Apparatus and Method for Generating Personalized Highlight

Abstract

A user-customized highlight generation device and method are disclosed. A user-customized highlight generation device according to an embodiment of the present invention includes a memory on which at least one program is recorded and a processor that executes the program. The program generates highlight candidate lists based on an artificial intelligence model that is pre-trained for each of the video data, background sound data, and commentary data separated from game data, and can generate highlights with user preferences applied to the highlight lists generated based on the generated highlight candidate lists.

Description

Technology Field

[0001] The described embodiments relate to a technology for generating highlights in games or e-sports. Background Technology

[0002] E-sports is the sport of greatest interest to the younger generation under the age of 30, following soccer and baseball. With various professional leagues organized by game and region, as well as numerous amateur leagues, and its adoption as an official event in the Asian Games, it is steadily solidifying its position as a sport. For instance, the number of viewers for live e-sports broadcasts in 2021 reached 662.6 million, a figure larger than the total viewership of the U.S. Major League Baseball, NFL, and NBA combined.

[0003] In e-sports, it is common to enjoy game footage on video platforms such as YouTube and Twitter rather than traditional broadcasting platforms, and the advantage is that one can watch sports competitions anytime and anywhere using mobile electronic devices. In other words, as the demand for on-demand viewing—enjoying videos whenever and wherever users want, in addition to real-time viewing—is very high, the demand for the production and utilization of e-sports content has increased significantly, and the global e-sports industry is achieving high-speed growth of over 10% annually.

[0004] As interest in e-sports and games grows and demand continues to rise, highlight services that allow users to quickly view key game content are also gaining popularity.

[0005] However, since most highlights are generated manually by individuals capable of understanding and analyzing match content, limitations in manpower and cost result in highlight videos being produced primarily for major matches of interest to a large number of users. In other words, the current situation does not provide highlight services that reflect the individual preferences of users.

[0006] Meanwhile, for some games, it is possible to automatically generate highlight videos by downloading gameplay data via APIs provided by the game company; however, since this depends on the policies of the API provider, it is difficult to generalize and apply it to various other game contents that do not provide APIs. The problem to be solved

[0007] The described embodiments are intended to reduce manpower and costs for generating real-time match highlights for e-sports matches and gameplay content.

[0008] The described embodiments are intended to allow the individual preferences of users to be reflected when generating real-time match highlights for e-sports matches and gameplay content.

[0009] The described embodiment is intended to enable game companies to generate highlights by applying it to various game content that does not support APIs. means of solving the problem

[0010] A user-customized highlight generation device according to an embodiment includes a memory on which at least one program is recorded and a processor that executes the program, wherein the program generates highlight candidate lists based on an artificial intelligence model that has been pre-trained for each of the video data, background sound data, and commentary data separated from the game data, and can generate highlights with user preferences applied to the highlight lists generated based on the generated highlight candidate lists.

[0011] At this time, the program can extract frames from video data at a predetermined period, generate a list of major event occurrences based on a video learning model from the extracted frames, generate a list of major event occurrence times based on analysis results of at least one predetermined video attribute change and character attribute change from the extracted frames, and generate a highlight candidate list by combining the list of major event occurrences and the list of major event occurrence intervals.

[0012] At this time, the video learning model may be pre-trained with video data including at least one of a character, indicator, skill, function, ability, and movement path.

[0013] At this time, the image attribute includes at least one of color, brightness, and saturation, and the character attribute may include at least one of movement, movement distance, and number.

[0014] At this time, the program can divide background sound data into intervals, generate a list of major event occurrences based on a background sound learning model from the divided background sound data, generate a list of major event occurrence times based on the analysis results of background sound frequency characteristics from the divided background sound data, and generate a list of highlight candidates by combining the list of major event occurrences and the list of major event occurrence intervals.

[0015] At this time, the background sound learning model may be pre-trained with acoustic data including at least one of the game background sound and the battle sound effect.

[0016] At this time, the program can divide the commentary data into intervals, generate a list of major event occurrences based on a commentary learning model from the divided commentary data, generate a list of major event occurrence times based on an analysis result for at least one predetermined voice attribute from the divided commentary data, and generate a highlight candidate list by combining the list of major event occurrences and the list of major event occurrence intervals.

[0017] At this time, the specified voice attribute may include at least one of intensity, pitch, and speed.

[0018] At this time, the program can extract a user highlight list with user preferences applied from the highlight list, generate video clips for each of the extracted user highlight lists, and combine the generated video clips to create a user highlight video.

