Video editing processing method

Abstract

The invention provides a video editing processing method. The video editing processing method comprises the following steps: identifying and marking required materials through a specified identification mode; the data formats according to the identification marks are arranged and compiled into record texts; selecting a data format in the compiling record text to execute editing storage; and sorting into a training text according to the existing recorded text, and executing into a finished video through a training text code. According to the video editing processing method, the purpose of artificial intelligence video production can be achieved, and automatic batch generation of one type of videos is realized, so that the video production efficiency is improved, the video production cost is reduced, and the requirement of how to automatically process basic video editing processing work is met.

Description

Technical Field

[0001] The present invention relates to the technical field of video processing, and in particular to a video editing and processing method. Background Art

[0002] It takes a lot of time for video media workers to use traditional Adobe and Vegas video editing software to produce a series of exquisite videos. For example, it takes at least one week for a B station UP host to repeatedly polish an exquisite MAD audio and video, and the storage and reuse of basic materials is also a problem. Although the production of movie videos similar to Douyin is more streamlined, the labor cost of the video production team is also a considerable expense. Therefore, in order to reduce the time, manpower and financial costs of video creation, how to realize the automatic batch generation of a type of video is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In view of the above, the present invention provides a video editing and processing method to achieve the purpose of artificial intelligence video production, realize automatic batch generation of a class of videos, thereby improving video production efficiency, reducing video production costs, and solving the need for how to automate basic video editing and processing work.

[0004] The technical solution of the present invention:

[0005] The present invention provides a video editing method.

[0006] Main step S1: identifying and marking the required material by a specified identification method;

[0007] Main step S2: compiling the data format according to the identification mark into a record text;

[0008] Main step S3: Select the data format in the compiled record text to perform clipping and storage;

[0009] Main step S4: organize the existing recorded text into training text, and execute the training text code to become a finished video.

[0010] The video editing processing method provided by the present invention has significant beneficial effects in the field of video production, which are specifically embodied in the following aspects:

[0011] 1. Improve video editing efficiency: By using recognition methods to automatically identify and mark video materials, the time cost of manual screening and classification is greatly reduced, making the video editing process more efficient. At the same time, the organization and application of recording text and training text further speeds up the speed of editing storage and finished video generation, improving overall editing efficiency.

[0012] Second, enhance the accuracy of video production: the video editing processing method of the present invention can accurately identify key information in the material and accurately organize it by recording text. This ensures that during the editing process, the selected data format is more accurate and the editing effect is more in line with the creator's expectations.

[0013] 3. Reduce production costs: Since the present invention realizes the automation and intelligence of video editing, it reduces the degree of manual participation, thus significantly reducing the cost of video production. In addition, the efficient editing process also means that more video production tasks can be completed in a shorter time, further improving production efficiency.

[0014] 4. Optimizing video quality: Through the video editing processing method of the present invention, the editing operation can be accurately performed, avoiding errors and omissions that may occur in traditional methods. At the same time, the application of training text makes the production of the finished video more refined and personalized, thereby improving the overall quality of the video.

[0015] 5. Improve user experience: For video creators and software application designers, the video editing processing method of the present invention not only improves work efficiency, but also brings them a better user experience through precise control and optimized quality. For the final audience, they can also enjoy more smooth, professional and high-quality video content.

[0016] Standardize and split each link of video media production, and record the process of video production. Use recognition, recording and editing technology to computerize the human process. According to the scene and actual needs, continuously fit and improve the entire processing and manufacturing process to realize the artificial intelligence production video assembly line.

[0017] Through the material library and training text, basic video editing and processing such as basic editing and synthesis, music addition, text, and image splicing can be automatically completed. The labor cost and capital cost related to basic work can be greatly reduced. At the same time, the data structure is convenient for storage and reuse, so as to maximize the use of time advantages to handle basic video editing and processing work. It enables media designers to focus more on creation and high-end special effects design, improve the quality of works, and generate high-quality works in batches.

[0018] The preferred embodiments of the present invention and their beneficial effects will be further described in detail in conjunction with specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but should not constitute a limitation of the present invention.

[0020] Figure 1It is the main flow chart of the video editing processing method of the present invention.

[0021] Figure 2 The present invention provides a flowchart of a video editing method according to the first embodiment.

[0022] Figure 3 This is a flowchart of the Naraka scene recognition process for this robot's visual image recognition.

[0023] Figure 4 A detailed illustration of how the Naraka battle scene is identified by features.

[0024] Figure 5 This is the recorded text of the material identification mark in Example 2.

[0025] Figure 6 This is the entire flow chart of generating clipping data for the second embodiment. DETAILED DESCRIPTION

[0026] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0027] In order to allow video creators and software application designers to better understand the technical solution of the present invention, the following will describe in detail, clearly and completely the technical and functional execution processes in the use cases of the present invention from the three aspects of identification, recording and editing, combined with examples and corresponding drawings. Of course, the described process is only a part of the use cases of the present invention, not all of the use cases. Based on the use cases in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0028] See also Figure 1 The present invention provides a video editing method, which specifically comprises the following steps:

[0029] Main step S1: marking the required material by identification method;

[0030] Main step S2: compiling the data format according to the identification mark into a record text;

[0031] Main step S3: Select the data format in the compiled record text to perform clipping and storage;

[0032] Main step S4: organize the existing recorded text into training text, and execute the training text code to become a finished video.

[0033] Each main step is explained in detail below.

[0034] Main step S1: marking the required material by identification method.

