A video metadata labeling method for film and television materials

Abstract

The application discloses a video metadata labeling method for film and television materials, and belongs to the technical field of digital film production. The method of the application is as follows: reading and preprocessing a video to obtain a video frame sequence, determining whether a scene board exists, reading the video frame sequence frame by frame, identifying label data items, and automatically extracting semantic information; mapping the semantic information to the semantic label data items, storing the semantic information to a semantic label field, cleaning and removing redundant data items, and performing associated storage to realize automatic labeling of video semantic labels; reading video metadata, retaining necessary video metadata label fields, appending the semantic label field, and realizing construction of video metadata of film and television materials. Through analysis and processing of film and television materials, the application automatically extracts frame picture semantic information, automatically labels semantic labels related to video content, stores the semantic labels to video metadata of the film and television materials, and improves the searching and management efficiency of the film and television materials.

Description

Technical Field

[0001] This invention relates to a method for annotating video metadata of film and television materials, belonging to the field of digital film production technology. Background Technology

[0002] In the digital film production process, film and television footage shot by digital cameras exists as files on storage media. The metadata of the film and television footage describes basic information related to the shooting, such as focal length, aperture, ISO, resolution, and frame rate. However, the metadata automatically written by the camera is relatively simple. Therefore, a script supervisor is employed on the film set to manually record the script and input the content information of the shot to supplement the metadata of the film and television footage.

[0003] With the increasing scale, industrialization, and efficiency of commercial film production, situations involving multi-camera shooting, rapid scene transitions, and coordination among multiple production departments are common, making manual annotation of basic information inefficient. Furthermore, the emergence of new film production technologies such as virtual shooting has led to more complex and diverse shooting methods, making manual annotation increasingly reliant on the script supervisor's experience and understanding of the shooting content, thus increasing the probability of omissions, duplicates, and errors.

[0004] Currently, post-production processing in the film and television industry supports the classification and management of metadata based on film and television materials to facilitate efficient retrieval and management of these materials during post-production. For example, patent CN111680189A describes a method and apparatus for retrieving film and television content, which generates summary information based on the video entity information of films and television dramas and constructs a film and television map to improve retrieval results. Another example is patent CN106484774A, which describes a method and system for associating multi-source video metadata. This method uses methods such as deduplication, field splitting, and format normalization to establish relationships between video metadata, enabling the extraction of video entity information from films and television works, such as film title, release date, and regional classification. Finally, patent CN109670080A describes a method, apparatus, device, and storage medium for determining film and television tags, using video entity information and audience description keywords as associated tags to serve user film recommendation algorithms. However, the aforementioned film and television content retrieval technologies are primarily based on video entity information such as film title, keywords, summaries, and creators. This content mainly describes the films and television works themselves and cannot meet the retrieval needs of large-scale film and television materials.

[0005] Existing technologies lack a video metadata annotation method that can, to some extent, replace traditional manual recording and automatically extract content information presented in the video, construct video semantic tags, and store them in video metadata to serve the retrieval and management of film and television materials. Summary of the Invention

[0006] The purpose of this invention is to provide a video metadata annotation method for film and television materials. By analyzing and processing film and television materials, extracting semantic information of frames, automatically annotating semantic tags related to video content, and storing them in the video metadata of film and television materials, the method can improve the efficiency of film and television material retrieval and management.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] The present invention provides a video metadata annotation method for film and television materials, comprising the following steps:

[0009] Step 1: Extract semantic information from the frames of the film and television footage, which includes the following sub-steps:

[0010] Step 1.1: Read the video and perform video preprocessing to obtain a video frame sequence. Use object detection methods to determine whether a clapperboard exists in the frame.

[0011] The video preprocessing includes: encoding and converting the video, reducing the resolution and sampling the frames, preserving the semantic information in the video frames, and outputting a video frame sequence;

[0012] The clapperboard refers to a device used during film and television shooting to write and identify the scene number, shot number, and number of takes of the current shooting content. The scene number, shot number, and number of takes are respectively referred to as scene number, shot number, and take number.

[0013] Step 1.2: Read the video frame sequence frame by frame. Using the clapperboard detection and recognition method, locate the clapperboard area in the frame where the clapperboard is present, identify the field number, shot number and sub-shot number marked on the clapperboard, and record the field number, shot number and sub-shot number marked on the clapperboard as the tag data item of the frame.

