A method and system for generating analysis reports and recommending videos based on content detection

Abstract

The present application discloses a method and system for generating analysis reports and recommending videos based on content detection, which belongs to the field of video content analysis technology. The application uses a video analysis module to analyze the social value and significance of specific scenes, objects, and human behaviors in the video to be detected, extracts and optimizes prompt words from the analysis results, generates an analysis report, and analyzes the effect of the analysis results to generate recommended video suggestions, solving the problem that current video recognition technology cannot form analysis reports for the content of the video and cannot provide viewing suggestions.

Description

Technical Field

[0001] The present application relates to the technical field of video content analysis, and in particular to a method and system for generating analysis reports and recommending videos based on content detection. Background Art

[0002] In the field of video playback terminals, especially home video playback terminals, including IPTV (Internet Protocoltelevision), NAS (Network Attached Storage Video), PAD (Portable Android Device), computers, etc., parents generally care about the positive impact of the content their children watch on their three views: they review the content their children are going to watch in advance; they understand the content their children have watched afterwards, and hope to have an intelligent analysis assistant to perform intelligent analysis of the content their children have watched, analyze the content they have watched and form an analysis report, analyze the impact of the video on the children's knowledge and values, and provide suggestions on what videos they should watch later.

[0003] At present, the main technologies involved in video content analysis are in the fields of image recognition and video recognition. There are relatively mature technologies for content recognition, including object recognition and text recognition in pictures and videos. However, current video recognition technology does not have the ability to determine the social value and significance of specific scenes, objects and human behaviors in videos and form video reports, nor does it have the ability to analyze values ​​and knowledge, or provide video viewing recommendations. Summary of the invention

[0004] The present application aims to provide a method and system for generating analysis reports and recommending videos based on content detection, so as to solve the deficiencies in the prior art. The technical problems to be solved by the present application are achieved through the following technical solutions.

[0005] In one aspect, a method for generating an analysis report and recommending videos based on content detection is provided, the method comprising:

[0006] S1, obtaining a video to be detected, and processing the video to be detected to obtain multiple video clips;

[0007] S2. Obtain video analysis results using a multi-dimensional data analysis model based on multiple video clips;

[0008] S3, obtaining a large model prompt word through a prompt word extraction method according to the video analysis result;

[0009] S4. Perform effect analysis and optimization based on the video analysis result and the large model prompt word to obtain content effect analysis results and optimized large model prompt words;

[0010] S5. According to the content effect analysis result, a recommended video is obtained according to a recommendation algorithm;

[0011] S6. Generate a report as required based on the optimized large model prompt words to obtain a video analysis report.

[0012] In S2, based on multiple video clips, a multi-dimensional data analysis model is used to obtain video analysis results, including:

[0013] S21, obtaining a plurality of sharpened video clips through noise reduction and enhancement processing according to the plurality of video clips;

[0014] S22, labeling detection frames according to the multiple sharpened video clips through a pre-trained model to obtain detection objects of the video clips;

[0015] S23, according to the detection object of the video clip, extracting the features of the detection object through a convolutional neural network model to obtain the features of the detection object of the video clip;

[0016] S24, obtaining the category of the detection object through a classifier according to the features of the detection object in the video clip;

[0017] S25, comparing the features of the detected object with a pre-established object feature library to obtain the name of the detected object;

[0018] S26. According to the category of the detection object of the video clip and the name of the detection object, in combination with the video clip, semantic analysis is performed on the scene in the video clip and the interactive behavior in the video clip is understood to obtain a video analysis result.

[0019] In S3, according to the video analysis result, a large model prompt word is obtained through a prompt word extraction method, including:

[0020] S31, according to the video analysis result, obtaining the object data recognized in each frame, and obtaining the position and time of the detected object;

[0021] S32, performing event recognition according to the category of the detection object, the name of the detection object, and the location and time of the detection object, and generating an event prompt word;

[0022] S33, selecting a language template according to the event prompt word to obtain the large model prompt word before optimization;

[0023] S34. Obtain the large model prompt words through filtering and optimization according to the large model prompt words before the optimization.