[0035] The main step S1 includes:

[0036] Sub-step S11: Identification method classification.

[0037] There are two ways to identify. The first is to automatically identify and record scenes based on obvious scene features through artificial intelligence recognition technologies such as image feature analysis. The second is to manually divide the required scenes for recording based on actual needs.

[0038] Sub-step S12: technical means of identification.

[0039] There are two ways to identify. The first is to use artificial intelligence recognition technologies such as image feature analysis to store and record the pictures, videos, time and corresponding information data of scenes that meet the characteristics. The second is to use tags and scene requirements to store and record the pictures, videos, time and corresponding information data of scenes that meet the required characteristics.

[0040] Main step S2: compile the data format according to the identification mark into a record text.

[0041] After all identification systems are marked, they will be organized into record texts according to the standard data format. The main step S2 includes:

[0042] Sub-step S21: Data format recording of related materials.

[0043] The video name, required duration, storage tag and tag name, execution level, source video source, and storage output method are organized into a standard data format model.

[0044] Sub-step S22: Arrange the data format and write it into the record text.

[0045] The identified and stored recorded data is sorted and matched according to the data format pattern of the record text to become the same format and input into the record text.

[0046] Sub-step S23: record the text for classification.

[0047] Classify the recorded text according to the scene and content, for example, material recorded text, music recorded text, source file recorded text.

[0048] Main step S3: Select the data format in the compiled record text to perform clipping and storage.

[0049] The main step S3 comprises:

[0050] Sub-step S31: Data format recording of related materials.

[0051] Extract the required data from the recorded text, and further compile the recorded data into executable clipping instructions according to the data interface and data format of the code.

[0052] For example, there is a line of data in the record text, including the video name, the required time, the storage tag and tag name, the execution level, the source video source, and the storage output path.

[0053] Sub-step S32: the conversion process of data format recording and editing instructions.

[0054] Compile into execution editing instructions: "Video name" + "Source video source" is converted to "Source video path" + "Video name" + ".mp4" and named "Input video source path".

[0055] The "required time length" is converted into (start time, end time) and named "clip data".

[0056] "Storage tag and tag name" + "Storage output path" is converted to "Storage output path" + "Storage tag folder" + "Storage tag name" + "Corresponding tag number" + ".mp4" and named "Output video storage path".

[0057] "Execution degree" is converted into "whether this data has been executed, if it has been executed, mark yes and then do not execute again, otherwise execute and then mark yes" and named "execution degree".

[0058] Sub-step S33: Edit the execution process of the instruction.

[0059] The editing instructions are input video source path, clip data, output video storage path, and execution level.

[0060] Automatic pipeline operation is performed according to the compilation instructions. The process is to clip the video data from the input video source path, then output the video storage path, and mark the execution degree.

[0061] Main step S4: organize the existing recorded text into training text, and execute the training text code to become a finished video.

[0062] The recorded text is further processed into the required finished product data for editing and storage. The required data is selected from the existing recorded text and combined to form a training text. For example, the recorded text includes a material recorded text, and the material file and music file are obtained from the music recorded text. The training text selects and further edits the material data in the material recorded text, and the training text selects the music file in the music recorded text.

[0063] The main step S4 comprises:

[0064] Sub-step S41: Data editing format of finished video.

[0065] The material data and music files, together with the file name and some notes, are organized into training text.

[0066] For example, the material data is Gilgamesh 1 (10, 20) [this means the start time is 10 seconds and the end time is 20 seconds], Gilgamesh 3 (24, 40) .... Gilgamesh 19 (12, 21), the music file is fakeit.mp3, the file name is The Legend of Gilgamesh, the output file path is G:\\video storage, note that this is the 24th training text, and this is the editing data structure of a training text.

[0067] Sub-step S42: Editing and synthesizing process of the finished video.

[0068] The edited data structure of the training text is encoded and read, and then the code is executed in sequence.

[0069] First, edit and merge Gilgamesh 1, 3...19 in the material data in this order, as (10,20), (24,40)...(12,21), and mix the music file: fakeit.mp3, use the file name: The Legend of Gilgamesh.mp4, store and then output the file path to G:\\video storage.

[0070] This completes the compilation and synthesis of a training text into a finished training text. Next, we will introduce it in detail through the Naraka and FATE animation series.

[0071] Example 1 (Naraka e-sports game)

[0072] Figure 2 This is a flow chart of a video editing method provided by the first embodiment of the present invention. This case is applicable to marking and recording video data according to the specific game scenes of e-sports games, and editing and classifying the data after sorting. This method is implemented in the form of software.

[0073] Main step S1: marking the required material by identification method.

[0074] This scene uses artificial intelligence visual analysis and recognition technology to extract the features of the Naraka battle scene and automatically identify and mark it. The following will explain the corresponding process in detail.

[0075] The main step S1 includes:

[0076] Sub-step S11A: Naraka scene segmentation.

[0077] Naraka's recorded videos generally have five types of scenes: login standby scene, game loading and entering the battlefield waiting scene, game scene, game battle scene, and game end scene. Therefore, we only need to identify the game battle scene and the game scene, extract the game battle scene to distinguish it from other types of scenes. At this time, the game battle scene is called the feature scene that needs to be identified, called the identification feature scene. Then, the entire video is framed into a video screen, and the video screen is analyzed based on the identification features to see if it is a game battle scene. When the identified features meet the judgment criteria and are judged as yes, it is recorded. The judgment criteria are that a white blood bar of a specific size range appears in a specific area and a specified judgment word (such as still alive, soul, and remaining number of soul returns) appears in the upper left corner. If the above features are met, the judgment criteria are met.