[0014] Step 1.3: Read the video frame sequence frame by frame, extract the content information appearing in the frame by using the target detection and recognition method in the semantic information extraction module, and store it as semantic tag data items;

[0015] As a preferred option, the data items of the semantic tag field include: actor, pose, shot type, inside / outside, scene, day / night, camera movement, object, and text description;

[0016] Preferably, the target detection and recognition method includes: a face detection and recognition model, a face key point detection model, a human skeleton key point detection model, a scene recognition model, a camera movement mode recognition model, an object detection and recognition model, and an image description generation model.

[0017] Step 2: Label the video with semantic tags based on semantic information, which includes the following sub-steps:

[0018] Step 2.1: Determine the semantic tag fields to be extracted from the video, map the semantic information extracted from the frame to the semantic tag data items, and store them in the semantic tag fields corresponding to each frame.

[0019] Step 2.2: Integrate all semantic tag fields extracted from the frame sequence, filter and analyze semantic tag data items, clean up and remove redundant data items based on inter-frame continuity, and generate valid video semantic tag data items;

[0020] Step 2.3: Associate and store the extracted video semantic tag data items with the corresponding semantic tag fields.

[0021] Step 3: Construct video metadata for film and television materials based on semantic tags, which includes the following sub-steps:

[0022] Step 3.1: Obtain video materials and read video metadata. Perform splitting, deletion, modification, and format normalization operations on the read video metadata, while retaining the necessary video metadata tag fields.

[0023] The film and television materials include videos shot with digital cameras, as well as source videos existing in Internet video websites, video content providers, and media asset storage systems, used for the production of movies, TV series, variety shows, and live online broadcasts;

[0024] Step 3.2: Add semantic tag fields to the read video metadata. Based on the existing metadata tag fields of the film and television materials, add semantic tag fields related to the video content, and write the marked video semantic tags into the video metadata according to the original data format of the current film and television material metadata.

[0025] The semantic tag fields that must be set include scene number, shot number, second shot number, actor, pose, human body key points, shot type, interior / exterior, scene, day / night, camera movement method, object category, and text description.

[0026] If a field fails to retrieve the corresponding data item, it is recorded as N / A;

[0027] Step 3.3: Using each material as a separate unit, integrate the initial metadata obtained in Step 3.1 with the semantic information tags and their corresponding data items obtained in Step 2, and store them in the format of a text file;

[0028] Step 3.4: Convert the metadata to the required format. For the specified non-linear editing software or digital intermediate software, rewrite it into a data structure or tag field that supports its reading and writing, and store it in a video metadata file of a specific format.

[0029] Beneficial effects:

[0030] This invention provides a video metadata annotation method for film and television materials. By analyzing the content information of video frames frame by frame, extracting each data item of its semantic tag field, analyzing and filtering the data items used to construct the video semantic tag field, and combining it with the metadata editing method of film and television materials, the method realizes the reading, writing and generation of video metadata. It can replace the traditional manual recording method, automatically extract the content information in the video frames, construct video semantic tags and store them in the video metadata, improve the efficiency of film and television material retrieval and management, and solve the problem that video tag information cannot meet the needs of large-scale film and television material retrieval. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the annotation system structure corresponding to the video metadata annotation method for film and television materials according to the present invention;

[0032] Figure 2 This is a flowchart illustrating the overall process of a video metadata annotation method for film and television materials according to the present invention.

[0033] Figure 3 This is a flowchart illustrating the method for extracting semantic information in this embodiment;

[0034] Figure 4 This is a flowchart illustrating the video semantic tagging method in the embodiment;

[0035] Figure 5 This is a flowchart illustrating the method for constructing video metadata for film and television materials in this embodiment.

[0036] Figure 6 This is a schematic diagram of the file structure of two types of video metadata in an embodiment;

[0037] Figure 7 This is an example diagram showing the video metadata parsing results of film and television materials used in this embodiment. Detailed Implementation

[0038] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and examples.

