same semantics as these sample source key frames (i.e., sample animation key frames). new . Further, based on each key frame, this transformation relationship T new The model is modeled in the form of standard keyframes and keyframe residuals. Specifically, when training a multi-keyframe redirection model, a sample keyframe under any animation video scene type is used for illustration.

[0137] First, it is necessary to collect a dataset of sample animation video scene types, including sample source keyframes and their corresponding sample animation keyframes. Then, for each sample, it is necessary to extract features from the sample source keyframes and sample animation keyframes. These features may include information such as the position and posture of key points. Next, a representative sample source keyframe and its corresponding sample animation keyframe are selected from the sample, and the sample source keyframe and the sample animation keyframe are aligned using computer vision technology (such as tracking algorithms), or selected through human experience to ensure that their spatial positions and postures are as similar as possible, and then an optimization algorithm (such as minimizing the mean square error) is used to calculate the conversion model or conversion function from the sample source keyframe to the sample animation keyframe, that is, the global keyframe conversion model.

[0138] For each frame, a global keyframe conversion model is used to map the coordinate points of the sample source keyframe to the space of the sample animation keyframe, thereby obtaining the sample standard keyframe. Furthermore, the radial basis function is used in the modeling of the keyframe residual term. The main reason why the radial basis function is selected in the present invention is that the radial basis function can make the data points with very small deviations pass approximately when modeling, that is, the source keyframe of the first frame input will obtain the corresponding animation keyframe of the first frame, and the source keyframe of the second frame input will obtain the corresponding animation keyframe of the second frame, and there will be no deviation in the position of the keyframe. At the same time, the radial basis function has a good local modeling characteristic, and can be modeled separately in different local areas. It can be understood that a residual is dispersed into multiple local areas, and the modeling of each local area has very weak mutual influence.

[0139] In the present invention, for each frame of the sample source key frame, the distance difference between it and other sample source key frames is calculated, wherein the distance difference can use the Euclidean distance or other distance metrics to calculate the distance between two key frames. Further, for each frame, the distance difference is used as the input of the radial basis function, and the sample animation key frame residual corresponding to the sample source key frame of each frame is calculated by weighted summing the distance difference. In the present invention, for the key frame residual corresponding to each frame, a kernel function (such as a Gaussian kernel function) is first determined, and the hyperparameters in the function can be adjusted according to actual needs (fixed after adjustment), and the distance function in the kernel function can use the Euclidean distance, and the distance function is used to calculate the distance difference between the current sample source key frame and each other sample source key frame, and then a model parameter to be solved is used as a weight coefficient. The distance difference is weighted summed to obtain the animation key frame residual corresponding to the current frame.

[0140] Furthermore, after given multiple sample key frames, the deviation value between the sample animation key frame of each frame and the corresponding sample standard key frame is calculated, and then a mapping relationship between the deviation value of each frame and the corresponding key frame residual is established, so that based on the radial basis function, a key frame residual model is constructed according to the established mapping relationship.

[0141] The multi-keyframe redirection model construction method provided by the present invention redirects each source keyframe in a source keyframe sequence to a corresponding character animation keyframe through a multi-keyframe redirection model, and obtains a character animation video sequence corresponding to the source object, thereby effectively avoiding the semantic deviation between the source object and the animation character and improving the accuracy of the keyframe redirection result.

[0142] The multi-key frame redirection system provided by the present invention is described below. The multi-key frame redirection system described below and the multi-key frame redirection method described above can be referred to each other.

[0143] Figure 3 The schematic diagram of the structure of the multi-keyframe redirection system provided by the present invention is as follows: Figure 3 As shown, the present invention provides a multi-keyframe redirection system, including a real keyframe acquisition module 301 and a multi-keyframe redirection module 302, wherein the source keyframe acquisition module 301 is used to redirect each source keyframe in the source keyframe sequence to a corresponding character animation keyframe based on a multi-keyframe redirection model, so as to obtain a character animation video sequence corresponding to the source object; wherein the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model; the global keyframe conversion model is constructed based on multiple sample source keyframes and sample animation keyframes corresponding to each of the sample source keyframes; the keyframe residual model is constructed based on the distance difference between the sample source keyframe of each frame and other sample source keyframes, and the deviation value between the sample animation keyframe and the corresponding sample standard animation keyframe.