[0078] In order to enable those skilled in the art to better understand the present invention, in this case, a portion of Naraka game screen recording will be used as experimental material for illustration.

[0079] This technology is based on robot visual image recognition and is an analysis mode for specific scenes in e-sports videos. The recognition method is to select all pictures that meet a type of feature, and add feature recognition basis and conditional fitting for visual recognition selection to the robot vision. For example, "a type of feature" is the appearance of a white blood bar of a specific size in a specific area and the appearance of a specified judgment word in the upper left corner (such as still alive, soul, and remaining number of soul returns).

[0080] Sub-step S12A: A method for recording identification marks of battle scenes.

[0081] Figure 3 This is a flowchart of the Naraka scene recognition process for this robot's visual image recognition.

[0082] The characteristics of image visual recognition are whether it belongs to a combat scene, whether it has entered a combat state, and strengthening combat state analysis.

[0083] Strengthening the battle status analysis refers to whether there is a red blood bar of a specific size range in the picture. This definition is the basis for strengthening the recognition results.

[0084] The technical solution is the image feature analysis algorithm for specific scenes and the image analysis tool of Opencv.

[0085] First of all, this recognition system is an identification solution for Naraka's game videos.

[0086] The basic features are combat scene recognition, whether to enter combat status, and strengthening combat status analysis.

[0087] The battle scene recognition method is whether there are "AA" and "BB" in the upper left corner, indicating that the game battle scene has been entered, and if there are "ACAC" and "DEDE" in the upper left corner, it means that the battle scene has not been entered in the preparation stage, among which AA, BB, ACAC, and DEDE are specific words.

[0088] How to judge whether to enter the combat state? In the combat state, a white health bar will appear on the character. The white RBG recognition area is between [x1, x2, x3] and [y1, y2, y3], and the size range is between [a1, b1] and [a2, b2] and between [a3, b3] and [a4, b4]. The unit size is px. The following features are recognized by OpenCV's inrange, boundingRect, and rectangle. Whether there is a combat standard is recognized within the range of (width*w1, height*h1) to (width*w2, height*h2).

[0089] The corresponding parameter explanations above are: Width is the full screen width, Height is the full screen height, X1, x2, x3 and y1, y2, y3 are two groups of parameters of rbg, [a1, b1], [a2, b2], [a3, b3], [a4, b4] are four groups of length and width categories, for example, [20, 30] is 20px long, 30px wide, and the unit size is px, w1, w2, h1, h2 are the scaling ratios of height and width, for example, w1=0.8h1=0.4 means that the width is 80% of the total and the height is 40% of the total.

[0090] How to strengthen the battle status analysis, whether there is a red blood bar icon of the enemy in the whole picture, the enemy has a red blood bar, the red rbg recognition area is between [x1,x2,x3] and [y1,y2,y3], the size range is between [a1,b1] to [a2,b2] and between [a3,b3] to [a4,b4], the unit size is px, through opencv's inrange,boundingRect,rectangle to identify the following features, and identify whether there is a battle standard within the scope of the entire picture.

[0091] The recognition features are summarized as whether the words "AA" and "BB" appear in the picture, whether a white blood bar appears in a specific category in the picture, and whether a red blood bar appears in the picture.

[0092] Therefore, after meeting the requirements of battle scene recognition and image recognition for entering combat status, the frame number is identified and the image and related information are saved.

[0093] Figure 4 A detailed illustration of how battle scenes are identified by features.

[0094] The pseudo code corresponding to step S1 is introduced as follows:

[0095] Video_nara.py

[0096] Function introduction:

[0097] Read the video file, capture the frame number and pictures, use opencv to get some eigenvalue pictures, and determine whether the eigenvalue pictures meet the selection criteria of the scene. If they meet the criteria, save them, otherwise delete them.

[0098] pseudocode:

[0099]

[0100]

[0101]

[0102]

[0103] Main step S2: compile the data format according to the identification mark into a record text.

[0104] After automatic feature recognition, the image, the corresponding data format and the corresponding source file name will be saved, and the data format of the above data materials will be further compiled into a record text. The following will explain the corresponding process in detail.

[0105] The main step S2 comprises:

[0106] Sub-step S21A: The characteristic picture data information is recorded and stored as characteristic data.

[0107] When the video screen is identified as a game battle scene, the feature screen is extracted and saved in a feature storage file. The video screen is stored in a battle scene folder and stored with a label.

[0108] For example, assuming that the video name is 2024-4-10naraka.mp4, the feature storage file is the feature picture of 2024-4-10naraka, the battle scene folder is the battle scene of 2024-4-10naraka, the 2889th frame of the video picture that meets the battle scene is stored as 2024-4-10naraka2889.jpg, and the 2889th frame of the feature picture has 5 pictures that meet the requirements, namely 2889+0.jpg, 2889+1.jpg, 2889+2.jpg, 2889+3.jpg, and 2889+4.jpg.

[0109] The feature storage file (for example: 2889+0.jpg) in the above files is used to identify whether the features meet the requirements and optimize the training text again. The battle scene file (for example: 2024-4-10naraka2889.jpg) is used to record whether the video screen belongs to the battle scene, which is convenient for recording and storage. In this process, if the parameters on the recognition data are not perfect, further learning and correction of parameters are carried out according to the recognized stored feature pictures.

[0110] Sub-step S22A: compile the feature data into a specific data format.

