A method and system for generating model actions based on image recognition

Abstract

The present invention discloses a model action generation method and system based on image recognition in the field of artificial intelligence technology. The model action generation method based on image recognition includes the following steps: S1, a step of constructing an action sample library through character features and action features; S11, a step of obtaining character features and action features based on multiple action videos, and constructing an action sample library through the character features and action features; S12, a step of obtaining the shortest period action through dynamic fitting lines using the action features; the present invention can identify actual pictures through the action sample library and generate action combinations that match the characters in the actual pictures; it can save the workload of staff and is relatively labor-saving; it can display the character actions in the action video in the form of dynamic lines, can obtain the character actions more intuitively, and the actions are relatively concise; it can make the generated action combination more realistic and smooth, and have a better degree of coherence and connection.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for generating model actions based on image recognition. Background Art

[0002] Image recognition and computer vision are important branches in the field of artificial intelligence. In recent years, with the rapid development of deep learning, especially convolutional neural networks, the application scope of image recognition and computer vision has continued to expand. They mainly focus on obtaining information from images or videos and understanding their content. Some model action generation also generates actions by obtaining features through image recognition. Generally, model action generation is also mainly used in film, television, games and other industries.

[0003] In the existing film, television and game industries, staff generally take makeup photos or draw original paintings first, and then design the character's movements based on the makeup photos or original paintings. This results in a large workload, is time-consuming and labor-intensive, and is also prone to poor continuity in the character's movements.

[0004] To this end, we propose a model action generation method and system based on image recognition to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for generating model actions based on image recognition to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for generating model actions based on image recognition.

[0007] A method for generating model actions based on image recognition, comprising the following steps:

[0008] S1. The steps of constructing an action sample library based on character features and action features include:

[0009] S11, obtaining character features and action features based on multiple action videos, and constructing an action sample library based on the character features and action features;

[0010] S12, obtaining the shortest period action by dynamically fitting a line based on the action feature, and dividing the shortest period action into a first action and a second action;

[0011] S13, associating the action features and character features in the same action video, and matching the action features to the associated character feature categories;

[0012] S2, the step of recognizing the actual image and generating the model action, including:

[0013] S21, inputting an actual image into an action sample library, performing character feature recognition on the actual image through the action sample library, determining action features based on the character features, and obtaining multiple action combinations based on the action features;

[0014] S22, obtaining the connection coherence of the first action and the second action in the same action combination, and outputting the action combination that meets the connection coherence requirement;

[0015] Preferably, in said S11, the step of constructing an action sample library includes:

[0016] S111. Acquire multiple action videos under a first preset rule. The action videos must contain both human figures and action elements. Perform frame processing on the action videos. The number of frames for the action videos can be set, generally 24 frames. Continuous framed images are obtained and recorded as a framed image set.

[0017] The first preset rule includes: the action video must be longer than 1 minute, contain human action, and the character's clothing, appearance, body elements and movements must be clear;

[0018] S112, selecting a person from a plurality of framed image sets through image recognition to obtain the selected person, overlaying the selected person with a standard image of a person model with marked body regions, thereby dividing the selected person into body regions; recording position changes of auxiliary points in the body regions, fitting the position changes of the auxiliary points into lines, recording the dynamic fitting lines, and obtaining motion features;

[0019] S113, presetting the same person's features as clothing elements, body elements, and facial elements, identifying and classifying the people in the multiple framed image sets according to the facial elements, clothing elements, and body elements to obtain person feature categories;

[0020] S114: Match the action feature to its associated character feature category.

[0021] By adopting the above technical solution, action features can be obtained through action videos, and character feature categories can be obtained through the three elements of preset character features. Action features can be classified through character feature categories, and then an action sample library can be constructed through action features and character features.

[0022] Preferably, in S112, each picture in the framed picture set is evenly divided into multiple pixel blocks, and the position of each pixel block in the first frame picture is marked, the position of the pixel block changes in each picture is recorded, and the pixel blocks with changed positions in the framed picture set are fitted into lines frame by frame to obtain dynamic fitting lines.