[0024] In S4, effect analysis and optimization are performed according to the video analysis result and the large model prompt word to obtain content effect analysis results and optimized large model prompt words, including:

[0025] S41, inputting the large model prompt word into the image generation model to obtain an image generation result;

[0026] S42. Perform comparative analysis based on the video analysis result and the image generation result, adjust the structure and content of the large model prompt word, repeat step S41, the image generation result is consistent with the video analysis result, and obtain the content effect analysis result and the optimized large model prompt word.

[0027] The recommendation algorithms in S5 include: content-based recommendation algorithm, collaborative filtering algorithm, hybrid recommendation algorithm, deep learning algorithm and knowledge graph-based recommendation algorithm.

[0028] In S6, a report is generated as required according to the optimized large model prompt words to obtain a video analysis report, including:

[0029] S61, extracting representative events of the video clip in chronological order according to the optimized large model prompt words;

[0030] S62, according to the representative events of the video clip, obtaining the core content of the video to be detected by scene analysis and inferring the specific semantics of the representative events of the video clip;

[0031] S63, generating a preliminary video analysis report according to the core content of the video to be detected and the analysis report template;

[0032] S64: Proofread and optimize the initial version of the video analysis report to obtain a video analysis report.

[0033] In an optional implementation manner, the method for processing the video to be detected in S1 includes: video slicing, video frame extraction, behavior extraction segmentation and character change segmentation.

[0034] In an optional implementation, the features of the detected object in the video clip in S23 include visual features, motion features, audio features, and posture estimation.

[0035] In an optional implementation, the analysis report template in S6 includes:

[0036] Requirement 1: The analysis report template contains the title, heading and details section;

[0037] Requirement 2: The detailed content section includes scenes, objects, human actions and time;

[0038] Requirement 3: If there are abnormal factors, the analysis report template includes an abnormal factor warning and records the time point in the video to be detected.

[0039] On the other hand, a system for generating an analysis report and a recommended video based on content detection is provided, and the system is applied to a method for generating an analysis report and a recommended video based on content detection, and the system includes:

[0040] The video playback module is used to obtain the video to be detected, process the video to be detected, and obtain multiple video clips;

[0041] A video analysis module is used to obtain video analysis results based on multiple video clips using a multi-dimensional data analysis model;

[0042] A large model prompt word module is used to obtain a large model prompt word through a prompt word extraction method according to the video analysis result;

[0043] An effect impact analysis module is used to perform effect analysis and optimization based on the video analysis results and the large model prompt words to obtain content effect analysis results and optimized large model prompt words;

[0044] A recommendation module is used to obtain recommended videos according to the content effect analysis results and the recommendation algorithm;

[0045] The report generation module is used to generate a report as required according to the optimized large model prompt words to obtain a video analysis report.

[0046] The embodiments of the present application include the following advantages:

[0047] The embodiments of the present application provide a method and system for generating analysis reports and recommending videos based on content detection. On the one hand, the application uses a video analysis module to analyze the social value and significance of specific scenes, objects, and human behaviors in the detected video, extracts and optimizes prompt words from the analysis results, and generates an analysis report, thereby solving the problem that current video recognition technology cannot form an analysis report for the content of the video. On the other hand, the application performs an effect analysis on the analysis results and generates recommended video suggestions, thereby solving the problem that current video recognition technology cannot provide viewing suggestions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flowchart of the steps of an embodiment of a method for generating an analysis report and recommending videos based on content detection of the present application;

[0049] Figure 2 This is a flowchart of the steps of obtaining video analysis results in the embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application;

[0050] Figure 3 This is a flowchart of the steps of obtaining large model prompt words in the embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application;

[0051] Figure 4 It is a flowchart of the steps of obtaining content effect analysis results and optimized large model prompt words in the method embodiment of the present application for generating analysis reports and recommending videos based on content detection;

[0052] Figure 5 It is a flowchart of the steps of generating a video analysis report in an embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application;

[0053] Figure 6 is a schematic diagram of Example 1 of a method for generating an analysis report and recommending videos based on content detection of the present application;