[0019] At this time, at least one of the length and subtitles of the video clip and highlight video can be edited through the previewer.

[0020] A user-customized highlight generation method according to an embodiment may include the steps of separating video data, background sound data, and commentary data from game data; generating highlight candidate lists based on an artificial intelligence model that has been pre-trained for each of the separated video data, background sound data, and commentary data; integrating the generated highlight candidate lists to generate a highlight list; and generating highlights with user preferences applied from the generated highlight list.

[0021] At this time, the step of generating highlight candidate lists may include the step of extracting frames from video data at a predetermined period, the step of generating a list of major event occurrences based on a video learning model from the extracted frames, the step of generating a list of major event occurrence times based on analysis results of at least one predetermined video attribute change and character attribute change from the extracted frames, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0022] At this time, the video learning model may be pre-trained with video data including at least one of a character, indicator, skill, function, ability, and movement path.

[0023] At this time, the image attribute includes at least one of color, brightness, and saturation, and the character attribute may include at least one of movement, movement distance, and number.

[0024] At this time, the step of generating highlight candidate lists may include the step of dividing background sound data into intervals, the step of generating a list of major event occurrences based on a background sound learning model from the divided background sound data, the step of generating a list of major event occurrence times based on the analysis results of background sound frequency characteristics from the divided background sound data, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0025] At this time, the background sound learning model may be pre-trained with acoustic data including at least one of the game background sound and the battle sound effect.

[0026] At this time, the step of generating highlight candidate lists may include the step of dividing commentary data into intervals, the step of generating a list of major event occurrences based on a commentary learning model from the divided commentary data, the step of generating a list of major event occurrence times based on an analysis result for at least one predetermined voice attribute from the divided commentary data, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0027] At this time, the specified voice attribute may include at least one of intensity, pitch, and speed.

[0028] At this time, the step of generating highlights may include the step of extracting a user highlight list with user preferences applied from a highlight list, the step of generating video clips for each of the extracted user highlight lists, and the step of integrating the generated video clips to generate a user highlight video.

[0029] At this time, at least one of the length and subtitles of the video clip and highlight video can be edited through the previewer. Effects of the invention

[0030] According to the described embodiments, manpower and costs for generating real-time match highlights of e-sports matches and gameplay content can be reduced.

[0031] According to the described embodiments, individual user preferences can be reflected when generating real-time match highlights for e-sports matches and gameplay content. That is, by recognizing detailed information such as major in-game events and situations involving game characters that occur during a match, highlight videos of events and situations desired by the user can be extracted based on this information, thereby generating personalized highlight videos that reflect user preferences. Consequently, customized highlight content can be produced and provided to meet real-time or on-demand viewing demands for a large volume of matches, ranging from major tournaments like the League of Legends World Championship to amateur matches played by ordinary individuals.

[0032] According to the described embodiments, various service support is possible, such as player training coaching, gamer play analysis, and records, which require immediate verification of game progress and the provision of accurate feedback.

[0033] According to the described embodiments, highlights can be generated even for various game contents that do not support APIs from game companies. In other words, without relying on game company APIs and data provision, game highlights are automatically generated by applying artificial intelligence or video and audio signal processing data analysis technologies to players' gameplay videos and commentary to identify key plays and moments of the match. This reduces the burden of service development by lowering dependence on game developers for highlight video generation and facilitates the generation of highlight videos for a large volume of game footage and various games. Brief explanation of the drawing

[0034] FIG. 1 is a schematic block diagram of a user-customized highlight generation device according to an embodiment. FIG. 2 is a detailed operation configuration diagram of a video / audio data extraction unit according to an embodiment. FIG. 3 is a detailed operational configuration diagram of a highlight list extraction unit according to an embodiment. FIG. 4 is a diagram illustrating the detailed operation of extracting major game events based on video data according to an embodiment. FIG. 5 is a diagram illustrating the detailed operation of extracting major match events based on game sound data according to an embodiment. FIG. 6 is a diagram illustrating the detailed operation of extracting major match events based on game narration data according to an embodiment. FIG. 7 is a diagram illustrating the detailed operation of the highlight automatic generation unit according to an embodiment. FIG. 8 is a flowchart illustrating a method for generating user-customized highlights according to an embodiment. FIG. 9 is a diagram showing the configuration of a computer system according to an embodiment. Specific details for implementing the invention

[0035] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0036] Although terms such as "first" or "second" are used to describe various components, these components are not limited by such terms. Such terms may be used merely to distinguish one component from another. Accordingly, the first component mentioned below may be the second component within the technical scope of the present invention.

