Video matching method and system based on AI

Abstract

The invention relates to the technical field of video matching, and particularly discloses an AI-based video matching method and system, and the method comprises the steps: obtaining hot videos containing a popularity value in a preset time range, extracting the video features of each hot video, and constructing a hot video library containing a popularity value item and a video feature item; when a to-be-matched video uploaded by a user is received, video features of the to-be-matched video are extracted based on the incremental resource amount quota, video feature items in the hotspot video library are traversed based on the extracted video features, and a target video is obtained through matching; reading a popularity value of the target video as feedback data; when the matching fails, performing popularity prediction on the to-be-matched video based on the AI, and outputting a popularity value; according to the method, a comparison type video matching scheme based on popularity data and an understanding type video analysis scheme based on A I are introduced, a gradient type video analysis architecture is built, the feedback result is accurate, the recognition range is high, and the cost is very low.

Description

Technical Field

[0001] The present invention relates to the technical field of video matching, and specifically to an AI-based video matching method and system. Background Art

[0002] With the popularization of intelligent devices and the progress of network technology, video media data has gradually become the mainstream media data. Many users will post some videos on video media platforms. When they post videos, they may be curious about the quality of their videos. If there is a popularity feedback mechanism, then users can adjust their videos according to the popularity feedback mechanism, which is beneficial to users themselves, the platform and viewers. Therefore, how to provide a popularity feedback mechanism for videos is the technical problem that this application wants to solve. Summary of the Invention

[0003] The purpose of the present invention is to provide an AI-based video matching method and system to solve the problems raised in the above background art.

[0004] To achieve the above purpose, the present invention provides the following technical solutions:

[0005] An AI-based video matching method, the method includes:

[0006] Obtain hot videos with popularity values within a preset time range, extract the video features of each hot video, and construct a hot video library containing popularity value items and video feature items; the video features are a feature set, and each feature corresponds to a resource quantity quota;

[0007] When receiving a video to be matched uploaded by a user, extract the video features of the video to be matched based on the increasing resource quantity quota, traverse the video feature items in the hot video library based on the extracted video features, and match to obtain a target video; the matching condition is that the matching degree reaches a preset threshold;

[0008] Read the popularity value of the target video as feedback data and feedback it to the user;

[0009] When the matching fails, perform popularity prediction on the video to be matched based on AI, output the popularity value, and feedback it to the user.

[0010] As a further solution of the present invention: the step of obtaining hot videos with popularity values within a preset time range, extracting the video features of each hot video, and constructing a hot video library containing popularity value items and video feature items includes:

[0011] Obtain hot videos with popularity values within a preset time range;

[0012] Extract the audio data of the hot video, recognize the audio data to obtain the audio text, and extract keywords from the audio text as the basic features;

[0013] Extract the image data of the hot video, recognize the image data with different resource amount quotas to obtain image features labeled with the resource amount quota as additional features;

[0014] Statistically analyze the basic features and additional features at different levels to obtain video features, and statistically analyze the hot videos, their popularity values, and video features to construct a hot video library.

[0015] As a further solution of the present invention: the step of obtaining hot videos containing popularity values within a preset time range includes:

[0016] Take the current moment as the right endpoint moment, and determine the left endpoint moment according to the preset popularity value threshold;

[0017] Statistically analyze the left endpoint moment and the right endpoint moment corresponding to each popularity value threshold to obtain the time range corresponding to each popularity value threshold; the span of the time range is proportional to the popularity value threshold;

[0018] Select hot videos whose popularity values reach the popularity value threshold within the time range.

[0019] As a further solution of the present invention: the step of extracting the image data of the hot video, recognizing the image data with different resource amount quotas to obtain image features labeled with the resource amount quota as additional features includes:

[0020] For any hot video, extract the image data of the hot video and convert the image data into an image sequence based on the time order;

[0021] Receive the resource amount quota pre-input by the staff;

[0022] For each resource amount quota, randomly select images in the image sequence, randomly determine the recognition area, recognize the images, and extract image features;

[0023] Accumulate the recognition resource amounts of the selected images. When the recognition resource amount reaches the resource amount quota, statistically analyze the extracted image features as the additional features at this level.