[0144] In the present invention, the source key frame acquisition module 301 captures the motion sequence of a real person through a motion capture acquisition device, such as a sensor suit, a camera array, etc., and converts it into digital motion data. Since the acquired motion capture data may contain a large amount of information, the source key frame acquisition module 301 extracts the source key frame sequence through data processing steps, including data cleaning, noise reduction, posture recognition and other steps, to ensure that the extracted data is accurate, complete and meets the requirements.

[0145] Furthermore, in the processed motion data, the source key frame acquisition module 301 needs to identify key frames, i.e., the most representative and important frames in the motion sequence. These key frames usually correspond to turning points, peaks or other important motion stages in the motion, which are crucial for redirection. In the redirection technology, determining key frames can be accomplished by analyzing and processing the motion sequence. When determining key frames, key frames can be selected by analyzing the importance of the motion, such as the start, end or turning point of the motion, to highlight the important parts of the motion.

[0146] Furthermore, the source key frame acquisition module 301 extracts source key frame sequences from the processed data. These source key frame sequences will be used as references in the redirection process. They contain important action information generated by the real person during the motion capture process and are the basis for redirection to a virtual character or an animated character, ensuring that the actions of the redirected animated character are consistent with those of the real person.

[0147] In the present invention, the multi-keyframe redirection model in the multi-keyframe redirection module 302 is used to redirect the actions of a real person to a virtual character or an animated character. The model uses multiple sample source keyframes and their corresponding sample animation keyframes as training data, and is constructed through a global keyframe conversion model and keyframe residuals. It should be noted that in the present invention, according to the actual redirection task requirements, the multi-keyframe redirection model can be constructed based on sample keyframes of different animation video scene types, so that the source keyframes can generate animation videos of corresponding scene styles through the multi-keyframe redirection models corresponding to different animation video scene types. Among them, the sample source keyframes are keyframe sequences extracted from the motion capture data of real people, and the sample animation keyframes are keyframes in the animation sequence of the corresponding virtual character or animated character. The global keyframe conversion model is a conversion model for converting source keyframes into animation keyframes, which describes the overall conversion relationship between source keyframes and animation keyframes. The keyframe residual is used to describe the residual information between the sample source keyframe and other sample source keyframes. In the present invention, the keyframe residual is converted into the residual of the animation keyframe through the mapping relationship between the deviation value between the sample animation keyframe and the corresponding sample standard animation keyframe and the keyframe residual, and then applied to the animation keyframe. During the training process, the multi-keyframe redirection model will try to minimize the difference between the sum of the sample standard animation keyframe and the residual and the sample animation keyframe, so as to ensure that the difference between the real character action and the character animation after redirection is minimized.

[0148] Furthermore, after the model training is completed, the multi-keyframe redirection model in the multi-keyframe redirection module 302 can redirect each frame in the source keyframe sequence to the corresponding character animation keyframe, including using the global keyframe conversion model to convert the source keyframe and fine-tuning according to the keyframe residuals to make the redirected action more accurate.

[0149] The multi-keyframe redirection system provided by the present invention redirects each source keyframe in a source keyframe sequence to a corresponding character animation keyframe through a multi-keyframe redirection model, and obtains a character animation video sequence corresponding to the source object, thereby effectively avoiding the semantic deviation between the source object and the animation character and improving the accuracy of the keyframe redirection results.

[0150] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0151] Figure 4 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor (Processor) 401, a communication interface (Communications Interface) 402, a memory (Memory) 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404. The processor 401 may call the logic instructions in the memory 403 to execute the above method, which includes: obtaining a source key frame sequence of a source object; based on a multi-key frame redirection model, redirecting each source key frame in the source key frame sequence to a corresponding character animation key frame, and obtaining a character animation video sequence corresponding to the source object; wherein the multi-key frame redirection model includes a global key frame conversion model and a key frame residual model; the global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames; the key frame residual model is constructed based on the distance difference between the sample source key frame of each frame and other sample source key frames, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame;

[0152] Or, construct a global key frame conversion model based on sample source key frames and sample animation key frames corresponding to each of the sample source key frames; obtain sample standard animation key frames corresponding to the sample source key frames based on the global key frame conversion model; construct the key frame residuals corresponding to the sample source key frames of each frame based on the distance difference between the sample source key frames of each frame and other sample source key frames; calculate the deviation value between the sample animation key frames of each frame and the corresponding sample standard animation key frames; based on the radial basis function, construct a key frame residual model according to the mapping relationship between the key frame residuals and the deviation values; obtain the multi-key frame redirection model through the global key frame conversion model and the key frame residual model.