[0037] The terms used herein are for describing embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used in this specification, "comprises" or "comprising" implies that the mentioned component or step does not exclude the presence or addition of one or more other components or steps.

[0038] Unless otherwise defined, all terms used in this specification may be interpreted in a sense that is commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0039] FIG. 1 is a schematic block diagram of a user-customized highlight generation device according to an embodiment, FIG. 2 is a detailed operation diagram of a video / audio data extraction unit according to an embodiment, and FIG. 3 is a detailed operation diagram of a highlight list extraction unit according to an embodiment.

[0040] Referring to FIG. 1, the user-customized highlight generation device according to the embodiment receives e-sports match data and game play data, such as game videos, sounds, and commentary, in real-time or non-real-time, analyzes them based on artificial intelligence, image processing, and signal processing, and automatically generates highlights that reflect user preferences.

[0041] To this end, the user-customized highlight generation device according to the embodiment may include a video / audio data extraction unit (110), a highlight list extraction unit (120), and a highlight automatic generation unit (130).

[0042] The video / audio data extraction unit (110) can extract video data, background sound data, and commentary data separated from the input game data.

[0043] At this time, the game data may include, for example, game videos, real-time game data, and real-time game commentary data, as illustrated in FIG. 1.

[0044] Referring to FIG. 2, the video / audio data extraction unit (110) can separate input game data into audio data and video data (111), and extract game background sound data and game commentary data (or game narration) from the audio data, respectively (112).

[0045] The highlight list generation unit (120) generates highlight candidate lists based on an artificial intelligence model that has been pre-trained for each of the video data, background sound data, and commentary data, and generates a highlight list by integrating the highlight candidate lists.

[0046] In other words, by separating video data, match commentary data, and game background music data, and applying methods suitable for the characteristics of each data set, it is possible to detect key events and points that can serve as highlights.

[0047] Referring to FIG. 3, the highlight list generation unit (120) can generate a highlight candidate list by extracting a major match event based on video data (121) using video data, generate a highlight candidate list by extracting a major match event based on game sound data (122) using game background sound and sound effects, and generate a highlight candidate list by extracting a major match event based on game narration data (123) using commentary from game commentators or narration data provided by the game.

[0048] A detailed description of the generation of such a highlight candidate list will be provided later with reference to FIGS. 4 to 6.

[0049] Additionally, the highlight list generation unit (120) combines the candidate lists extracted for each data to generate a single integrated highlight list.

[0050] Here, the highlight list may include additional information such as the time of occurrence of major events and moments in the game, associated game characters, and details of the events that occurred.

[0051] Referring again to FIG. 1, the highlight automatic generation unit (130) can generate highlights by applying user preferences to the generated highlight list.

[0052] That is, the highlight automatic generation unit (130) generates a highlight video by selecting key moments that the user wants or needs from an integrated highlight list. For example, if the user selects a game target that the user supports or is interested in, such as a team or champion, or a major event that occurs during the game, a highlight video of the key moments containing relevant information is automatically generated.

[0053] A detailed explanation of the generation of these highlights will be described later with reference to Fig. 7.

[0054] FIG. 4 is a diagram illustrating the detailed operation of extracting major game events based on video data according to an embodiment.

[0055] Referring to FIG. 4, frames are extracted from video data at a predetermined period (210).

[0056] Then, a list of major events can be generated (220) based on a video learning model from the extracted frames.

[0057] In other words, it analyzes frames to detect major events occurring during gameplay within the game screen in detail. This game event detection identifies screen areas that provide various information within the game, enabling the automatic inference and acquisition of detailed information about the generated events through an image learning model.

[0058] At this time, the video learning model (221) may be pre-trained with video data including at least one of a character, an indicator, a skill, a function, an ability, and a movement path.

[0059] Meanwhile, a list of major event occurrence times (230) can be generated based on the analysis results of at least one predetermined image attribute change and character attribute change from the extracted frame.

[0060] At this time, the image attribute includes at least one of color, brightness, and saturation, and the character attribute may include at least one of movement, movement distance, and number.

[0061] In other words, the timing of key gameplay events can be estimated by analyzing changes in color, brightness, and saturation in general gameplay footage and situations where combat or confrontation occurs, as well as character movements, movement distances, and the number of characters within the screen, using image processing algorithms.

[0062] Then, a list of highlight candidates (240) can be generated by integrating the list of major event occurrences and the list of major event occurrence intervals.