[0024] As a further solution of the present invention: when receiving a to-be-matched video uploaded by a user, the step of extracting video features of the to-be-matched video based on an increasing resource amount quota, traversing the video feature items in the hot video library based on the extracted video features, and matching to obtain the target video includes:

[0025] Receive the to-be-matched video uploaded by the user;

[0026] Extract the audio data of the video to be matched, recognize the audio data to obtain the audio text, and extract keywords from the audio text as the basic features;

[0027] Extract the image data of the hot video, determine the resource volume quota, and extract features from the image data based on the resource volume quota to obtain the additional features corresponding to the resource volume quota;

[0028] Traverse the video feature items in the hot video library according to the basic features and additional features, and calculate the matching degree in real time;

[0029] When the matching degree reaches the preset matching degree threshold, read the corresponding hot video as the target video;

[0030] When the matching degrees of all hot videos do not reach the matching degree threshold, increase the resource volume quota, execute in a loop, and select the hot video corresponding to the maximum matching degree as the target video.

[0031] As a further solution of the present invention: the step of traversing the video feature items in the hot video library according to the basic features and additional features and calculating the matching degree in real time includes:

[0032] Compare the basic features with the basic features in the video feature items in the hot video library, screen the hot video library for videos, and obtain a simplified database;

[0033] Compare the additional features with the additional features in the video feature items in the simplified database, and calculate the matching degree in real time;

[0034] Among them, the comparison process of the basic features adopts a set comparison scheme, and the comparison process of the additional features adopts an image comparison scheme.

[0035] The technical solution of the present invention also provides an AI-based video matching system, and the system includes:

[0036] A video library construction module, which is used to obtain hot videos with heat values within a preset time range, extract the video features of each hot video, and construct a hot video library containing heat value items and video feature items; the video features are a feature set, and each feature corresponds to a resource volume quota;

[0037] A target video matching module, which is used to, when receiving a video to be matched uploaded by a user, extract the video features of the video to be matched based on an increasing resource volume quota, traverse the video feature items in the hot video library based on the extracted video features, and match to obtain a target video; the matching condition is that the matching degree reaches a preset threshold;

[0038] A heat value reading module, which is used to read the heat value of the target video as feedback data and feedback it to the user;

[0039] An AI prediction module, which is used to predict the popularity of the video to be matched based on AI when the matching fails, output the popularity value, and feedback it to the user.

[0040] As a further solution of the present invention: the video library construction module includes:

[0041] A video query unit, which is used to obtain hot videos containing popularity values within a preset time range;

[0042] An audio recognition unit, which is used to extract the audio data of the hot video, recognize the audio data to obtain the audio text, and extract keywords from the audio text as the basic features;

[0043] An image recognition unit, which is used to extract the image data of the hot video, recognize the image data with different resource quantity quotas, and obtain image features labeled with the resource quantity quotas as additional features;

[0044] A feature statistics unit, which is used to count the basic features and additional features at different levels to obtain video features, count the hot videos and their popularity values and video features, and construct a hot video library.

[0045] As a further solution of the present invention: the video query unit includes:

[0046] An endpoint determination subunit, which is used to use the current moment as the right endpoint moment and determine the left endpoint moment according to the preset popularity value threshold;

[0047] A range determination subunit, which is used to count the left endpoint moment and the right endpoint moment corresponding to each popularity value threshold to obtain the time range corresponding to each popularity value threshold; the span of the time range is proportional to the popularity value threshold;

[0048] A video selection subunit, which is used to select hot videos whose popularity values reach the popularity value threshold within the time range.

[0049] As a further solution of the present invention: the image recognition unit includes:

[0050] A sequence generation subunit, which is used to extract the image data of any hot video and convert the image data into an image sequence based on the time sequence for any hot video;

[0051] A quota receiving subunit, which is used to receive the resource quantity quota pre-input by the staff;

[0052] An identification execution subunit, which is used to randomly select images in the image sequence for each resource quantity quota, randomly determine the identification area, recognize the images, and extract the image features;

[0053] An accumulation subunit is configured to accumulate the recognition resource amounts of the selected images. When the recognition resource amount reaches the resource amount quota, the extracted image features are statistically analyzed as the additional features at this level.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention introduces a comparison-based video matching scheme based on heat data and an understanding-based video analysis scheme based on AI, constructs a gradient video analysis architecture, with accurate feedback results, high recognition breadth, and low cost. Description of the Drawings

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0056] Figure 1 It is a flowchart of a video matching method based on AI.