[0054] Figure 7 It is a system block diagram of a system embodiment of the present application for generating analysis reports and recommending videos based on content detection. DETAILED DESCRIPTION

[0055] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments. Example

[0056] This embodiment 1 provides a video to be detected of a person jogging in the park in the morning, such as Figure 1 The step flow chart of the method embodiment of the present application for generating an analysis report and recommending videos based on content detection and the like Figure 6 The schematic diagram of Embodiment 1 of the method for generating an analysis report and recommending videos based on content detection of the present application is shown. The present application provides a method for generating an analysis report and recommending videos based on content detection, and the method includes:

[0057] S1, obtaining a video to be detected, and processing the video to be detected to obtain multiple video clips;

[0058] Specifically, a video of a person jogging in the park in the morning is obtained. The quality and source of the video meet the analysis requirements, the format is correct and the content is complete. The video of the morning jogging is divided into continuous frame images, so that each frame can be regarded as an independent image for processing and key frames are extracted. Key frames can be extracted according to certain rules, such as selecting a frame at a certain time interval, or selecting frames with larger changes according to the degree of change in the video content. For some videos with relatively slow scene changes, the shot segmentation method can be used to segment the video into different shots. The starting frame and the ending frame of each shot can be used as key frames. The video of the morning jogging is divided into multiple video clips according to the key frames.

[0059] S2. Obtain video analysis results using a multi-dimensional data analysis model based on multiple video clips;

[0060] Specifically, Figure 2 The flowchart of the steps for obtaining the video analysis results in the embodiment of the method for generating analysis reports and recommending videos based on content detection of the present application is shown. According to multiple video clips, multiple sharpened video clips are obtained through denoising, brightness adjustment and contrast enhancement processing, and an appropriate deep learning model is selected to detect the object in each frame of the image. The deep learning model can be YOLO (You Only Look Once, an object detection system for computer vision tasks), Faster R-CNN (Towards Real-Time Object Detection with Region Proposal Networks, real-time target detection based on region proposal networks), etc. These deep learning models can identify common objects such as people, cars, animals, etc. The bounding box of the detected object in each frame of the image is annotated to obtain the detected object of the video clip, ensuring that the detected object can continue to be tracked in subsequent frames, even if the detected object is blocked or moved for a short time.

[0061] Then, according to the detection object in the video clip, the features of the detection object are extracted through the convolutional neural network model to obtain the color histogram of the video, the texture feature vector of the video, the edge contour of the video and other information, the trajectory of the detection object, the speed of the detection object, the acceleration and other movements of the detection object, the audio spectrum of the detection object, the energy of the detection object, the tone of the detection object and the skeleton structure of the detection object, etc.

[0062] The features of the detected object are classified by a classifier, and a predefined category is assigned to the detected object to obtain the category of the detected object. The classifier can be a SVM (Support Vector Machine) and a deep neural network. The features of the detected object are compared with a pre-established object feature library to obtain the name of the detected object.

[0063] Finally, according to the category of the detection object of the video clip and the name of the detection object, in combination with the video clip, a semantic analysis is performed on the scene in the video clip and the interactive behavior in the video clip is understood to obtain a video analysis result.

[0064] The video analysis results can output the object, time, location and other information obtained from the analysis in the form of tables, JSON (JavaScript Object Notation, a lightweight data exchange format), etc., and can also be visualized to mark the bounding box and category of the identified object in the video, and generate annotated video to intuitively display the analysis results.

[0065] S3, obtaining a large model prompt word through a prompt word extraction method according to the video analysis result;

[0066] Specifically, Figure 3 The flowchart of the steps of obtaining the large model prompt words in the embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application is shown, and according to the video analysis result, the object data recognized in each frame is obtained to obtain the position and time of the detected object;

[0067] Extract the detected object data and obtain the detected object data identified in each frame from the video analysis results, including:

[0068] The category of the detected object: such as animals, plants and buildings, etc.

[0069] The name of the detection object: such as people, cars, animals and office buildings, etc.

[0070] Object features: color, shape, posture, and movement, etc.