[0039] Example 1:

[0040] like Figure 1 As shown in the embodiment, the annotation system of the video metadata annotation method for film and television materials of the present invention includes: a semantic information extraction module M10, a video semantic annotation module M20, and a video metadata construction module M30;

[0041] The semantic information extraction module M10 is used to extract semantic information based on the frames of film and television materials;

[0042] In this embodiment, by collecting a semantic information database related to film and television content, the corresponding model required in the semantic information extraction module M10 is trained, specifically including:

[0043] (1) By training with the clapperboard image dataset, a clapperboard detection and recognition model is constructed. Image processing is performed on various types of clapperboard images to obtain standard brightness clapperboard images without perspective and distortion. The corresponding positions of field lens information in the clapperboard images are labeled, and a clapperboard image segmentation model is trained and constructed.

[0044] (2) Train and build a handwritten character recognition model using a handwritten character dataset;

[0045] (3) Using actor face datasets, train and build face detection and recognition models and face key point detection models;

[0046] (4) Train and build a human skeleton key point detection model using actor motion dataset;

[0047] (5) Train and build a scene recognition model using a film and television scene classification dataset;

[0048] (6) Train and build a scene recognition model using scene datasets;

[0049] (7) Train and build a camera movement recognition model using a film and television film movement classification dataset;

[0050] (8) Train and build an object detection and recognition model using prop and object datasets;

[0051] (9) Train and build an image description generation model using a dataset of film and television materials;

[0052] The M20 video semantic annotation module is used to annotate video semantic tags based on semantic information;

[0053] The video metadata module M30 is constructed to build video metadata for film and television materials based on semantic tags;

[0054] The construction module is used to construct video metadata based on the semantic tag information already labeled in the video and the required data format, and store the semantic tag information in a video metadata file of a specified format.

[0055] During post-production, importing film and television footage in batches into post-production software allows the reading of initial metadata data items, including basic information related to shooting, such as frame rate, resolution, duration, and audio information.

[0056] In the post-production process, the video metadata annotation method for film and television materials of this invention is applied to analyze and process batch-uploaded film and television materials, generate semantic tags and data items related to the film and television material content required by the user, and store them in the video metadata. This facilitates the reading of post-production software, enabling the classification, retrieval, and management of materials based on metadata information, such as... Figure 3 As shown, it includes the following steps:

[0057] Step 1: Extract semantic information from the frames of the film and television footage, such as... Figure 3 As shown, it includes: video preprocessing step 101, detection and recognition of the clapperboard step 102, and frame semantic information extraction step 103;

[0058] Video preprocessing step 101: Read the video and perform video preprocessing, including the following sub-steps:

[0059] Step 101.1: Obtain the video files of the film and television materials, and sort and organize the film and television materials to be processed according to the file name, file type, and file generation time;

[0060] Step 101.2: Read the video footage to be processed, obtain video information, and perform decoding and conversion;

[0061] Decoding and conversion specifically include: decoding video data using appropriate video decoding algorithms, downsampling the decoded video data, converting it to a specific depth, pixel format, resolution, frame rate, and color space, and converting the audio data of the original film and television file into audio data with a specific depth, specific sampling frequency, and specific encoding.

[0062] Decoding and conversion are performed to facilitate subsequent analysis and processing;

[0063] Step 101.3: Obtain the encoded and converted video file, and extract the frame sequence according to a certain sampling frequency;

[0064] The frame rate of video footage is usually 24fps, 30fps, or 60fps. Only a certain amount of video frame sequences need to be extracted to satisfy the analysis and understanding of the video content.

[0065] The obtained video frame sequence is sent to the semantic information extraction module M10;

[0066] Step 101.4: Construct a semantic information database;

[0067] Film and television materials and user needs vary. By constructing a semantic information dataset, it is possible to supplement the semantic tags to be detected, so as to adjust the functions and steps of the semantic information extraction module M10 according to user needs.

[0068] Step 101.5: Using the semantic information database constructed in step 101.4, train the semantic information extraction model, organize the categories of semantic information to be identified, and adjust the specific recognition and detection model to be used by the semantic information extraction module M10.

[0069] Based on deep learning and computer vision technologies, the semantic information extraction module M10 will perform content analysis and understanding on the video and film frames processed by 101.2 and 101.3.

[0070] Step 102, by detecting and recognizing the clapperboard, extracts the scene information of the frame, including the following sub-steps:

[0071] Step 102.1: Read the film and television footage frames and detect whether a clapperboard exists in the current frame;

[0072] If it exists, proceed to step 106.2; if it does not exist, proceed to step 103.1.