[0057] Figure 2 It is a block diagram of the composition structure of a video matching system based on AI. Detailed Embodiments

[0058] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] Figure 1 It is a flowchart of a video matching method based on AI. In an embodiment of the present invention, a video matching method based on AI includes:

[0060] Step S100: Obtain hot videos with heat values within a preset time range, extract the video features of each hot video, and construct a hot video library containing heat value items and video feature items; the video features are a feature set, and each feature corresponds to a resource amount quota.

[0061] This application is applied to a media data management platform. For example, in some existing short video apps, the heat values of the videos in the short video apps are known data. Each platform will analyze the heat values of each video. The simplest way is to use the view count as the heat value; the time range can be one week or half a month. The hot videos are the videos whose heat values reach the preset heat value threshold. By performing recognition and analysis on the hot videos, the video features of each hot video can be extracted, and the heat values and video features of the hot videos can be statistically analyzed to construct a hot video library within the time range.

[0062] It is worth mentioning that the video feature is a feature set that contains multiple features, and each feature corresponds to a different resource quota. The resource quota represents how much resources are invested in extracting video features. The larger the resource quota, the more resources are invested, the more video features are extracted, and the easier it is for the subsequent matching process to succeed.

[0063] Step S200: When receiving a to-be-matched video uploaded by a user, extract the video features of the to-be-matched video based on an increasing resource quota, traverse the video feature items in the hot video library based on the extracted video features, and obtain a target video through matching; the matching condition is that the matching degree reaches a preset threshold.

[0064] When receiving a to-be-matched video uploaded by a user, first extract features of the to-be-matched video with a relatively small resource quota to obtain fewer video features. Correspondingly, compare them with the video features of each hot video in the hot video library under the same resource quota, calculate the matching degree. If the matching degree is high enough, read the corresponding hot video as the matched target video. If the matching degree is not high enough, increase the resource quota and perform another match. This method can optimize the comparison process and is very likely to succeed in matching under a low resource quota. At this time, there is no need to perform subsequent matching processes with a high resource quota. When the number of to-be-matched videos is large enough, the cost saved will be very significant.

[0065] It is worth mentioning that the resource quota involved in this application is the same resource quota. Generally, a gradient resource quota is adopted. For example, five resource quotas are preset by staff, and these five resource quotas are increasing. Whether it is for feature extraction of hot videos or for feature extraction of to-be-matched videos, these five pre-determined resource quotas are used.

[0066] Step S300: Read the popularity value of the target video as feedback data and feedback it to the user.

[0067] Step S400: When the matching fails, predict the popularity of the to-be-matched video based on AI, output the popularity value, and feedback it to the user.

[0068] After matching the target video, read the popularity value of the target video and feedback it to the user. However, there is such a situation where the maximum matching degree obtained in the matching process under the maximum resource amount limit is still very small. At this time, even if a target video is matched, the reference significance of this target video is very small, and it is difficult to determine its popularity value. For this, this application introduces an AI recognition process to predict the popularity of such videos. The existing AI has extremely strong capabilities. As long as a large number of hot videos are fed to it, even for unknown new videos, it can make inferences based on known data and obtain a predicted popularity value. This prediction process with the help of AI greatly enriches the prediction breadth.

[0069] Combined with the above content, when the user uses this platform and inputs a video to be matched, this platform can feedback a popularity value. The calculation process of the popularity value first is a comparison and prediction process based on hot data, and then is an understanding and prediction process based on AI. The resource consumption of the understanding and prediction process based on AI is very large, and the resource consumption of the comparison and prediction process based on hot data is less. In addition, the comparison and prediction process based on hot data is also a gradient prediction process. Therefore, this application provides a video analysis architecture that gradually improves the prediction ability and optimizes the cost.

[0070] As a preferred embodiment of the technical solution of the present invention, regarding step S100, the steps of obtaining hot videos with popularity values within a preset time range, extracting video features of each hot video, and constructing a hot video library containing popularity value items and video feature items include:

[0071] Obtain hot videos with popularity values within a preset time range;

[0072] Extract the audio data of the hot video, perform recognition on the audio data to obtain an audio text, and extract keywords in the audio text as basic features;

[0073] Extract the image data of the hot video, perform recognition on the image data with different resource amount limits to obtain image features labeled with resource amount limits as additional features;

[0074] Statistically analyze the basic features and additional features at different levels to obtain video features, and statistically analyze the hot videos and their popularity values and video features to construct a hot video library.