[0071] The location of the detected object: the time, position and duration of its appearance in the video.

[0072] Then, according to the category of the detected object, the name of the detected object, and the location and time of the detected object, event recognition is performed to generate event prompt words, including:

[0073] Scene segmentation: Divide the detected object into different scene segments, each of which may represent a complete event or action (such as a person walking across the road).

[0074] Scene description: Extract scene information from the detected object (such as indoor, outdoor, day, night, etc.), as well as the relationship description between objects (such as "a person walking on the street", "two cars driving on the road").

[0075] Action and behavior extraction: Analyze the actions of the detected object (such as "walking", "running", "driving", etc.) and use these actions as part of the prompt word.

[0076] Event recognition: Identify possible events based on the dynamic and interactive behaviors of the detected objects (such as distance changes between multiple objects, motion trajectories, etc.). For example, pedestrians approaching vehicles may represent "crossing the road", and people sitting on a chair may represent "resting".

[0077] Finally, select the natural language template: Based on the category, action, and scene information of the object, construct a sentence structure that can naturally describe the video content, such as: a person jogging in the park in the morning or two cars driving towards each other on the road. When multiple scenes and objects are involved, more complex prompt words can be generated by combining sentences. For example: a person is walking on the street and two cars pass by him at the same time.

[0078] According to the large model prompt words before the optimization, ensure that the prompt words do not contain repeated or irrelevant information. For example, if a detection object appears in multiple scenes and the action or scene does not change, the repeated description can be omitted to obtain the large model prompt words, for example: simplifying "a person running on the grass in the park" to "a person running in the park".

[0079] S4. Perform effect analysis and optimization based on the video analysis result and the large model prompt word to obtain content effect analysis results and optimized large model prompt words;

[0080] Specifically, Figure 4 The flowchart of the steps for obtaining content effect analysis results and optimized large model prompt words in the method embodiment of the present application for generating analysis reports and recommending videos based on content detection is shown, the large model prompt words are input into the image generation model to obtain image generation results, and then a comparative analysis is performed based on the video analysis results and the image generation results, and the structure and content of the large model prompt words are adjusted in real time until the image generation results are consistent with the video analysis results, so as to obtain content effect analysis results and optimized large model prompt words.

[0081] S5. According to the content effect analysis result, a recommended video is obtained according to a recommendation algorithm;

[0082] Specifically, the recommendation algorithms in S5 include: content-based recommendation algorithm, collaborative filtering algorithm, hybrid recommendation algorithm, deep learning algorithm and knowledge graph-based recommendation algorithm.

[0083] Compare and match the content effect analysis results with the video content in the existing video library. Calculate the similarity between the contents. The higher the similarity, the more similar the two videos are in content. Similarity calculation methods can be used, including: cosine similarity and Euclidean distance. Then, based on the similarity, select a batch of videos that are most similar to the video to be recommended as recommended videos.

[0084] S6. Generate a report as required based on the optimized large model prompt words to obtain a video analysis report.

[0085] Specifically, Figure 5 The flowchart of the steps for generating a video analysis report in an embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application is shown. First, based on the optimized large model prompt words, representative events of the video clip are extracted in chronological order. Based on the representative events of the video clip, the core content of the video to be detected is obtained through scene analysis and inference of the specific semantics of the representative events of the video clip.

[0086] Then, based on the core content of the video to be detected, an initial version of the video analysis report is generated according to the analysis report template. Based on the initial version of the video analysis report, check whether the initial version of the video analysis report accurately reflects the content in the video, ensure that the sentences are fluent and the description is consistent with the video, and manually adjust some erroneous or ambiguous descriptions. At the same time, ensure that the behavior and scene descriptions of each detected object in the video are consistent with the timeline of the event to prevent the generated text from having semantic deviations from the video content, and finally generate a video analysis report.

[0087] An example video analysis report is as follows:

[0088] Title: Morning jog in the park

[0089] Introduction: This video shows a peaceful morning in a green park, where a runner is concentrating on his morning run. The surrounding environment is quiet, with occasional vehicles passing by, creating a picture of harmonious coexistence between man and nature.