[0023] By adopting the above technical solution, the action features in the action video can be obtained, and the character actions in the action video can be displayed in the form of dynamic lines, so that the character actions can be obtained more intuitively and the actions are relatively concise.

[0024] Preferably, in said S12, the step of obtaining the shortest period action includes:

[0025] S121, assigning values to the dynamic fitting lines corresponding to the motion features to obtain motion parameters of the dynamic fitting lines;

[0026] S122, setting the action parameter as the Y axis and the time corresponding to the action parameter as the X axis to obtain a function curve;

[0027] S123. Select a regular periodic curve from the function curve, determine the shortest function period in the regular periodic curve, save the action characteristics within the function period and the corresponding dynamic fitting line, and obtain the shortest period action.

[0028] By adopting the above technical solution, regular movements in the movement characteristics can be obtained, and the shortest period movement is the main movement in the corresponding movement characteristics, thereby obtaining the regular main movement in the movement characteristics.

[0029] Preferably, in S123, the step of dividing the shortest period action into a first action and a second action includes:

[0030] S1231, selecting multiple shortest-period actions in the same area, and calculating discrete values of action parameters corresponding to the multiple shortest-period actions;

[0031] S1232. Record the shortest period action with an action parameter discrete value less than or equal to a threshold value X as the first action;

[0032] S1233. Record the shortest period action with the action parameter discrete value greater than the threshold value X as the second action;

[0033] S1234: Associating the first action and the second action with the character feature categories in the same action video.

[0034] By adopting the above technical solution, the shortest period actions can be classified into a first action and a second action, which facilitates the subsequent combination of the first action and the second action to generate an action combination.

[0035] Preferably, in said S21, the step of identifying the actual picture includes:

[0036] S211, input the actual picture into the action sample library;

[0037] S212, identifying clothing elements, body elements, and facial elements of the actual image using the action sample library to obtain the intersection of the three elements;

[0038] S213 , determining the corresponding character features through the intersection, obtaining matching action features through the character features, and determining the corresponding first action and second action according to the action features.

[0039] By adopting the above technical solution, the first action and the second action can be determined by identifying the features of the person in the actual picture.

[0040] Preferably, the first action under the same character feature category is recorded as a first action set, and the second action under the same character feature category is recorded as a second action set;

[0041] In said S21, the step of generating the action of the actual picture includes:

[0042] S214, selecting first actions in the first action set that have the same area as the second action, and alternating or replacing multiple first actions with the second actions to form different action combinations;

[0043] S215, obtaining the coherence and connection degree between the first action and the second action in the plurality of action combinations;

[0044] S216: Output action combinations that meet the standard coherence and connection degree. These action combinations are the actions generated by the actual image.

[0045] By adopting the above technical solution, multiple action combinations can be generated through the first action set and the second action set.

[0046] Preferably, in S22, the step of obtaining the degree of coherence and connection between the first action and the second action includes:

[0047] S221, comparing the ending action of the first action in the action combination with the starting action of the second action to obtain a degree of coherence;

[0048] S222: Determine a coherent action with a coherence degree greater than or equal to 70% as a coherent action, a coherence degree less than 70% but greater than 50% as an adjustable action, and a coherence degree less than 50% as an incoherent action;

[0049] S223: Delete the action combinations corresponding to the incoherent actions, and adjust the action combinations corresponding to the adjustable actions.

[0050] By adopting the above technical solution, the coherence and connection of the action combination can be judged, making the generated actions more realistic and smooth.

[0051] Preferably, in said S224, the step of adjusting the first action among the adjustable actions includes:

[0052] S2241, determining the movement change area of the first movement, and recording it as the adjustment area;

[0053] S2242. The angle between the first and second actions of the continuous action within the adjustment area is obtained by dynamically fitting the line, and is recorded as θcoherence. The angle between the first and second actions within the adjustment area is obtained by dynamically fitting the line, and is recorded as θto be measured. Thus, the formula can be obtained:

[0054] θadjustment = θcoherence - θto be measured;

[0055] S2243. If θ is adjusted to a positive number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is increased frame by frame; if θ is adjusted to a negative number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is decreased frame by frame.