[0075] The time range is preset. Obtain hot videos with popularity values within a preset time range. The video itself contains an audio channel and an image channel. Extract the audio data of the hot video, perform recognition on the audio data to obtain an audio text. The process of converting audio to text belongs to known technology and will not be elaborated here. Extract keywords in the audio text as basic features. The audio recognition process is very simple. The features (keywords) extracted from the audio can be used as the basic features of the video.

[0076] Further, extract the image data of the hot video, perform identification with different resource quantity quotas on the image data, obtain the image features labeled with the resource quantity quotas as additional features. The image recognition process is relatively complex and consumes resources. In the image recognition stage of the present application, a differential recognition process is introduced to perform identification with different resource quantity quotas on the image data, obtain the image features labeled with the resource quantity quotas as additional features. The simplest solution is to determine an area ratio for different resource quantity quotas. The larger the resource quantity quota, the larger the area ratio. Intercept a small area outward from the center point of the image. The ratio of the small area to the image is the area ratio. Extract features from the small area to obtain the video features corresponding to the resource quantity quota. Thus, a technical solution is obtained where the larger the resource quantity quota, the more image features are extracted. The obtained image features are called additional features.

[0077] Finally, count the basic features and additional features at different levels to obtain video features, count the hot videos, their popularity values, and video features, and construct a hot video library.

[0078] As a preferred embodiment of the technical solution of the present invention, the step of obtaining hot videos containing popularity values within a preset time range includes:

[0079] Take the current moment as the right endpoint moment, and determine the left endpoint moment according to the preset popularity value threshold;

[0080] Count the left endpoint moment and the right endpoint moment corresponding to each popularity value threshold to obtain the time range corresponding to each popularity value threshold; the span of the time range is proportional to the popularity value threshold;

[0081] Select hot videos whose popularity values reach the popularity value threshold within the time range.

[0082] In an example of the technical solution of the present invention, the selection process of hot videos is limited. Specifically, the parameter of the time range is limited. Take the current moment as the right endpoint moment, determine the left endpoint moment according to the preset popularity value threshold, count the left endpoint moment and the right endpoint moment corresponding to each popularity value threshold to obtain the time range corresponding to each popularity value threshold. The larger the popularity value threshold, the larger the time range. Select hot videos whose popularity values reach the popularity value threshold within the time range. The practical significance of this process is to select videos with higher popularity values as hot videos within a longer time range, such as videos with millions of views within one or two months. Videos with not very high popularity values can also be used as hot videos within a shorter time range, such as videos with tens of thousands of views within one or two days.

[0083] As a preferred embodiment of the technical solution of the present invention, the steps of extracting the image data of the hot video, identifying the image data with different resource amount quotas, and obtaining the image features labeled with the resource amount quotas as additional features include:

[0084] For any hot video, extract the image data of the hot video and convert the image data into an image sequence based on the time sequence;

[0085] Receive the resource amount quota pre-input by the staff;

[0086] For each resource amount quota, randomly select images in the image sequence, randomly determine the recognition area, recognize the images, and extract the image features;

[0087] Accumulate the recognition resource amount of the selected images. When the recognition resource amount reaches the resource amount quota, count the extracted image features as the additional features of this level.

[0088] In an example of the technical solution of the present invention, the image recognition process is described. For any hot video, extract the image data of the hot video and convert the image data into an image sequence based on the time sequence; receive the resource amount quota pre-input by the staff, which means that the resource amount quota is determined by the staff when analyzing the hot video. Once determined, all resource amount quotas in the entire matching process will no longer be changed.

[0089] Furthermore, for each resource amount quota, the present application introduces a more random image recognition scheme applied to videos. The video itself contains multiple images. Randomly select images in the image sequence, then randomly determine the recognition area, recognize the images, and extract the image features. At this time, there will be a resource amount input for the recognition of each image. This process is executed in a loop. For each recognized image, the consumed resource amount increases by one point. Accumulate the recognition resource amount of the selected images. When the recognition resource amount reaches the resource amount quota, count the extracted image features as the additional features of this level.