[0111] Assuming that 30 frames are 1 second, the 2889th frame is the 96th second. The number of frames is compiled into seconds and recorded according to the length of the source file. The recording format is as follows: the video name is 2024-4-10naraka.mp4, the video path is G: / NARAKA / source video / , and the battle scene storage path is G: / NARAKA / storage video / 2024-4-10naraka (automatically saved as the video name) / battle clip / .

[0112] The target file of the label text is the battle scene label file, which connects and distinguishes continuous and discontinuous label images. The basis for determining whether it is continuous or discontinuous is that if the gap between the two images is more than 10 frames, it is discontinuous, and if it is less than 10 frames, it is continuous. The format of the label text is as follows:

[0113] Snippet Start file name End file name First battle scene 2024-4-10naraka750.jpg 2024-4-10naraka3250.jpg Second battle scene 2024-4-10naraka5700.jpg 2024-4-10naraka7200.jpg The third battle scene 2024-4-10naraka9000.jpg 2024-4-10naraka12600.jpg Fourth battle scene 2024-4-10naraka15000.jpg 2024-4-10naraka16500.jpg

[0114] Compile according to 30 frames per second. Take the first battle scene as an example to compile the recorded text. The video name is 2024-4-10naraka.mp4. Remove the ".mp4" in the video name and keep "2024-4-10naraka" as the marker text. Remove the ".jpg" and marker text of the start and end file names, leaving only "750" and "3250". Convert "750" of the start file name to a positive integer and round it down to an even number that divides 30. Lower(int(750) / 30)=25 is defined as the start time (seconds). Convert "3250" of the end file name to a positive integer and round it up to an even number that divides 30. Lower(int(750) / 30)=25 is defined as the start time (seconds).

[0115] The target files of the material video cropping text are the material video and the source video. This format is used to crop and store the compiled battle scene clips, the start time and the end time of the source video. The format of the material video cropping text is as follows:

[0116] Snippet Start time (seconds) End time (seconds) First battle scene 25 109 Second battle scene 190 240 The third battle scene 300 420 Fourth battle scene 500 550

[0117] Sub-step S23A: Arrange the data format and corresponding materials into training text.

[0118] The training text for simulated synthesis effects is to distinguish battle scenes by different source videos, heroes, weapons, and modes, and to generate different finished product effects by adding different background music and audio. It can also merge all related battle materials without distinguishing these factors to generate a total battle scene collection. The basic structure of the training text for simulated synthesis effects is material clip, start time, duration, remarks, corresponding theme, music file, file name, and source of material file. By further processing the start time and duration of the material clip, recording the relevance through remarks, adding music files, indicating the source path of the material file, and storing it in the corresponding theme with the file name.

[0119] The reprocessing of the synthetic video text is to extract a part of the existing simulated synthetic effect training text, and further refine it according to different attributes, such as different source videos, different heroes, different weapons, and different modes, and differentiate the required materials from the relatively rough total collection, and extract the finished product with higher processing degree and better viewing experience according to the needs of refinement. The reprocessing of the synthetic video text is to extract a part of the editing content from the simulated synthetic effect training text, so the relevance and source of the content extraction are added, and the basic structure is consistent with the record structure.

[0120] Music text and audio text record the required music, the related information of the used music and audio, and the used materials for easy repeated use.

[0121] Main step S3: Select the data format in the compiled record text to perform clipping and storage.

[0122] After the record text is generated, the record text will record the path and processing information of the material file and the relevant required data information, and the data extracted according to the compiled record text data format will be edited and stored to obtain the corresponding basic material. The following will explain the corresponding process in detail.

[0123] The main step S3 comprises:

[0124] Sub-step S31A: Basic material clipping and storage method.

[0125] After the required scenes in the source video are identified and marked, the format of the source video and the label text is recorded, and then the text is compiled using an operator to become the basic material video clipping text. A corresponding storage space is established for the source file and the corresponding material file for editing and storage.

[0126] Sub-step S32A: The process of clipping and storing basic materials.

[0127] The video file in the video path (G: / NARAKA / source video / 2024-4-10naraka.mp4) is processed according to the recorded text and stored in the battle scene storage path (G: / NARAKA / storage video / 2024-4-10naraka (automatically saved as video name) / battle clip / ).

[0128] Edit the video files in the video path according to the recorded text, and clip them into the first battle scene.mp4, the second battle scene.mp4, the third battle scene.mp4, and the fourth battle scene.mp4, and then store them in the battle scene storage path (G: / NARAKA / Storage Video / 2024-4-10naraka(automatically saved as video name) / Battle Clip / ), for example G: / NARAKA / Storage Video / 2024-4-10naraka(automatically saved as video name) / Battle Clip / First Battle Scene.mp4.

[0129] Main step S4: organize the existing recorded text into training text, and execute the training text code to become a finished video.

[0130] Different recording texts record different materials and files. Select the required recording text, extract the corresponding material data, design the training text, and then obtain the corresponding data through the training text. Compile and execute the code in the corresponding data format to generate the required finished video. The following will explain the corresponding process in detail.

[0131] The main step S4 comprises:

[0132] Sub-step S41A: Basic production process of finished video.

[0133] Choose to automatically synthesize all battle scenes, so that the finished product can be fully automated from recognition to recording, and finally to editing and synthesis.

[0134] Further use the simulated synthesis effect training text to extract the required files in the battle scene, taking the first battle scene.mp4 and the fourth battle scene.mp4 as examples, add the background music all my people, adjust the contrast volume: the volume of BGM and the battle scene are consistent, and finally synthesize: 2024-4-10naraka wonderful battle collection.mp4, stored in G: / NARAKA / Storage Video / 2024-4-10naraka (automatically saved as video name) / .