[0090] Details:

[0091] At the beginning of the video, we see the sun just rising, shining through the leaves onto the ground. The runner is wearing a blue sportswear and walking briskly.

[0092] As the camera moves, we can see the runner running along the park path. The green grass beside him and the woods in the distance set off the tranquil atmosphere of the early morning.

[0093] In the middle of the video, a few cars drove slowly on the road in the distance, but they did not break the tranquility of the park. The whole scene seemed leisurely and comfortable.

[0094] in conclusion:

[0095] Through this video, we can feel the vitality brought by early morning exercise, and at the same time appreciate the tranquility and harmony in the natural environment. The whole video gives people a sense of relaxation and calmness, showing the harmonious coexistence of man and nature. Example

[0096] This embodiment 2 provides a 5-minute animation video, such as Figure 1 The step flow chart of the method embodiment of the present application for generating an analysis report and recommending videos based on content detection is shown. The present application provides a method for generating an analysis report and recommending videos based on content detection, and the method includes:

[0097] S1, obtaining a video to be detected, and processing the video to be detected to obtain multiple video clips;

[0098] Specifically, a 5-minute animation video is obtained, the quality and source of which meet the analysis requirements, the format is correct and the content is complete. The 5-minute animation video can be extracted into a video segment of 30 seconds in length, and the following segments are divided:

[0099] Clip 1, Clip 2 (0:00 - 1:00): The main characters, a boy and a girl, are playing and telling stories in the park. The girl suggests exploring deeper into the park.

[0100] Clip 3, Clip 4 (1:00 - 2:00): The boy belittles the girl's suggestion, saying that girls are timid and unsuitable for adventure, and that only boys have the courage to explore and take risks, and sets off alone to the depths of the park;

[0101] Clip 5 and Clip 6 (2:00 - 3:00): The boy encounters a dangerous monster and begins to fight it. The fight scene involves weapons and violence.

[0102] Clip 7, Clip 8 (3:00 - 4:00): The girl then arrives and tries to help the boy, suggesting that the boy talk to the monster to find out what it needs and whether they can make friends, but the boy refuses and continues to use violence to solve the problem;

[0103] Clip 9, Clip 10 (4:00 - 5:00): The boy wins the battle by beating the monster away, then laughs at the girl, saying that monsters are inherently dangerous and that the suggestion of making friends with monsters is stupid. Only fighting monsters can show the boy's bravery, and the girl is not qualified to participate.

[0104] S2. Obtain video analysis results using a multi-dimensional data analysis model based on multiple video clips;

[0105] Specifically, Figure 2 The flowchart of the steps for obtaining video analysis results in the embodiment of the method for generating analysis reports and recommending videos based on content detection of the present application is shown. According to multiple video clips, multiple sharpened video clips are obtained through denoising, brightness adjustment and contrast enhancement processing, and the YOLO deep learning model is selected to detect objects in each frame of the image.

[0106] By detecting violent fighting scenes (such as using weapons to attack and monsters being injured), motion detection technology is used to identify fighting behaviors and injuries. The violent scenes are mainly concentrated in segments 5 to 8, lasting about 2 minutes, accounting for 40% of the entire video. Through image analysis, the weapons used by the boy in the fight, such as swords or other attack tools, are identified. The fighting scenes contain intense violent actions, such as monsters being attacked and injured, which reinforces the means of solving problems with violence.

[0107] NLP (Natural Language Processing) technology was used to analyze the boy’s conversation, especially in segments 2, 3, 7 and 10, where the boy used language to belittle the girl’s abilities and suggestions. Role analysis revealed that the girl was marginalized, and although she offered to help, she was rejected by the boy, showing a sexist stereotype.

[0108] The features of the detected object are classified by a classifier, and a predefined category is assigned to the detected object to obtain the category of the detected object. The classifier can be a SVM (Support Vector Machine) and a deep neural network. The features of the detected object are compared with a pre-established object feature library to obtain the name of the detected object.