[0090] The above recognition scheme actually increases the randomness of the image recognition process. Compared with the fixed recognition scheme, the random recognition scheme may occasionally improve the efficiency or may also reduce the efficiency. It can be used as an alternative scheme.

[0091] As a preferred embodiment of the technical solution of the present invention, when receiving the video to be matched uploaded by the user, the steps of extracting the video features of the video to be matched based on the increasing resource amount quota, traversing the video feature items in the hot video library based on the extracted video features, and matching to obtain the target video include:

[0092] Receive the video to be matched uploaded by the user;

[0093] Extract the audio data of the video to be matched, recognize the audio data to obtain the audio text, and extract keywords from the audio text as the basic features;

[0094] Extract the image data of the hot video, determine the resource quantity quota, and perform feature extraction on the image data based on the resource quantity quota to obtain the additional features corresponding to the resource quantity quota;

[0095] Traverse the video feature items in the hot video library according to the basic features and additional features, and calculate the matching degree in real time;

[0096] When the matching degree reaches the preset matching degree threshold, read the corresponding hot video as the target video;

[0097] When the matching degrees of all hot videos do not reach the matching degree threshold, increase the resource quantity quota, execute in a loop, select the hot video corresponding to the maximum matching degree as the target video.

[0098] In an example of the technical solution of the present invention, the specific matching process is described. Receive the video to be matched uploaded by the user, and process the video to be matched using the same scheme as the hot video. Extract the audio data of the video to be matched, recognize the audio data to obtain the audio text, extract keywords from the audio text as the basic features, extract the image data of the hot video, determine the resource quantity quota, perform feature extraction on the image data based on the resource quantity quota to obtain the additional features corresponding to the resource quantity quota, traverse the video feature items in the hot video library according to the basic features and additional features, calculate the matching degree in real time. When the matching degree is large enough, read the corresponding hot video as the target video. When, under the corresponding resource quantity quota condition, all matching degrees are not large enough, increase the resource quantity quota and perform another loop recognition (only loop the image feature comparison process). When all matching degrees under the condition of the highest resource quantity quota are not large enough, select the hot video corresponding to the maximum matching degree as the target video. At this time, the target video is only a reference, and the specific hot spot prediction process is completed by AI.

[0099] As a preferred embodiment of the technical solution of the present invention, the step of traversing the video feature items in the hot video library according to the basic features and additional features and calculating the matching degree in real time includes:

[0100] Compare the basic features with the basic features in the video feature items in the hot video library, screen the hot video library to obtain a simplified database;

[0101] Compare the additional features with the additional features in the video feature items in the simplified database, and calculate the matching degree in real time;

[0102] Among them, the comparison process of the basic features adopts a set comparison scheme, and the comparison process of the additional features adopts an image comparison scheme.

[0103] In an example of the technical solution of the present invention, the application processes of the basic features and the additional features are described. The basic features are compared with the basic features in the video feature items in the hot video library, and the hot video library is screened for videos to obtain a simplified database. The additional features are compared with the additional features in the video feature items in the simplified database, and the matching degree is calculated in real time.

[0104] Among them, when comparing the basic features with the basic features in the video feature items in the hot video library, since the basic features are a set of keywords, the comparison scheme adopted is a set comparison scheme. For example, the ratio of the intersection to the union is calculated as the first similarity. The additional features are image features, and an image comparison algorithm is adopted. For example, the image is divided into multiple small regions, and the small regions are compared pairwise to obtain the similarity of the small regions, and the total value of the similarity of the small regions is calculated to obtain the second similarity. Based on the preset weight, the first similarity and the second similarity are added to obtain the matching degree.

[0105] Figure 2 For the composition structure block diagram of the AI-based video matching system, the technical solution of the present invention also provides an AI-based video matching system. The system 10 includes:

[0106] A video library construction module 11, configured to obtain hot videos containing heat values within a preset time range, extract the video features of each hot video, and construct a hot video library containing heat value items and video feature items; the video features are a feature set, and each feature corresponds to a resource amount quota;

[0107] A target video matching module 12, configured to, when receiving a to-be-matched video uploaded by a user, extract the video features of the to-be-matched video based on the increasing resource amount quota, traverse the video feature items in the hot video library based on the extracted video features, and match to obtain a target video; the matching condition is that the matching degree reaches a preset threshold;

[0108] A heat value reading module 13, configured to read the heat value of the target video as feedback data and feedback it to the user;

[0109] An AI prediction module 14, configured to, when the matching fails, perform heat prediction on the to-be-matched video based on AI, output a heat value, and feedback it to the user.