[0135] Sub-step S42A: Processing and production flow of finished video.

[0136] We can further refine the training texts from these simulated synthesis effects. For example, the content of the first battle scene.mp4 is more relevant to the hero's ultimate move. We only need to extract this clip, add different background music, several textures and texts, and refine it into a highlight collection. The second battle scene.mp4, the third battle scene.mp4, and the fourth battle scene.mp4 are not needed.

[0137] The pseudocode corresponding to step S2, step S3 and step S4 is introduced.

[0138] Video_train.py

[0139] Function introduction:

[0140] Get the corresponding material images, decode and store the compiled information, write the compiled information into the database and Excel through data sorting, and edit, merge and store them through the database and Excel

[0141] pseudocode:

[0142]

[0143]

[0144]

[0145]

[0146] Example 2 (FATE animation series)

[0147] First of all, this technology is based on scene division and is an identification and division mode for specific label scenes of animation themes.

[0148] The Fate anime series is a big IP, with more than 15 works in the collection, and the largest and most complete series of peripherals, games, animations, comics, and secondary creations. Take the Fate anime series and movies as source videos. Classify by tags, further learn the creation methods of film and television works, and combine them with material tags.

[0149] Main step S1: marking the required material by identification method.

[0150] In this scenario, there are many scenes that need to be identified and marked, and there are many source file categories, so manual classification and recognition is used to classify the scenes by labels, manually record the corresponding data information for storage, and then train and learn based on the labels and scenes to generate data text and data sets that can be repeatedly called. The following will explain the corresponding process in detail.

[0151] The main step S1 includes:

[0152] Sub-step S11B: Select and design the recognition mark mode of the animation theme.

[0153] Character recognition tools for anime themes are not yet mature. The scenes are mostly divided by UP hosts and anime fans. Image recognition is difficult, and the themes and label categories are numerous and complex. We adopt the label classification recognition method, use training text to record label divisions, and after training the recognition classification data, repeat the recognition call many times.

[0154] First, divide the target files to be identified and the type structure of the corresponding record generation files, then specify the required labels, design the label format, enter the label format, compare the existing label formats and contents for duplication, and learn to supplement the label format.

[0155] The training-learning process includes:

[0156] 1. The data sets include: source file training text and material training record text.

[0157] 2. Use xlrd to import excel file.

[0158] 3. Use metatransfer.data to establish an input pipeline. Metatransfer, referred to as mt, refers to a code method for editing, recording, and storing metadata of videos, pictures, audio, and text. It is called metatransformation and is used to compile video editing instructions and provide source material information as well as classified storage records. Finally, it generates mt.data for transmission between training texts.

[0159] 4. Use the model correspondence between training texts to map the columns in mt.data to the next training model. The specific process is as follows:

[0160] The source file training text is meta-transformed through mt.data to generate material training record text and material video.

[0161] The material training record text, music text and audio text are transformed through mt.data to generate simulated synthesis effect training text and finished video.

[0162] 5. Use the MT video editing instruction model (MT model or meta-transformation model for short) to execute the construction, generate the video, and classify and store the material files in the video generation storage file. The material files include audio files, video files, picture files, and corresponding texts; train and modify the meta-transformation model and training text according to the effect of the video.

[0163] The metatransfer model is a model that extracts elements from metadata of images, texts, audios, and videos and stores them as data sets, selects required feature series from the data sets, performs matrix transformation, and executes operation commands through matrix transformation. The selected images, texts, audios, and videos are used as basic elements to edit, combine, and split their features, and run them into a video generation analytical model that meets a class of video creation design models.

[0164] The type structures of the identified target files and the corresponding record generated files are source video recording structure, material video division and identification structure, simulated synthetic video design structure, re-processed synthetic video design structure, music and audio recording structure.

[0165] Source video record structure: Source video data records are formed by crawling the source video and folders and reading and entering file information.

[0166] The format of the material video division and recognition structure is as follows:

[0167]

[0168] Clear meaning, used to record the information of a basic material video and expand its related data record set, including the material's file name, start time, end time, tag definition, related notes, whether it has been marked, the tag marking code, whether it meets the editing specifications, and related provision of editing secondary creation works.

[0169] in:

[0170] The file name is the source of the source video file.

[0171] The start time and end time are the time periods selected for the source video file.

[0172] Tags are editorial definitions of the video content and time period for this enumeration.

[0173] Related notes are annotations to the enumerated video content and time periods and attempts to further expand the developed structural style.

[0174] Whether the execution mark has been passed refers to whether the material information has been executed by the operation instruction.

[0175] The execution tag code refers to the Xth material information to be executed in the tag according to the previous tag.

[0176] Whether it complies with the specifications refers to whether the video file exists and whether the start and end times are within the time period of the video file. The related secondary work refers to the name of the theme in which this material video is applied to the simulated composite video design structure.

[0177] The design structure of the simulated composite video includes: corresponding theme, music file, file name, start time, and duration.

[0178] The re-processed synthetic video design structure (re-synthesized structure for short) is similar to the analog synthetic video design structure (re-synthesized structure for short). The difference is that the re-synthesized structure selects similar editing information segments from the analog synthetic structure according to the similarity of themes and styles, and then selects the required materials from the material video in a composite editing mode.