[0109] Finally, according to the category of the detection object of the video clip and the name of the detection object, in combination with the video clip, a semantic analysis is performed on the scene in the video clip and the interactive behavior in the video clip is understood to obtain a video analysis result.

[0110] The video analysis results can output the object, time, location and other information obtained from the analysis in the form of tables, JSON (JavaScript Object Notation, a lightweight data exchange format), etc., and can also be visualized to mark the bounding box and category of the identified object in the video, and generate annotated video to intuitively display the analysis results.

[0111] S3, obtaining a large model prompt word through a prompt word extraction method according to the video analysis result;

[0112] Specifically, Figure 3The flowchart of the steps of obtaining the large model prompt words in the embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application is shown, and according to the video analysis result, the object data recognized in each frame is obtained to obtain the position and time of the detected object;

[0113] Extract the detected object data and obtain the detected object data identified in each frame from the video analysis results, including:

[0114] The category of the detected object: such as animals, plants and buildings, etc.

[0115] The names of the detected objects: such as boy, girl, monster, and battle site.

[0116] Object features: color, shape, posture, and movement, etc.

[0117] The location of the detected object: the time, position and duration of its appearance in the video.

[0118] Then, according to the category of the detected object, the name of the detected object, and the location and time of the detected object, event recognition is performed to generate event prompt words, including:

[0119] Scene segmentation: Divide the detected object into different scene segments, each of which may represent a complete event or action (such as using a weapon to attack, a monster being injured).

[0120] Scene description: Extract scene information from detected objects (such as outdoor, daytime, nighttime, etc.), as well as descriptions of relationships between objects (such as "the weapons used by the boy in battle").

[0121] Action and behavior extraction: Analyze the actions of the detected object (such as "the monster is attacked and injured").

[0122] Event recognition: Identify possible events based on the dynamic and interactive behaviors of the detected objects (such as distance changes and motion trajectories between multiple objects).

[0123] Finally, select a natural language template: Based on the object category, action, and scene information, construct a sentence structure that can naturally describe the video content. Based on the large model prompt words before optimization, ensure that the prompt words do not contain repeated or irrelevant information.

[0124] S4. Perform effect analysis and optimization based on the video analysis result and the large model prompt word to obtain content effect analysis results and optimized large model prompt words;

[0125] Specifically, Figure 4The flowchart of the steps for obtaining content effect analysis results and optimized large model prompt words in the method embodiment of the present application for generating analysis reports and recommending videos based on content detection is shown, the large model prompt words are input into the image generation model to obtain image generation results, and then a comparative analysis is performed based on the video analysis results and the image generation results, and the structure and content of the large model prompt words are adjusted in real time until the image generation results are consistent with the video analysis results, so as to obtain content effect analysis results and optimized large model prompt words.

[0126] S5. According to the content effect analysis result, a recommended video is obtained according to a recommendation algorithm;

[0127] Specifically, the content effect analysis results show that the recommendation algorithm uses a combination of collaborative filtering and hybrid recommendation algorithms to recommend a series of suitable alternative videos to help children avoid negative influences and educate them on correct values. For example, reduce educational videos with violent content, recommend some children's adventure videos without violent content, and focus on showing plots where problems are solved in a peaceful way.

[0128] Recommended video 1: An animated film that shows the protagonist solving problems through teamwork and communication rather than through violent fighting;

[0129] Recommended video 2: A documentary about nature adventures that shows children learning new knowledge through exploration rather than through violence. Educational videos that convey gender equality: Animations with positive gender equality content are recommended, showing that both male and female characters can equally participate and solve problems;

[0130] Recommended Video 3: An animated short film showing boys and girls working together and contributing equally, emphasizing the complementary abilities of both sides;

[0131] Recommended video 4: Animation with female heroes as the main characters, showing that women are brave, smart, and powerful, helping to break gender stereotypes.

[0132] S6. Generate a report as required based on the optimized large model prompt words to obtain a video analysis report.

[0133] Specifically, Figure 5 The flowchart of the steps for generating a video analysis report in an embodiment of the method for generating an analysis report and recommending videos based on content detection of the present application is shown. First, based on the optimized large model prompt words, representative events of the video clip are extracted in chronological order. Based on the representative events of the video clip, the core content of the video to be detected is obtained through scene analysis and inference of the specific semantics of the representative events of the video clip.