[0056] By adopting the above technical solution, the adjustable action can be adjusted to achieve better connection and higher coherence between the adjustable action and the second action, and it is also conducive to outputting more action combinations.

[0057] Preferably, a model action generation system based on image recognition comprises a construction module and a generation module;

[0058] A construction module, which constructs an action sample library through character features and action features; the construction module includes an acquisition unit and a classification storage unit;

[0059] An acquisition unit, which obtains character features and action features based on multiple action videos, constructs an action sample library based on the character features and action features, obtains the shortest period action by dynamically fitting a line based on the action features, and divides the shortest period action into a first action and a second action;

[0060] A classification storage unit associates action features with character features in the same action video and matches the action features to the associated character feature categories;

[0061] The generation module recognizes the actual image and generates the model action; the generation module includes an input recognition unit and an output judgment unit;

[0062] Input recognition unit, input the actual picture into the action sample library, use the action sample library to identify the character features of the actual picture, determine the action features based on the character features, and obtain multiple action combinations based on the action features;

[0063] The output determination unit obtains the connection coherence of the first action and the second action in the same action combination, and outputs the action combination that meets the connection coherence requirement.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] By obtaining action features and preset character features, the action features can be classified according to the character features, and an action sample library can be constructed based on the character features and action features, thereby facilitating the recognition of actual images and generating action combinations that match the characters in the actual images;

[0066] By fitting the pixel blocks with changing positions in the frame image into lines frame by frame, dynamic fitting lines are obtained. The movements of the characters in the action video can be displayed in the form of dynamic lines, which can obtain the movements of the characters more intuitively and concisely.

[0067] By obtaining the shortest cycle action and dividing the shortest cycle action into a first action and a second action, the first action and the second action can be arranged alternately to generate multiple action combinations, which can save the workload of the staff and is more labor-saving and time-saving;

[0068] By judging the degree of coherence between the first action and the second action in the action combination, it is possible to easily generate an action combination that conforms to the actual picture, making the generated action combination more realistic and smooth, and also conducive to outputting more action combinations. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0070] Figure 1 Schematic diagram of the method of the present invention.

[0071] Figure 2 It is a structural flow diagram of the present invention.

[0072] Figure 3 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0073] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention. Example 1

[0074] Reference Figure 1 and Figure 2 The first embodiment further illustrates a method for generating model actions based on image recognition proposed by the present invention.

[0075] A method for generating model actions based on image recognition, comprising:

[0076] Based on multiple action videos, character features and action features are obtained, and an action sample library is constructed based on the character features and action features. The steps of constructing the action sample library include: obtaining multiple action videos under a first preset rule, where the action videos must contain both character and action elements; performing frame processing on the action videos, where the number of frames of the action videos can be set by the user, generally 24 frames, and obtaining continuous framed images, which are recorded as a framed image set;

[0077] Through image recognition, a person in a plurality of framed picture sets is framed to obtain the framed person, and a standard figure model with marked body regions is overlapped with the framed person to divide the framed person into body regions; the position changes of auxiliary points in the body regions are recorded, and the position changes of the auxiliary points are fitted into lines, which are recorded as dynamic fitting lines to obtain motion features; each picture in the framed picture set is evenly divided into multiple pixel blocks, and the position of each pixel block in the first frame is marked, and the position of the pixel block changes in each picture is recorded, and the pixel blocks with position changes in the framed picture set are fitted into lines frame by frame to obtain dynamic fitting lines;

[0078] In practical applications, by fitting the pixel blocks with changing positions in the framed images into lines frame by frame, dynamic fitting lines are obtained. This allows the movements of characters in action videos to be displayed in the form of dynamic lines, making it possible to obtain the movements of the characters more intuitive and concise. Example 2

[0079] Reference Figure 2 The first embodiment further illustrates a method for generating model actions based on image recognition proposed by the present invention.