[0073] Step 102.2: Determine the bounding box where the clapperboard is located according to the clapperboard positioning method;

[0074] Step 102.3: Crop the image according to the clapperboard bounding box, and segment the clapperboard region image obtained in step 106.3 using the image segmentation method, and send the different segmented regions to the next step;

[0075] Step 102.4: Input the segmented clapperboard region image into the handwritten character recognition module;

[0076] Step 102.5: Obtain the text information recorded in the clapperboard, such as the scene number, shot number, and sequence number;

[0077] Among them, the field number, mirror number, and sub-number data items take the range of integer natural numbers, and the text information is a string;

[0078] Step 103: Extract semantic information from the frame using the semantic information extraction module M10, including the following sub-steps:

[0079] Step 103.1: Read the film and television frame, and combine the face detection and recognition model and the face key point detection model to detect whether there is a person in the current frame. If there is, perform face detection and recognition to obtain actor information data items.

[0080] The input frame is fed into a trained facial landmark detection model to perform face detection and face bounding box regression. All face bounding boxes are filtered, and the model outputs face bounding boxes and several facial key feature points to determine whether a face exists.

[0081] If no face is found, the actor information will not be extracted, and step 103.4 will continue.

[0082] If a face is present, the detected face region is aligned based on several key facial feature points. The aligned face image is then input into the face detection and recognition model and compared one by one with the actor's face data to obtain the corresponding actor information and output it.

[0083] The actor information includes the position of the face bounding box and the corresponding actor's name. Multiple faces are allowed in one frame. The actor information is arranged in the order from left to right and from top to bottom in the frame. Steps 103.2 and 103.3 are executed.

[0084] Step 103.2: Based on the key facial feature points detected in 103.1, estimate the facial pose using the pose detection method to determine the current facial orientation and shooting angle of the actor.

[0085] Postures include: frontal, lateral, 3 / 4 lateral, tilted, prone, and supine;

[0086] Step 103.3: Based on the face bounding box detected in 103.1, and combining the face bounding box with the frame size, determine the height of the frame occupied by the face bounding box using the scene recognition model; based on conventional scene shooting experience, set corresponding thresholds for different scene sizes, and obtain the shooting scene data item of the frame by comparing with the threshold.

[0087] Shot type data items include: wide shot, medium shot, close-up, extreme close-up, and extreme close-up;

[0088] Step 103.4: Read the video footage frames and detect whether there is a person in the current frame. If there is, combine the human skeleton key point detection model to obtain human motion information.

[0089] For some shooting scenarios, such as when the actor's back is to the camera or the face is not clear, the proportion of the height of the human body bounding box to the height of the image is determined by the human skeleton key point detection model. Based on the experience of shooting conventional shot types, corresponding thresholds are set for different shot types. The shooting shot type data items of the frame are obtained by comparing with the threshold.

[0090] If no person is present, the scene information will rely on the object bounding box output by the object detection and recognition model in step 103.8 to determine the shooting scene data item of the frame;

[0091] Step 103.5: Read the film and television footage frames, classify the scenes in the current frame using the scene recognition model, and compare the scene categories with the scene directory to obtain the interior and exterior scene classification results.

[0092] Step 103.6: Read the film and television frame images, combine the scene category and the indoor / outdoor scene classification results, and comprehensively consider the average brightness, color temperature, chroma, and other values ​​of the R, G, and B channels of the frame images to analyze and obtain the daytime and nighttime scene classification results of the frame images;

[0093] Step 103.7: Read the film and television footage frames and analyze and identify the camera movement patterns in the current frame using the camera movement recognition model;

[0094] The current frame and several frames before and after are input as a frame sequence. The sparse optical flow map of each frame is determined using the optical flow method, and the optical flow trajectory in the temporal domain is obtained. Combined with the optical flow trajectory features of different camera movement methods, the classification and recognition of camera movement methods are realized.

[0095] The camera movement data items include: fixed position, push, pull, pan, tilt, track movement, and handheld;

[0096] Step 103.8: Read the film and television footage frames, and use the object detection and recognition model to detect the objects and props present in the current frame, and obtain the bounding boxes and categories of objects of different sizes;

[0097] The object category data item takes the prop type and the pre-marked related objects, including: microphone stand, fill light, tripod;

[0098] Step 103.9: Read the film and television frame, combine the image description generation model and the semantic information obtained in the above steps, and generate a text description of the current frame.