[0179] The music and audio recording structure includes: music storage name, music name, music album, usage text, keynote and style. The usage text refers to the record of the music as background music in the training text. The keynote and style refer to the music's tendency and suitability to serve as background music of this keynote and style.

[0180] Sub-step S12B: Identify the marking process and specific cases.

[0181] The basic process of the identification method is to define a basic tag, record the identification material tag, and store it according to the data editing mode of the material. When writing a new identification record material, if the content is similar to the existing tag record, it will not be entered or supplemented under the existing tag record. The source video file compares the material tag to a similar source video based on the existing material tag. When identifying and recording related similar source videos, the material tag is directly labeled. The original tag is directly used when it is compared with the content required by the source video, and a new tag is added if the original tag does not have it. Use this identification tag recording method to identify and classify the source video materials that meet the needs, correspond these materials to the source video, store them, and then further plan and improve the tag definition.

[0182] For example, there is a character named Altria in Fate, whose job is Saber. She has two identification tags, Altria and Saber. In one of Fate's works, FateUBW, these two tags are used to identify records. She also appears in the two series FateHF and FateZero. Since there are also Bedivere and Nero who have the Saber tag, we still use these two tags to identify her character prototype in these two series, so that she is classified into two material libraries in three series through two tags. In FateHF, she has two new tags, namely battle scene and treasure release, so we add them to the tag library related to her, and also put new tags in the total tag library. At this time, she appears again in the FGO Round Table. At this time, this character has four tags, namely Altria, Saber, battle scene, and treasure release. There are dozens of characters related to these four tags, forming a cross-record marking model of characters, scenes, series, and tags.

[0183] Sub-step S13B: Identify the recording format and form a corresponding recording structure according to different materials.

[0184] The above examples show that the relevant source videos include fate zero, fate UBW, fate grand order, UBW theatrical version, FGO round table, fate HF, etc. Popular element tags include battle scenes, release of treasures, story narration, eyes, posing, etc. Character element tags include Altria, Hero King, Emiya Kiritsugu, Emiya Shirou, Toosaka Rin, etc. Job element tags include saber, lancer, archer, berserker, rider, caster, assassin, etc. Source video episodes and different types of tags form a complete set of recognition record formats, so that the materials related to the main factors are all in this tag recognition network. The recognition record format of the material video is the source video file name, start time, end time, tag, and remarks.

[0185] There are several types of recognition record formats, source video record structure (source video, duration, video quality, notes, video source), material video division recognition structure (file name, start time, end time, label, notes, whether the execution mark has been executed, execution mark code), simulation synthesis video design structure (segment, start time, duration, notes, corresponding theme, music file, file name, source of material file), re-processing synthesis video design structure (segment, start time, duration, notes, corresponding theme, music file, file name, source of material file, corresponding simulation synthesis video text), music and audio record structure (music name, volume adjustment, musician, music album, music lyrics).

[0186] Main step S2: compile the data format according to the identification mark into a record text.

[0187] After designing through the scene, the identification mark is made through the label, and the information data of the identification mark is sorted into different data formats according to the category of the record text and stored in different record texts. The corresponding process will be explained in detail below.

[0188] The main step S2 comprises:

[0189] Sub-step S21B: Record the classification and storage space of the text.

[0190] The synthesis of a video requires recording the source files, editing the source files into materials as needed, selecting the materials according to the required effects, and generating a synthesis workstation for simulated synthesis. Some cases of simulated synthesis effects are overall effects. The overall effect is split into multiple effects to generate them again, and a general synthesis workstation is split into multiple subordinate synthesis workstations, and then the music files are organized into a provided text.

[0191] This type of classified record text not only records the footage and forms multiple calls for one record, but also saves the storage space of the material, turning the video storage space into text storage space. At the same time, it supports multi-person cloud editing, so that the creativity and insights of multiple people are concentrated in a record text, and then a source video material is shared, forming a cloud record text and an original video material. Multiple people and multiple machines extract and generate a large amount of material, and delete some of the space-consuming materials when they are not in use, and run editing when they are needed.

[0192] The type structures of the record-generated files are: source video text record, material video text record, simulated synthetic video text, re-processed synthetic video text, music and audio text, and finished product record text.

[0193] The above file classification space stations include source file space station, material file storage station, synthesis effect workstation, multi-level effect synthesis workstation, finished product display library, and music file storage space.

[0194] Sub-step S22B: Record data format and content and relevance of the record text.

[0195] Video production requires a large number of files of various types, which need to be sorted and classified, and related information needs to be recorded. The key to recording information is to encode and record the required type of target files, and to encode and record the required features and related information. The core lies in the target file, the required features, and the related information. The target file is processed through the features and supplemented and modified with related information.

[0196] The purpose of recording is to classify the required resources to facilitate reading and computer recognition, provide instructions for editing operations, and facilitate further classification, storage and processing.

[0197] The recorded texts classified according to functions and target files are source video recorded text, material video cropped text, simulated synthesis effect training text, re-production training text, music text and audio text.

[0198] The target file of the source video recording text is the source video. The required features are the duration and source of the source video. The related information includes labels and supplementary instructions, as well as whether it has been replaced.

[0199] The target files of the material video cropping text are the material video and the source video. The required features are the tags, the clipping duration of the source video, and the storage method of the material file. The related information is whether it has been clipped, the notes, and the execution markup code.

[0200] The target files of the training text for simulating synthesis effects are the source video and the synthesized video. The required features are the source video and the cut duration of the source video and the name of the synthesized video. The related information is the synthesis theme, music file, work number, source video, and tag of the source video. Creating training text again is similar.