[0080] The same character features are pre-set to be composed of clothing elements, body elements, and facial elements. Characters in multiple framed image sets are identified and classified according to facial elements, clothing elements, and body elements to obtain character feature categories. Action features are matched to their associated character feature categories. The shortest-period action is obtained by dynamically fitting lines between action features, and the shortest-period action is divided into a first action and a second action. Action features in the same action video are associated with character features, and the action features are matched to their associated character feature categories.

[0081] Assign values to the dynamic fitting lines corresponding to the motion features to obtain motion parameters of the dynamic fitting lines; set the motion parameters as the Y-axis and the time corresponding to the motion parameters as the X-axis to obtain a function curve; select a regular periodic curve in the function curve, determine the shortest function period in the regular periodic curve, save the motion features and the corresponding dynamic fitting lines within this function period, and obtain the shortest period motion;

[0082] Select multiple shortest-period actions in the same area and calculate the discrete values of the action parameters corresponding to the multiple shortest-period actions; record the shortest-period action with the action parameter discrete value less than or equal to the threshold X as the first action; record the shortest-period action with the action parameter discrete value greater than the threshold X as the second action; associate the first and second actions with the feature categories of the person in the same action video;

[0083] In practical applications, by obtaining the shortest cycle action and dividing the shortest cycle action into the first action and the second action, the first action and the second action can be arranged alternately to generate multiple action combinations, which can save the workload of the staff and is more labor-saving. Example 3

[0084] Reference Figure 2 The first embodiment further illustrates a method for generating model actions based on image recognition proposed by the present invention.

[0085] S21, inputting an actual image into an action sample library, performing character feature recognition on the actual image through the action sample library, determining action features based on the character features, and obtaining multiple action combinations based on the action features;

[0086] By inputting an actual image into an action sample library; using the action sample library to identify clothing elements, body elements, and facial elements of the actual image, the intersection of the three elements is obtained; the corresponding character features are determined through the intersection, and the matching action features are obtained through the character features. The corresponding first action and second action are determined based on the action features; the first action under the same character feature category is recorded as the first action set, and the second action under the same character feature category is recorded as the second action set;

[0087] Select the first action that has the same area as the second action in the first action set, and alternate and replace multiple first actions with the second actions to form different action combinations; obtain the connection and coherence of the first action and the second action in the same action combination, and output the action combination that meets the connection and coherence requirements; obtain the connection and coherence of the first action and the second action in multiple action combinations; output the action combinations that meet the standard connection and coherence requirements, and these action combinations are the actions generated by the actual image;

[0088] In actual applications, by obtaining action features and preset character features, action features can be classified according to character features, and an action sample library can be constructed through character features and action features, which can facilitate the recognition of actual pictures and generate action combinations that match the characters in the actual pictures. Example 4

[0089] Reference Figure 2 The first embodiment further illustrates a method for generating model actions based on image recognition proposed by the present invention.

[0090] Compare the ending action of the first action in the action combination with the starting action of the second action to obtain the degree of continuity and coherence; define actions with a degree of continuity and coherence greater than or equal to 70% as continuous actions, define actions with a degree of continuity and coherence less than 70% and greater than 50% as adjustable actions, and define actions with a degree of continuity and coherence less than 50% as incoherent actions; delete the action combinations corresponding to incoherent actions, and adjust the action combinations corresponding to adjustable actions; determine the action change area of the first action, and record it as the adjustment area;

[0091] The angle between the first and second actions of the continuous movement in the adjustment area is obtained by dynamically fitting the line, which is recorded as θcoherence; the angle between the first and second actions in the adjustment area is obtained by dynamically fitting the line, which is recorded as θto be measured. Thus, the formula can be obtained:

[0092] θadjustment = θcoherence - θto be measured;

[0093] If θ is adjusted to a positive number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is increased frame by frame; if θ is adjusted to a negative number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is decreased frame by frame;

[0094] In practical applications, by judging the degree of coherence between the first and second actions in an action combination, it is possible to generate an action combination that conforms to the actual picture, making the generated action combination more realistic and smooth, and also conducive to outputting more action combinations. Example 5

[0095] Reference Figure 2 and Figure 3 The first embodiment further illustrates a method for generating model actions based on image recognition proposed by the present invention.