[0110] Further, the video library construction module 11 includes:

[0111] A video query unit, configured to obtain hot videos containing heat values within a preset time range;

[0112] An audio recognition unit, which is used to extract the audio data of a hot video, recognize the audio data to obtain an audio text, and extract keywords from the audio text as basic features;

[0113] An image recognition unit, which is used to extract the image data of a hot video, recognize the image data with different resource quantity quotas to obtain image features labeled with the resource quantity quotas as additional features;

[0114] A feature statistics unit, which is used to count the basic features and additional features at different levels to obtain video features, count the hot video, its popularity value and video features, and construct a hot video library.

[0115] Specifically, the video query unit includes:

[0116] An endpoint determination subunit, which is used to use the current moment as the right endpoint moment and determine the left endpoint moment according to a preset popularity value threshold;

[0117] A range determination subunit, which is used to count the left endpoint moment and the right endpoint moment corresponding to each popularity value threshold to obtain the time range corresponding to each popularity value threshold; the span of the time range is proportional to the popularity value threshold;

[0118] A video selection subunit, which is used to select hot videos whose popularity values reach the popularity value threshold within the time range.

[0119] Furthermore, the image recognition unit includes:

[0120] A sequence generation subunit, which is used to extract the image data of any hot video and convert the image data into an image sequence based on the time sequence;

[0121] A quota receiving subunit, which is used to receive the resource quantity quota pre-input by the staff;

[0122] An identification execution subunit, which is used to randomly select images in the image sequence for each resource quantity quota, randomly determine the identification area, identify the images, and extract image features;

[0123] An accumulation subunit, which is used to accumulate the identification resources of the selected images, and when the identification resources reach the resource quantity quota, count the extracted image features as the additional features at this level.

[0124] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present invention by the same token.

Claims

1. An AI-based video matching method, characterized in that, The method includes: Obtaining hot videos with heat values within a preset time range, extracting video features of each hot video, and constructing a hot video library containing heat value items and video feature items; the video features are a feature set, and each feature corresponds to a resource amount quota; When receiving a to-be-matched video uploaded by a user, extracting video features of the to-be-matched video based on an increasing resource amount quota, traversing the video feature items in the hot video library based on the extracted video features, and obtaining a target video through matching; the matching condition is that the matching degree reaches a preset threshold; Reading the heat value of the target video as feedback data and feeding it back to the user; When the matching fails, performing heat prediction on the to-be-matched video based on AI, outputting a heat value, and feeding it back to the user.

2. The AI-based video matching method according to claim 1, wherein The steps of obtaining hot videos with heat values within a preset time range, extracting video features of each hot video, and constructing a hot video library containing heat value items and video feature items include: Obtaining hot videos with heat values within a preset time range; Extracting the audio data of the hot video, identifying the audio data to obtain an audio text, and extracting keywords from the audio text as basic features; Extracting the image data of the hot video, identifying the image data with different resource amount quotas to obtain image features labeled with resource amount quotas as additional features; Statistically analyzing the basic features and additional features at different levels to obtain video features, statistically analyzing the hot videos and their heat values and video features, and constructing a hot video library.

3. The AI-based video matching method according to claim 2, wherein The steps of obtaining hot videos with heat values within a preset time range include: Taking the current moment as the right endpoint moment and determining the left endpoint moment according to a preset heat value threshold; Statistically analyzing the left endpoint moment and the right endpoint moment corresponding to each heat value threshold to obtain the time range corresponding to each heat value threshold; the span of the time range is proportional to the heat value threshold; Selecting hot videos with heat values reaching the heat value threshold within the time range.