[0201] The target files of music text and audio text are music files and audio files. The required features are the source of the audio file and the synthesis theme corresponding to the music file. The related information is the lyrics, author, album, and which work the music file is used for.

[0202] There are certain connections and correspondences between different recorded texts. The source video recorded text and the material video cropped text are matched through the relevance of the source video, and the verification method is whether (duration) corresponds to (end time). The material video cropped text and the simulated synthesis effect training text are matched through the relevance of the material name, and the verification method is (start time, end time) corresponds to (start time, start time + duration). The simulated synthesis effect training text and the music text and audio text are matched through the relevance of the music name, and the verification method is whether the music name exists.

[0203] Main step S3: Select the data format in the compiled record text to perform clipping and storage.

[0204] After the record text is generated, the record text will record the path and processing information of the material file and the relevant required data information, and the data extracted according to the compiled record text data format will be edited and stored to obtain the corresponding basic material. The following will explain the corresponding process in detail.

[0205] The main step S3 comprises:

[0206] Sub-step S31B: record text data format import.

[0207] The source video recording text imports the information required for the source video into the editing data operation as the source video content standard. The content standard is compared to see whether the file exists, whether the duration is consistent, etc.

[0208] Sub-step S32B: Record the text clipping execution process.

[0209] The material video cropping relationship text reads each piece of material editing information, and then compares it with the source video content standard. If it meets the standard, the material editing information is decoded and executed according to the decoding instructions recorded in the material video text (file name, start time, end time, label, remarks, whether the marking has been executed, execution marking code), and clips are performed according to the start and end time of the file name. The clips are stored in the label classification folder, and it is indicated that the marking has been executed. The storage name is label + execution marking code, and the execution marking code is that this material is the Nth marking code of this type of label that has been executed.

[0210] Figure 5 Record text for material identification tags.

[0211] Main step S4: organize the existing recorded text into training text, and edit the text code to execute it into a finished video.

[0212] During the editing process, the recorded text is cut according to the source video recorded text to the material video, and then to the simulated synthesis effect training text. Finally, a round of material information of the recorded text in the training text is extracted, assisted by the music text and audio text, and the training text of the recorded product is added.

[0213] Through the above recorded texts, the instructions form a self-circulating closed loop to generate the editing data process. The automatic editing data process will be introduced in detail below.

[0214] The main step S4 comprises:

[0215] Sub-step S41B: Basic production process of finished video.

[0216] By extracting a part of the material file, the required art data editing mode is operated according to the specifications of (segment, start time, duration, notes, corresponding theme, music file, file name, source of material file) in the simulation synthesis effect training text, wherein (segment, start time, duration, notes) are compared with the content standard of the material video cropping relationship text. After meeting the content standard, check whether the (music file) is consistent with the music text and audio text standards. If it is consistent, the source of the material file is processed again according to the start time and duration of the segment of the material file, and stored in the synthesis effect workstation. The workstation synthesizes the second-edited material according to the editing sequence in the simulation synthesis effect training text, adds the music file as the mixed BGM, and stores it according to the file name.

[0217] The simulation synthesis effect of the first training is relatively comprehensive or the subject matter is relatively broad. When re-producing, part of the effect is extracted from the first training text as a reference, or a part of the total collection is extracted as the re-production training text, which saves the cost of extracting video text and has a high reuse rate, making it convenient for multiple people to operate and mark.

[0218] Sub-step S42B: Processing and production process of finished video and examples.

[0219] Take the MAD production of the FATE series as an example for analysis. The FATE series is the series with the most works, the largest IP, the most complete two-dimensional creation and comprehensive industry in recent years. We selected the main animation and film works of FATE as the basic material units. According to the popular elements of the MAD video, the labels of the division elements are determined. And the theme elements with the theme and character design designed by ourselves as the core.

[0220] For example, there is a saber-round table in the simulation synthesis effect training text, which is a collection synthesis of saber (134-168) with saber tag record data extracted from the material, and is divided into 7 themes according to the character theme content. Taking two themes as examples, the sacred round table field Camelot-Saber-Gawain and the sacred round table field Camelot-Saber-Mordred, the materials of the sacred round table field Camelot-Saber-Gawain are saber (134, 135, 137, 156, 163), and the materials of the sacred round table field Camelot-Saber-Mordred are saber (139, 140, 160). They are all extracted and edited from the summary of saber-round table, and the music, theme and materials are replaced for further processing. This processing makes it convenient to separate the total collection and short video effects, so that both refined processing and comprehensive design content are covered.

[0221] In the above process, multiple files can be generated in the same batch, and multiple modifications and editing can be performed. The reproducibility is strong, and the command can be automatically executed. The same sequence can generate multiple training text products, and use a material library to generate multiple training texts in the same batch through the editing data process.

[0222] For example, the Sacred Round Table Realm Camelot-Saber-Bedivere, the Sacred Round Table Realm Camelot-Saber-The Lion King, and the Sacred Round Table Realm Camelot-Saber-Gawain are the training texts for three videos. These three videos are generated and processed at the same time, the music and audio are modified, and the editing parameter information is modified. The contents of these three videos are from the same drama, so part of the content of "Bedivere" is copied to "The Lion King" and simple modifications are made. This can supplement and improve the content effect of "The Lion King", and other more relevant content of training texts can also be added to "Gawain" or "Bedivere". By simply modifying and copying the editing parameter information to add and improve the video content, three videos are generated in batches.