[0063] At this time, according to the embodiment, both AI-based event occurrence inference (220) and image processing-based game screen feature analysis (230) may be performed, and only one may be performed as needed.

[0064] FIG. 5 is a diagram illustrating the detailed operation of extracting major match events based on game sound data according to an embodiment.

[0065] Referring to FIG. 5, background sound data can be divided at regular intervals (310). That is, game play sound data of the game in progress is received and divided at regular intervals by moving the data input window, so that artificial intelligence-based data analysis and sound processing-based data analysis are performed, thereby allowing major events and moments of the game to be searched.

[0066] That is, a list of major events can be generated (320) based on a background sound learning model from the divided background sound data.

[0067] At this time, the background sound learning model (321) may be pre-learned with acoustic data including at least one of the game background sound and the battle sound effect.

[0068] Additionally, a list of major event occurrence times (330) can be generated based on the analysis results of background sound frequency characteristics from the segmented background sound data. That is, sound processing-based data analysis infers or detects the occurrence of major combat sections based on the existing background sound frequency characteristics and the frequency characteristics of combat sound effects.

[0069] Then, a list of highlight candidates (340) can be generated by combining a list of major event occurrences and a list of major event occurrence intervals.

[0070] At this time, according to the embodiment, both AI-based major sound effect generation inference (320) and signal processing-based game background sound feature analysis (330) may be performed, and only one may be performed as needed.

[0071] FIG. 6 is a diagram illustrating the detailed operation of extracting major match events based on game narration data according to an embodiment.

[0072] Referring to FIG. 6, commentary data can be divided at regular intervals (410). That is, commentator or in-game voice explanation data is input and divided by moving the data input window at regular intervals, so that key moments of the game are searched by applying artificial intelligence-based data analysis (420) and voice processing-based data analysis (430) methods.

[0073] That is, a list of major event occurrences (420) can be generated based on the commentary learning model (421) from the divided commentary data. In other words, the major words explained by the commentator or the commentary sounds occurring in the game are analyzed based on artificial intelligence to infer major moments and game event occurrences.

[0074] In addition, a list of major event occurrence times (430) can be generated based on the analysis results for at least one predetermined voice attribute based on the frequency characteristics of human voice from the divided commentary data.

[0075] At this time, the specified voice attribute may include at least one of intensity, pitch, and speed.

[0076] Then, a list of highlight candidates (440) can be generated by combining the list of major event occurrences and the list of major event occurrence intervals.

[0077] At this time, according to the embodiment, both AI-based major explanatory sound generation inference (420) and signal processing-based speech feature analysis (430) may be performed, and only one may be performed as needed.

[0078] FIG. 7 is a diagram illustrating the detailed operation of the highlight automatic generation unit according to an embodiment.

[0079] Referring to Fig. 7, a highlight image can be generated from a highlight list.

[0080] In other words, the highlight list contains information such as the time, section, and event details where each highlight point occurred. For instance, from the generated highlight list, users can select only major events involving their favorite champions, or choose to create highlights specifically by selecting intense battles such as teamfights. Additionally, it is possible to generate summary highlight videos for the entire match by selecting game-specific key gameplay events that help grasp the main flow of the game, such as character KDA (Kill, Death, Assistance) events or neutral character and structure removal events in MOBA (Multiplayer Online Battle Arena) games.

[0081] To this end, as the user preference option is selected (410), a user highlight list with the user preference applied is extracted (420) from the highlight list.

[0082] Then, video clips (430) are generated for each of the extracted user highlight lists.

[0083] At this time, the video clip can have at least one of its length and subtitles edited (440) through a previewer.

[0084] Then, the generated video clips can be combined to create a user highlight video (450).

[0085] At this time, the video clip and highlight video can have at least one of the length and subtitles edited (460) through a previewer. That is, the user previews the generated integrated video and, if necessary, edits the length and subtitles of the entire video to complete the creation of the end-user highlight.

[0086] FIG. 8 is a flowchart illustrating a method for generating user-customized highlights according to an embodiment.

[0087] Referring to FIG. 8, a user-customized highlight generation method according to an embodiment may include the steps of: separating video data, background sound data, and commentary data from game data (S510); generating highlight candidate lists based on an artificial intelligence model that has been pre-trained for each of the separated video data, background sound data, and commentary data (S520); integrating the generated highlight candidate lists to generate a highlight list (S530); and generating a highlight with user preferences applied from the generated highlight list (S540).