[0099] In this embodiment, the semantic information includes scene number, shot number, second shot number, actor, pose, human body key points, shot type, interior / exterior shot, scene, day / night, camera movement method, object category, and text description, as shown in Table 1. The extracted semantic information is stored and recorded in the semantic tag field corresponding to the frame.

[0100] Table 1

[0101] Frame number field number Mirror number Second number actor posture Shot type inside and outside Scene day and night Camera movement object Text description 000001 3 5 1 Xiaoming side panoramic Inside living room day Fixed position football Xiaoming is playing soccer in the living room.

[0102] Step 2: Label the video with semantic tags based on semantic information:

[0103] Step one involves acquiring semantic information from video frames. For the video as a whole, all semantic information in the frame sequence is analyzed and filtered. Based on the semantic tag fields required by the user, duplicate, redundant, and erroneous data items are removed, while data items valuable for video retrieval and management are retained. This process constructs video semantic tags, such as... Figure 4 As shown, it includes the following sub-steps:

[0104] Step 201: Determine the semantic tag fields to be extracted according to user needs, obtain the semantic information obtained in Step 1, and map the data items of each category of the extracted semantic information to the semantic tag fields corresponding to the frame based on the frame index.

[0105] Step 202: Obtain all semantic tag fields and data items constructed in step 201 above. Integrate all semantic tags and data items of the entire frame sequence of the film and television materials in order, perform statistical analysis, clean up and remove redundant data items according to the continuity between frames, filter the semantic tags required by the user, and extract and retain effective video semantic tag data items through multi-frame statistical methods.

[0106] Step 203: Construct the semantic tag field for the video and associate and store the data items filtered in step 202.

[0107] Due to different user needs, supplementary models are trained by constructing corresponding sample datasets to expand or delete semantic tag fields and supplement other semantic tag fields related to video content.

[0108] Step 3: Construct video metadata for film and television materials based on semantic tags:

[0109] Based on the video semantic tags obtained in step two, semantic tag fields are appended to the video metadata according to the original video metadata format of the film and television materials. These fields are then associated and stored with the corresponding data items. Depending on the post-production software used by the user, and considering the data structure of the video metadata that the software can parse, the video metadata is converted into a different format to obtain video metadata for film and television material retrieval and management. Figure 5 As shown, it includes the following sub-steps:

[0110] Step 301: Since the metadata storage data formats of digital cameras from different manufacturers are different, the video information tags and data items of the current film and television materials are extracted by using the video metadata parsing method provided by the film and television film shooting equipment manufacturer or by reading the video metadata file of the film and television materials with the help of post-processing software.

[0111] Step 302: Based on the video information tags of the current film and television materials, add the semantic tag fields required by the user, that is, the video semantic tags constructed in step two;

[0112] Step 303: Using each material as a separate unit, store the tags and their corresponding data items from Step 301 and Step 2 in text file format.

[0113] Step 304: Convert the text file format stored in step 303 into a metadata specification format that can be parsed by post-production or digital intermediate software for easy application, and store it in a specific file.

[0114] like Figure 6 As shown, the examples use two mainstream film post-production software programs:

[0115] The metadata file format supported by the editing software Avid Media Composer is ALE file, and the file data structure is as follows: Figure 6 As shown in (a), the file header field is used to record the basic information and category information of the film and television materials, and the rest is used to record the semantic tag information of the film and television materials. Each piece of material occupies one line and is separated by tabs.

[0116] The color correction software Davinci Resolve supports CSV files as the metadata file format, and the file data structure is as follows: Figure 6 As shown in (b), the fields in the first row are the tag category information, and the subsequent fields are used to record the semantic tag information of the film and television materials. Each piece of material occupies one line and is separated by commas.

[0117] The present invention provides a video metadata annotation method for film and television materials, which supports users to upload film and television materials in batches for analysis and processing, supports users to customize the required semantic information tags, and stores them in the metadata file format selected by the user, and exports the corresponding metadata file according to the requirements.