[0223] Figure 6 Generate a data flow chart for the entire editing process, which specifically includes: importing the required training text, extracting the required materials and processing the materials according to the data instructions, merging the processed materials in the order of the training text, mixing the merged video with music effects, naming and storing the video according to the naming and storage path of the training text, and outputting the finished video.

[0224] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0225] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A video editing method, characterized in that: Main step S1: identifying and marking the required material by a specified identification method; Main step S2: compiling the data format according to the identification mark into a record text; Main step S3: Select the data format in the compiled record text to perform clipping and storage; Main step S4: organize the existing recorded text into training text, and execute the training text code to become a finished video.

2. The video editing method according to claim 1, characterized in that: In the main step S1, There are two ways of recognition. The first is to use image feature analysis artificial intelligence recognition technology to store and record the pictures, videos, time and corresponding information data of scenes that meet the characteristics. The second is to use tags and scene requirements to store and record the pictures, videos, time and corresponding information data of scenes that meet the required characteristics.

3. The video editing method according to claim 1, characterized in that: In the main step S2, Arrange the video name, required duration, storage tag and tag name, execution level, source video source, storage output method into a standard data format mode; The identified and stored recorded data is sorted and matched according to the data format mode of the recorded text so as to be input into the recorded text in the same format; Classify the recorded text according to the scene and content.

4. The video editing method according to claim 1, characterized in that: In the main step S3, Extract the required data from the recorded text, and further compile the recorded data into executable clipping instructions according to the data interface and data format of the code; The editing instructions are input video source path, editing data, output video storage path, and execution level; Automatic pipeline operation is performed according to the compilation instructions. The process is to clip the video data from the input video source path, then output the video storage path, and mark the execution degree.

5. The video editing method according to claim 1, characterized in that: In the main step S4, The training text is organized from the material data and music files, plus the file name and some notes; The edited data structure of the training text is encoded and read, and then the code is executed in sequence.

6. The video editing method according to claim 1 is applied to Naraka game video processing, characterized in that: In the main step S1, It is only necessary to identify the game battle scene and the game progress scene, extract the game battle scene and distinguish it from other types of scenes. At this time, the game battle scene is called the feature scene that needs to be identified. Then, the entire video is extracted into video images, and the video images are analyzed based on the identification features to determine whether they are game battle scenes. When the identified features meet the judgment criteria and are judged as yes, they are recorded. The judgment criteria are that a white blood bar of a specified size appears in a specified area and a specified judgment word appears in the upper left corner; The robot adopts the recognition and analysis mode for the specified scenes of e-sports videos based on the robot visual image recognition. The recognition method is to select all the pictures that meet a certain type of features, and add the feature recognition basis and conditional fitting selected by the robot vision. The first type of features is the appearance of a white blood bar of a specified size in the specified area and the appearance of a specified judgment word in the upper left corner. The characteristics of image visual recognition are whether it belongs to a combat scene, whether it has entered a combat state, and strengthening combat state analysis; Strengthening the battle status analysis refers to whether there is a red health bar of a specified size in the picture. This definition is the basis for strengthening the identification of the battle status results; After meeting the requirements of battle scene recognition and image recognition for entering combat status, the frame number is identified and the image and related information are saved.

7. The video editing method according to claim 6, characterized in that: The main step S2 comprises: Sub-step S21A: recording and storing characteristic picture data information as characteristic data; Sub-step S22A: compiling the feature data into a specified data format; Sub-step S23A: Arrange the data format and corresponding materials into training text.

8. The video editing method according to claim 6, characterized in that: In the main step S3, After the required scenes in the source video are identified and marked, the format of the source video and the label text is recorded, and then the text is compiled using an operator to become the basic material video clipping text. A corresponding storage space is established for the source file and the corresponding material file for editing and storage.

9. The video editing method according to claim 1 is applied to the processing of FATE animation series videos, characterized in that: In the main step S1, The label classification recognition method is adopted. The training text is used to record the label division. After the recognition classification data is trained, the recognition call is repeated many times. First, divide the target files to be identified and the type structures of the corresponding record generation files, then define the required labels, design the label format, enter the label format, compare the existing label formats and contents for duplication, and learn to supplement the label format; The type structures of the identified target files and the corresponding record generated files are source video recording structure, material video division identification structure, simulated synthetic video design structure, re-processed synthetic video design structure, music and audio recording structure.

10. The video editing method according to claim 9, characterized in that: The main step S2 comprises: Sub-step S21B: The synthesis of a video requires recording the source file, editing the source file into materials as needed, selecting the materials according to the required effects, generating a synthesis workstation for simulated synthesis, some cases of simulated synthesis effects are overall effects, splitting the overall effect into multiple re-generated effects, splitting a general synthesis workstation into multiple subordinate synthesis workstations, and then organizing the music files into a provided text; The types and structures of the record-generated files are: source video text record, material video text record, simulated synthetic video text, reprocessed synthetic video text, music and audio text, and finished product record text; The above file classification spaces include source file space, material file storage, synthesis effect workstation, multi-level effect synthesis workstation, finished product display library, and music file storage space; Sub-step S22B: The number of files required for video production is numerous and the types are disorderly. They need to be sorted and classified, and the relevant information needs to be recorded. The key to recording information is to encode and record the required type of target files, encode and record the required features and relevant information. The core lies in the target file, the required features, and the relevant information. The target file is processed by the features, and supplemented and modified with the relevant information; The recorded texts classified according to functions and target files are source video recorded text, material video cropped text, simulated synthesis effect training text, re-production training text, music text and audio text.

Images