[0153] In addition, the logic instructions in the above-mentioned memory 403 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0154] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer, the computer can execute the above methods, the method comprising: obtaining a source key frame sequence of a source object; based on a multi-key frame redirection model, redirecting each source key frame in the source key frame sequence to a corresponding character animation key frame, to obtain a character animation video sequence corresponding to the source object; wherein the multi-key frame redirection model includes a global key frame conversion model and a key frame residual model; the global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames; the key frame residual model is constructed based on the distance difference between the sample source key frame of each frame and other sample source key frames, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame;

[0155] Or, construct a global key frame conversion model based on sample source key frames and sample animation key frames corresponding to each of the sample source key frames; obtain sample standard animation key frames corresponding to the sample source key frames based on the global key frame conversion model; construct the key frame residuals corresponding to the sample source key frames of each frame based on the distance difference between the sample source key frames of each frame and other sample source key frames; calculate the deviation value between the sample animation key frames of each frame and the corresponding sample standard animation key frames; based on the radial basis function, construct a key frame residual model according to the mapping relationship between the key frame residuals and the deviation values; obtain the multi-key frame redirection model through the global key frame conversion model and the key frame residual model.

[0156] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to execute the methods provided by the above embodiments, the method comprising: obtaining a source key frame sequence of a source object; based on a multi-key frame redirection model, redirecting each source key frame in the source key frame sequence to a corresponding character animation key frame, and obtaining a character animation video sequence corresponding to the source object; wherein the multi-key frame redirection model comprises a global key frame conversion model and a key frame residual model; the global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames; the key frame residual model is constructed based on the distance difference between the sample source key frame of each frame and other sample source key frames, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame;

[0157] Or, construct a global key frame conversion model based on sample source key frames and sample animation key frames corresponding to each of the sample source key frames; obtain sample standard animation key frames corresponding to the sample source key frames based on the global key frame conversion model; construct the key frame residuals corresponding to the sample source key frames of each frame based on the distance difference between the sample source key frames of each frame and other sample source key frames; calculate the deviation value between the sample animation key frames of each frame and the corresponding sample standard animation key frames; based on the radial basis function, construct a key frame residual model according to the mapping relationship between the key frame residuals and the deviation values; obtain the multi-key frame redirection model through the global key frame conversion model and the key frame residual model.

[0158] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0159] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-keyframe redirection method, characterized in that: include: Get the source key frame sequence of the source object; Based on a multi-keyframe redirection model, each source keyframe in the source keyframe sequence is redirected to a corresponding character animation keyframe to obtain a character animation video sequence corresponding to the source object; Wherein, the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model; The global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames; The key frame residual model is constructed based on the distance difference between the sample source key frame and other sample source key frames of each frame, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame; the sample standard animation key frame is obtained by mapping the coordinate points of the sample source key frame to the space of the sample animation key frame through the global key frame conversion model.

2. The multi-keyframe redirection method according to claim 1, characterized in that: The global key frame conversion model is obtained by the following steps: The global key frame conversion models corresponding to different sample animation video scene types are obtained according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames.

3. The multi-keyframe redirection method according to claim 2, characterized in that: The key frame residual model is obtained by the following steps: Obtaining a sample standard animation key frame corresponding to the sample source key frame through the global key frame conversion model; Constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames; Calculate the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame; Based on the radial basis function, the key frame residual model is constructed according to the mapping relationship between the key frame residual and the deviation value.

4. The multi-keyframe redirection method according to claim 2, characterized in that: The acquiring the global key frame conversion model corresponding to different sample animation video scene types according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames includes: Acquire a first sample standard posture and a second sample standard posture, wherein the first sample standard posture is a standard posture corresponding to the sample source key frame, and the second sample standard posture is a standard posture corresponding to the sample animation key frame; Based on the sample animation video scene type, the corresponding global key frame conversion model is calculated according to the first sample standard posture and the second sample standard posture.