[0118] During post-production, the metadata file exported using this method is imported into the post-production software, such as... Figure 7 As shown, the post-production software can parse the metadata file and data items exported by the method of this invention, realize the reading of the data items of the initial metadata and the semantic data items completed by the method of this invention, and realize the classification, retrieval and management of film and television materials based on the metadata information through various tags in the metadata, thereby improving the efficiency of film and television material management in the post-production process.

[0119] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for annotating video metadata of film and television materials, characterized in that: Includes the following steps: Step 1: Extract semantic information from the frames of the film and television footage, which includes the following sub-steps: Step 1.1: Read the video and perform video preprocessing to obtain a video frame sequence. Use object detection methods to determine whether a clapperboard exists in the frame. Step 1.2: Read the video frame sequence frame by frame. Using the clapperboard detection and recognition method, locate the clapperboard area in the frame where the clapperboard is present, identify the field number, shot number and sub-shot number marked on the clapperboard, and record the field number, shot number and sub-shot number marked on the clapperboard as the tag data item of the frame. Step 1.3: Read the video frame sequence frame by frame, extract the content information appearing in the frame by using the target detection and recognition method in the semantic information extraction module, and store it as semantic tag data items; Step 2: Label the video with semantic tags based on semantic information, which includes the following sub-steps: Step 2.1: Determine the semantic tag fields to be extracted from the video, map the semantic information extracted from the frame to the semantic tag data items, and store them in the semantic tag fields corresponding to each frame. Step 2.2: Integrate all semantic tag fields extracted from the frame sequence, filter and analyze semantic tag data items, clean up and remove redundant data items based on inter-frame continuity, and generate valid video semantic tag data items; Step 2.3: Associate and store the extracted video semantic tag data items with the corresponding semantic tag fields. Step 3: Construct video metadata for film and television materials based on semantic tags, which includes the following sub-steps: Step 3.1: Obtain video materials and read video metadata. Perform splitting, deletion, modification, and format normalization operations on the read video metadata, while retaining the necessary video metadata tag fields. Step 3.2: Add semantic tag fields to the read video metadata. Based on the existing metadata tag fields of the film and television materials, add semantic tag fields related to the video content, and write the marked video semantic tags into the video metadata according to the original data format of the current film and television material metadata. Step 3.3: Using each material as a separate unit, integrate the initial metadata obtained in Step 3.1 with the semantic information tags and their corresponding data items obtained in Step 2, and store them in the format of a text file; Step 3.4: Convert the metadata to the required format. For the specified non-linear editing software or digital intermediate software, rewrite it into a data structure or tag field that supports its reading and writing, and store it in a video metadata file of a specific format.

2. The video metadata annotation method for film and television materials as described in claim 1, characterized in that: The video preprocessing in step 1.1 includes: encoding and converting the video, reducing the resolution and sampling the frames, preserving the semantic information in the video frames, and outputting the video frame sequence; The target detection method in step 1.1 is implemented through a clapperboard detection and recognition model; The clapperboard in step 1.1 is a device used in film and television shooting to write and mark the scene number, shot number, and number of shots of the current shooting content; Scene number, shot number, and number of shots are abbreviated as scene number, shot number, and shot number, respectively.

3. The video metadata annotation method for film and television materials as described in claim 1, characterized in that: The clapperboard detection and recognition method in step 1.2 is implemented using a handwritten character recognition model.

4. The video metadata annotation method for film and television materials as described in claim 1, characterized in that: The data items in the semantic tag field in step 1.3 include: actor, pose, shot type, inside / outside, scene, day / night, camera movement, object, and text description; The target detection and recognition methods in step 1.3 are implemented through a face detection and recognition model, a face key point detection model, a human skeleton key point detection model, a scene recognition model, a camera movement recognition model, an object detection and recognition model, and an image description generation model, respectively.

5. The video metadata annotation method for film and television materials as described in claim 1, characterized in that: The film and television materials in step 3.1 include videos shot with digital cameras, as well as source videos existing in Internet video websites, video content providers, and media asset storage systems, used for the production of movies, TV series, variety shows, and live online broadcasts.

6. The video metadata annotation method for film and television materials as described in claim 1, characterized in that: In step 3.2, the semantic tag fields that must be set include scene number, shot number, second shot number, actor, pose, human body key points, shot type, interior / exterior, scene, day / night, camera movement, object category, and text description. If a field fails to retrieve the corresponding data item, it is recorded as N / A.

Images