[0134] Then, based on the core content of the video to be detected, an initial version of the video analysis report is generated according to the analysis report template. Based on the initial version of the video analysis report, check whether the initial version of the video analysis report accurately reflects the content in the video, ensure that the sentences are fluent and the description is consistent with the video, and manually adjust some erroneous or ambiguous descriptions. At the same time, ensure that the behavior and scene descriptions of each detected object in the video are consistent with the timeline of the event to prevent the generated text from having semantic deviations from the video content, and finally generate a video analysis report.

[0135] An example video analysis report is as follows:

[0136] Title: An analysis of violence and sexism in children's animation videos

[0137] Introduction: This video is suitable for children aged 6-12, but contains inappropriate violence and sexist content.

[0138] Details:

[0139] 1. Violent content: The violent scenes in the video take up about 2 minutes, mainly focusing on fighting scenes, including weapon attacks, physical conflicts and injury effects, which may cause children to imitate violent behaviors.

[0140] 2. Gender discrimination: The boy character in the video repeatedly belittles the girl character through words and actions, which reinforces gender stereotypes and may affect children's gender concepts.

[0141] in conclusion:

[0142] Psychological impact assessment: This video may trigger some children to imitate violence, especially those in primary school who do not have sufficient discernment. Sexist content may cause boys to form prejudices against the opposite sex and believe that girls are not as good as boys in taking risks or solving problems.

[0143] Parental advice: Parents are advised to supervise children watching such videos and have appropriate discussions with their children to explain the harm of violence and gender stereotypes. In addition, parents can choose to provide more positive, gender-equal children's content to balance the negative impact.

[0144] Figure 7 A system for generating analysis reports and recommending videos based on content detection is shown according to an exemplary embodiment. The system is applied to a method for generating analysis reports and recommending videos based on content detection. The system includes a video playback module, a video analysis module, a large model prompt word module, an effect impact analysis module, a recommendation module and a report generation module. Specifically,

[0145] The video playback module is used to obtain the video to be detected, process the video to be detected, and obtain multiple video clips;

[0146] A video analysis module is used to obtain video analysis results based on multiple video clips using a multi-dimensional data analysis model;

[0147] A large model prompt word module is used to obtain a large model prompt word through a prompt word extraction method according to the video analysis result;

[0148] An effect impact analysis module is used to perform effect analysis and optimization based on the video analysis results and the large model prompt words to obtain content effect analysis results and optimized large model prompt words;

[0149] A recommendation module is used to obtain recommended videos according to the content effect analysis results and the recommendation algorithm;

[0150] The report generation module is used to generate a report as required according to the optimized large model prompt words to obtain a video analysis report.

[0151] The embodiments of the present application provide a method and system for generating analysis reports and recommending videos based on content detection. The application uses a video analysis module to analyze the social value and significance of specific scenes, objects, and human behaviors in the detected video, extracts and optimizes prompt words from the analysis results, generates an analysis report, and at the same time, performs an effect analysis on the analysis results to generate recommended video suggestions, thereby solving the problem that current video recognition technology cannot form analysis reports for the content of the video and cannot provide viewing suggestions.