[0096] A model action generation system based on image recognition, comprising a building module and a generation module;

[0097] A construction module, which constructs an action sample library through character features and action features; the construction module includes an acquisition unit and a classification storage unit;

[0098] An acquisition unit obtains character features and action features based on multiple action videos, constructs an action sample library based on the character features and action features, obtains the shortest period action through dynamic fitting lines of the action features, and divides the shortest period action into a first action and a second action;

[0099] A classification storage unit associates action features with character features in the same action video and matches the action features to the associated character feature categories;

[0100] The generation module recognizes the actual image and generates the model action; the generation module includes an input recognition unit and an output judgment unit;

[0101] Input recognition unit, input the actual picture into the action sample library, use the action sample library to identify the character features of the actual picture, determine the action features based on the character features, and obtain multiple action combinations based on the action features;

[0102] An output determination unit obtains the connection and coherence of the first and second actions in the same action combination, and outputs the action combination that meets the connection and coherence requirements;

[0103] In actual applications, the system can identify actual pictures through the action sample library and generate action combinations that match the characters in the actual pictures; it can save the workload of staff and is relatively labor-saving; it can display the character actions in the action video in the form of dynamic lines, and can obtain the character's actions more intuitively, and the actions are relatively simple; it can make the generated action combinations more realistic and smooth, with better coherence and connection.

[0104] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0105] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A model action generation method based on image recognition, characterized in that: The following steps are involved: S1. The steps of constructing an action sample library based on character features and action features include: S11, obtaining character features and action features based on multiple action videos, and constructing an action sample library based on the character features and action features; S12, obtaining the shortest period action by dynamically fitting a line based on the action feature, and dividing the shortest period action into a first action and a second action; Record the position of pixel blocks that change in each image, fit the pixel blocks that change in position in the framed image set into lines frame by frame, and obtain dynamic fitting lines; assign values to the dynamic fitting lines corresponding to the action features to obtain the action parameters of the dynamic fitting lines; set the action parameters as the Y-axis and the time corresponding to the action parameters as the X-axis to obtain a function curve; select a regular periodic curve in the function curve, determine the shortest function period in the regular periodic curve, save the action features within this function period and the corresponding dynamic fitting lines, and obtain the shortest period action; S13, associating the action features and character features in the same action video, and matching the action features to the associated character feature categories; S2, the step of recognizing the actual image and generating the model action, including: S21, inputting an actual image into an action sample library, performing character feature recognition on the actual image through the action sample library, determining action features based on the character features, and obtaining multiple action combinations based on the action features; S22: Obtain the connection coherence between the first action and the second action in the same action combination, and output the action combination that meets the connection coherence requirement.

2. The method for generating model actions based on image recognition according to claim 1, characterized in that: In said S11, the step of constructing an action sample library includes: S111. Acquire multiple action videos under a first preset rule. The action videos must contain both human figures and action elements. Perform frame processing on the action videos to obtain continuous framed images, which are recorded as a framed image set. S112, selecting a person from a plurality of framed image sets through image recognition to obtain the selected person, overlaying the selected person with a standard image of a person model with marked body regions, thereby dividing the selected person into body regions; recording position changes of auxiliary points in the body regions, fitting the position changes of the auxiliary points into lines, recording the dynamic fitting lines, and obtaining motion features; S113, presetting the same person's features as clothing elements, body elements, and facial elements, identifying and classifying the people in the multiple framed image sets according to the facial elements, clothing elements, and body elements to obtain person feature categories; S114: Match the action feature to its associated character feature category.