5. The multi-keyframe redirection method according to claim 3, characterized in that: The step of constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames includes: Based on the distance function, the distance difference between the sample source key frame of each frame and other sample source key frames is calculated; According to the distance difference, a key frame residual corresponding to the sample source key frame of each frame is constructed.

6. The multi-keyframe redirection method according to claim 3, characterized in that: Before acquiring the global key frame conversion model corresponding to different sample animation video scene types according to the sample source key frames and the sample animation key frames corresponding to each of the sample source key frames, the method further includes: Based on the farthest point sampling algorithm, the frame sequence is sampled to obtain the corresponding key frames in the frame sequence, wherein the frame sequence includes a sample source frame sequence and / or a sample animation frame sequence.

7. The multi-keyframe redirection method according to claim 6, characterized in that: The sampling of the frame sequence based on the farthest point sampling algorithm to obtain the corresponding key frame in the frame sequence includes: Step 1, randomly sampling in the frame sequence, and adding the randomly sampled frames as sampling frames to a sampling pool; Step 2, obtaining a matching frame from the remaining frames of the frame sequence, and adding the matching frame to the sampling pool, wherein the matching frame is a frame with the maximum and minimum distances from the sampling frames in the sampling pool; Step 3, repeating step 2, after the number of sampled frames in the sampling pool meets the preset number of frames, taking the sampled frames in the sampling pool as the key frames.

8. The multi-keyframe redirection method according to claim 7, characterized in that: The method further comprises: It is determined whether there is a frame semantics missing situation between the key frames. If so, a key frame is added between the key frames with the frame semantics missing situation according to the frame semantics information of the key frames with the frame semantics missing situation.

9. The multi-keyframe redirection method according to claim 1 or 2, characterized in that: After obtaining the source key frame sequence of the source object, the method further includes: Receive a first input, wherein the first input includes an operation of determining a character animation video scene type, wherein the character animation video scene type is an animation video scene type corresponding to a current redirection task; In response to the first input, the multi-keyframe retargeting model required for the current retargeting task is obtained according to the character animation video scene type.

10. A method for constructing a multi-keyframe redirection model, characterized in that: The construction method is used to construct a multi-keyframe redirection model in the multi-keyframe redirection method according to any one of claims 1 to 9, comprising: Constructing a global keyframe conversion model according to sample source keyframes and sample animation keyframes corresponding to each of the sample source keyframes; According to the global key frame conversion model, obtaining the sample standard animation key frame corresponding to the sample source key frame; Constructing a key frame residual corresponding to the sample source key frame of each frame according to the distance difference between the sample source key frame of each frame and other sample source key frames; Calculate the deviation value between the sample animation key frame and the corresponding sample standard animation key frame of each frame; Based on the radial basis function, a key frame residual model is constructed according to the mapping relationship between the key frame residual and the deviation value; The multi-key frame redirection model is obtained through the global key frame conversion model and the key frame residual model.

11. A multi-keyframe redirection system, characterized in that: include: A source key frame acquisition module, used to acquire a source key frame sequence of a source object; A multi-keyframe redirection module, used for redirecting each source keyframe in the source keyframe sequence to a corresponding character animation keyframe based on a multi-keyframe redirection model, so as to obtain a character animation video sequence corresponding to the source object; Wherein, the multi-keyframe redirection model includes a global keyframe conversion model and a keyframe residual model; The global key frame conversion model is constructed based on multiple sample source key frames and sample animation key frames corresponding to each of the sample source key frames; The key frame residual model is constructed based on the distance difference between the sample source key frame and other sample source key frames of each frame, and the deviation value between the sample animation key frame and the corresponding sample standard animation key frame; the sample standard animation key frame is obtained by mapping the coordinate points of the sample source key frame to the space of the sample animation key frame through the global key frame conversion model.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-keyframe redirection method according to any one of claims 1 to 9 or the multi-keyframe redirection model construction method according to claim 10 is implemented.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-keyframe redirection method according to any one of claims 1 to 9 or the multi-keyframe redirection model construction method according to claim 10 is implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-keyframe redirection method according to any one of claims 1 to 9 or the multi-keyframe redirection model construction method according to claim 10 is implemented.

Images