4. The AI-based video matching method according to claim 3, wherein The steps of extracting the image data of the hot video, identifying the image data with different resource amount quotas to obtain image features labeled with resource amount quotas as additional features include: For any hot video, extracting the image data of the hot video and converting the image data into an image sequence based on the time sequence; Receiving the resource amount quota pre-input by the staff; For each resource amount quota, randomly selecting images in the image sequence, randomly determining the recognition area, identifying the images, and extracting image features; Accumulating the recognition resource amounts of the selected images, and when the recognition resource amount reaches the resource amount quota, statistically analyzing the extracted image features as the additional features at this level.

5. The AI-based video matching method according to claim 2, wherein The steps of, when receiving a to-be-matched video uploaded by a user, extracting video features of the to-be-matched video based on an increasing resource amount quota, traversing the video feature items in the hot video library based on the extracted video features, and obtaining a target video through matching include: Receiving the to-be-matched video uploaded by the user; Extracting the audio data of the to-be-matched video, identifying the audio data to obtain an audio text, and extracting keywords from the audio text as basic features; Extract the image data of the hot video, determine the resource volume quota, perform feature extraction on the image data based on the resource volume quota, and obtain additional features corresponding to the resource volume quota; Traverse the video feature items in the hot video library according to the basic features and additional features, and calculate the matching degree in real time; When the matching degree reaches the preset matching degree threshold, read the corresponding hot video as the target video; When the matching degrees of all hot videos do not reach the matching degree threshold, increase the resource volume quota, execute in a loop, and select the hot video corresponding to the maximum matching degree as the target video.

6. The AI-based video matching method according to claim 5, wherein The step of traversing the video feature items in the hot video library according to the basic features and additional features and calculating the matching degree in real time includes: Compare the basic features with the basic features in the video feature items in the hot video library, screen the videos in the hot video library, and obtain a simplified database; Compare the additional features with the additional features in the video feature items in the simplified database, and calculate the matching degree in real time; Among them, the comparison process of the basic features adopts a set comparison scheme, and the comparison process of the additional features adopts an image comparison scheme.

7. An AI-based video matching system, characterized in that, The system includes: A video library construction module, which is used to obtain hot videos containing heat values within a preset time range, extract the video features of each hot video, and construct a hot video library containing heat value items and video feature items; the video features are a feature set, and each feature corresponds to a resource volume quota; A target video matching module, which is used to, when receiving a video to be matched uploaded by a user, extract the video features of the video to be matched based on an increasing resource volume quota, traverse the video feature items in the hot video library based on the extracted video features, and match to obtain a target video; the matching condition is that the matching degree reaches a preset threshold; A heat value reading module, which is used to read the heat value of the target video as feedback data and feedback it to the user; An AI prediction module, which is used to, when the matching fails, perform heat prediction on the video to be matched based on AI, output the heat value, and feedback it to the user.

8. The AI-based video matching system according to claim 7, wherein The video library construction module includes: A video query unit, which is used to obtain hot videos containing heat values within a preset time range; An audio recognition unit, which is used to extract the audio data of the hot video, recognize the audio data to obtain an audio text, and extract keywords in the audio text as basic features; An image recognition unit, which is used to extract the image data of the hot video, recognize the image data with different resource volume quotas, and obtain image features labeled with the resource volume quota as additional features; A feature statistics unit, which is used to count the basic features and additional features at different levels, obtain video features, count the hot videos and their heat values and video features, and construct a hot video library.

9. The AI-based video matching system according to claim 8, wherein The video query unit includes: An endpoint determination subunit, which is used to use the current moment as the right endpoint moment and determine the left endpoint moment according to the preset heat value threshold; A range determination subunit, which is used to count the left endpoint moment and the right endpoint moment corresponding to each heat value threshold, and obtain the time range corresponding to each heat value threshold; the span of the time range is proportional to the heat value threshold; A video selection subunit, which is used to select hot videos whose heat values reach the heat value threshold within the time range.

10. The AI-based video matching system according to claim 9, wherein The image recognition unit includes: A sequence generation subunit, configured to extract image data of any hot video and convert the image data into an image sequence based on chronological order; An amount receiving subunit, configured to receive the resource amount limit pre-input by a staff member; An identification execution subunit, configured to randomly select an image from the image sequence for each resource amount limit, randomly determine the identification area, perform identification on the image, and extract image features; An accumulation subunit, configured to accumulate the recognized resource amounts of the selected images, and when the recognized resource amount reaches the resource amount limit, count the extracted image features as additional features at this level.

Images