3. The method for generating model actions based on image recognition according to claim 2, characterized in that: In the S111, each picture in the framed picture set is evenly divided into multiple pixel blocks, and the position of each pixel block in the first frame picture is marked, the position of the pixel block changes in each picture is recorded, and the pixel blocks with changed positions in the framed picture set are fitted into lines frame by frame to obtain dynamic fitting lines.

4. The method for generating model actions based on image recognition according to claim 3, characterized in that: In said S12, the step of dividing the shortest cycle action into a first action and a second action includes: S121, selecting multiple shortest-period actions in the same area, and calculating discrete values of action parameters corresponding to the multiple shortest-period actions; S122, record the shortest period action with the action parameter discrete value less than or equal to the threshold value X as the first action; S123, record the shortest period action with the action parameter discrete value greater than the threshold value X as the second action; S124: Associating the first action and the second action with the character feature categories in the same action video.

5. The method for generating model actions based on image recognition according to claim 1, characterized in that: In the step S21, the actual picture recognition step includes: S211, input the actual picture into the action sample library; S212, identifying clothing elements, body elements, and facial elements of the actual image using the action sample library to obtain the intersection of the three elements; S213 , determining the corresponding character features through the intersection, obtaining matching action features through the character features, and determining the corresponding first action and second action according to the action features.

6. The method for generating model actions based on image recognition according to claim 1, characterized in that: The first action under the same character feature category is recorded as a first action set, and the second action under the same character feature category is recorded as a second action set; In said S21, the step of generating the action of the actual picture includes: S214, selecting first actions in the first action set that have the same area as the second action, and alternating or replacing multiple first actions with the second actions to form different action combinations; S215, obtaining the coherence and connection degree between the first action and the second action in the plurality of action combinations; S216: Output action combinations that meet the standard coherence and connection degree. These action combinations are the actions generated by the actual image.

7. The method for generating model actions based on image recognition according to claim 1, characterized in that: In the step S22, obtaining the degree of coherence between the first action and the second action includes: S221, comparing the ending action of the first action in the action combination with the starting action of the second action to obtain a degree of coherence; S222: Determine a coherent action with a coherence degree greater than or equal to 70% as a coherent action, a coherence degree less than 70% but greater than 50% as an adjustable action, and a coherence degree less than 50% as an incoherent action; S223: Delete the action combinations corresponding to the incoherent actions, and adjust the action combinations corresponding to the adjustable actions.

8. The method for generating model actions based on image recognition according to claim 7, characterized in that: In the step S222, the step of adjusting the first action among the adjustable actions includes: S2221, determining the movement change area of the first movement, and recording it as the adjustment area; S2222. Determine the angle between the first and second consecutive actions within the adjustment region by dynamically fitting the line, denoted as θcoherence; and determine the angle between the first and second consecutive actions within the adjustment region by dynamically fitting the line, denoted as θtobemeasured. Thus, the formula: θadjustment = θcoherence - θto be measured; S2223. If θ is adjusted to a positive number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is increased frame by frame; if θ is adjusted to a negative number, the angle of the dynamic fitting line of the first action is adjusted, and the angle value of θ to be measured is decreased frame by frame.

9. A model action generation system based on image recognition, applied to the model action generation method based on image recognition according to any one of claims 1 to 8, characterized in that: Includes building blocks and generation modules; A construction module, which constructs an action sample library through character features and action features; the construction module includes an acquisition unit and a classification storage unit; An acquisition unit, which obtains character features and action features based on multiple action videos, constructs an action sample library based on the character features and action features, obtains the shortest period action by dynamically fitting a line based on the action features, and divides the shortest period action into a first action and a second action; A classification storage unit associates action features with character features in the same action video and matches the action features to the associated character feature categories; The generation module recognizes the actual image and generates the model action; the generation module includes an input recognition unit and an output judgment unit; Input recognition unit, input the actual picture into the action sample library, use the action sample library to identify the character features of the actual picture, determine the action features based on the character features, and obtain multiple action combinations based on the action features; The output determination unit obtains the connection coherence of the first action and the second action in the same action combination, and outputs the action combination that meets the connection coherence requirement.

Images