[0152] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0153] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0154] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0155] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0156] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0157] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for generating analysis reports and recommending videos based on content detection, It is characterized in that The method comprises: S1, obtaining a video to be detected, and processing the video to be detected to obtain multiple video clips; S2. Obtaining video analysis results using a multi-dimensional data analysis model according to the multiple video clips; S3, obtaining a large model prompt word through a prompt word extraction method according to the video analysis result; S4. Perform effect analysis and optimization according to the video analysis results and the large model prompt words to obtain content effect analysis results and optimized large model prompt words; S5. Obtaining a recommended video according to the content effect analysis result and a recommendation algorithm; S6. Generate a report as required based on the optimized large model prompt words to obtain a video analysis report; The step S2 uses a multi-dimensional data analysis model to obtain video analysis results based on the multiple video clips, including: S21, obtaining a plurality of sharpened video clips through noise reduction and enhancement processing according to the plurality of video clips; S22, labeling detection frames according to the plurality of sharpened video clips by using a pre-trained model to obtain detection objects of the video clips; S23, extracting features of the detection object according to the detection object of the video clip by using a convolutional neural network model to obtain features of the detection object of the video clip; S24, obtaining a category of the detection object through a classifier according to the features of the detection object in the video clip; S25, comparing the features of the detection object with a pre-established object feature library to obtain the name of the detection object; S26, according to the category of the detection object of the video clip and the name of the detection object, in combination with the video clip, performing semantic analysis on the scene in the video clip and understanding the interactive behavior in the video clip to obtain a video analysis result; In S3, according to the video analysis result, a large model prompt word is obtained by a prompt word extraction method, including: S31, according to the video analysis result, obtaining the object data recognized in each frame, and obtaining the position and time of the detected object; S32, performing event recognition according to the category of the detection object, the name of the detection object, and the location and time of the detection object, and generating an event prompt word; S33, selecting a language template according to the event prompt word to obtain a large model prompt word before optimization; S34, obtaining a large model prompt word through filtering and optimization according to the large model prompt word before optimization; In S4, effect analysis and optimization are performed according to the video analysis results and the large model prompt words to obtain content effect analysis results and optimized large model prompt words, including: S41, inputting the large model prompt word into the image generation model to obtain an image generation result; S42, performing comparative analysis based on the video analysis result and the image generation result, adjusting the structure and content of the large model prompt word, repeating step S41, the image generation result is consistent with the video analysis result, and obtaining the content effect analysis result and the optimized large model prompt word; The recommendation algorithm in S5 includes: content-based recommendation algorithm, collaborative filtering algorithm, hybrid recommendation algorithm, deep learning algorithm and knowledge graph-based recommendation algorithm; In S6, a report is generated as required according to the optimized large model prompt words to obtain a video analysis report, including: S61, extracting representative events of the video clip in chronological order according to the optimized large model prompt words; S62, obtaining the core content of the video to be detected by scene analysis and inferring the specific semantics of the representative events of the video clips according to the representative events of the video clips; S63, generating a preliminary video analysis report according to the core content of the video to be detected and the analysis report template; S64: Proofread and optimize the initial version of the video analysis report to obtain a video analysis report.

2. The method for generating analysis reports and recommending videos based on content detection according to claim 1, It is characterized in that The method for processing the video to be detected in S1 includes: video slicing, video frame extraction, behavior extraction and segmentation, and character change segmentation.

3. The method for generating analysis reports and recommending videos based on content detection according to claim 1, It is characterized in that The features of the detected object of the video clip in S23 include visual features, motion features, audio features and posture estimation.

4. The method for generating analysis reports and recommending videos based on content detection according to claim 1, It is characterized in that The S6 analysis report template includes: Requirement 1: The analysis report template contains the title, heading and details section; Requirement 2: The detailed content section includes scenes, objects, human actions and time; Requirement 3: If there are abnormal factors, the analysis report template includes an abnormal factor warning and records the time point in the video to be detected.

5. A system for generating an analysis report and a recommended video based on content detection, wherein the system for generating an analysis report and a recommended video based on content detection is used to implement the method for generating an analysis report and a recommended video based on content detection as claimed in any one of claims 1 to 4, It is characterized in that The system comprises: The video playback module is used to obtain the video to be detected, and process the video to be detected to obtain multiple video clips; A video analysis module, configured to obtain a video analysis result based on the plurality of video clips using a multi-dimensional data analysis model; A large model prompt word module, used to obtain a large model prompt word through a prompt word extraction method according to the video analysis result; An effect impact analysis module, used to perform effect analysis and optimization according to the video analysis results and the large model prompt words, to obtain content effect analysis results and optimized large model prompt words; A recommendation module, used to obtain a recommended video according to the content effect analysis result and a recommendation algorithm; The report generation module is used to generate a report as required according to the optimized large model prompt words to obtain a video analysis report.

Images