[0088] At this time, the step of generating highlight candidate lists (S520) may include the step of extracting frames from video data at a predetermined period, the step of generating a list of major event occurrences based on a video learning model from the extracted frames, the step of generating a list of major event occurrence times based on analysis results of at least one predetermined video attribute change and character attribute change from the extracted frames, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0089] At this time, the video learning model may be pre-trained with video data including at least one of a character, indicator, skill, function, ability, and movement path.

[0090] At this time, the image attribute includes at least one of color, brightness, and saturation, and the character attribute may include at least one of movement, movement distance, and number.

[0091] At this time, the step of generating highlight candidate lists (S520) may include the step of dividing background sound data into intervals, the step of generating a list of major event occurrences based on a background sound learning model from the divided background sound data, the step of generating a list of major event occurrence times based on the analysis results of background sound frequency characteristics from the divided background sound data, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0092] At this time, the background sound learning model may be pre-trained with acoustic data including at least one of the game background sound and the battle sound effect.

[0093] At this time, the step of generating highlight candidate lists (S520) may include the step of dividing commentary data into intervals, the step of generating a list of major event occurrences based on a commentary learning model from the divided commentary data, the step of generating a list of major event occurrence times based on an analysis result for at least one predetermined voice attribute from the divided commentary data, and the step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals.

[0094] At this time, the specified voice attribute may include at least one of intensity, pitch, and speed.

[0095] At this time, the step of generating highlights (S540) may include the step of extracting a user highlight list with user preferences applied from a highlight list, the step of generating video clips for each of the extracted user highlight lists, and the step of integrating the generated video clips to generate a user highlight video.

[0096] At this time, at least one of the length and subtitles of the video clip and highlight video can be edited through the previewer.

[0097] FIG. 9 is a diagram showing the configuration of a computer system according to an embodiment.

[0098] A user-customized highlight generation device according to an embodiment can be implemented in a computer system (1000), such as a computer-readable recording medium.

[0099] A computer system (1000) may include one or more processors (1010), memory (1030), user interface input device (1040), user interface output device (1050), and storage (1060) that communicate with each other via a bus (1020). Additionally, the computer system (1000) may further include a network interface (1070) connected to a network (1080). The processor (1010) may be a semiconductor device that executes programs or processing instructions stored in a central processing unit, memory (1030), or storage (1060). The memory (1030) and storage (1060) may be storage media comprising at least one of a volatile medium, a non-volatile medium, a removable medium, a non-removable medium, a communication medium, or an information transfer medium. For example, the memory (1030) may include a ROM (1031) or a RAM (1032).

[0100] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without changing its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.

Claims

Claim 1 A user-customized highlight generation device comprising: a memory in which at least one program is recorded; and a processor for executing the program, wherein the program separates video data, background sound data, and commentary data from game data, generates a list of highlight candidates corresponding to each based on a pre-trained artificial intelligence model corresponding to each of the separated video data, background sound data, and commentary data, generates a highlight list by integrating the generated list of highlight candidates, and generates highlights selected based on user preferences from the generated list of highlight candidates, wherein each of the list of highlight candidates is generated by combining a list of major event occurrences generated by a learning model corresponding to each of the video data, background sound data, and commentary data, and a list of major event occurrence times based on the results of analyzing each of the video data, background sound data, and commentary data. Claim 2 A user-customized highlight generation device according to claim 1, wherein the program extracts frames from video data at a predetermined period, generates a list of major event occurrences based on an image learning model from the extracted frames, generates a list of major event occurrence times based on analysis results of at least one predetermined image attribute change and character attribute change from the extracted frames, and generates a highlight candidate list by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 3 In claim 2, the image learning model is a user-customized highlight generation device that is pre-trained with image data including at least one of a character, an indicator, a skill, a function, an ability, and a movement path. Claim 4 A user-customized highlight generation device according to claim 2, wherein the image attribute includes at least one of color, brightness, and saturation, and the character attribute includes at least one of movement, movement path distance, and number. Claim 5 A user-customized highlight generation device according to claim 1, wherein the program divides background sound data into intervals, generates a list of major event occurrences based on a background sound learning model from the divided background sound data, generates a list of major event occurrence times based on analysis results of background sound frequency characteristics from the divided background sound data, and generates a list of highlight candidates by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 6 In claim 5, the background sound learning model is a user-customized highlight generation device that is pre-trained with acoustic data including at least one of game background sound and combat sound effect. Claim 7 A user-customized highlight generation device according to claim 1, wherein the program divides commentary data into intervals, generates a list of major event occurrences based on a commentary learning model from the divided commentary data, generates a list of major event occurrence times based on an analysis result for at least one predetermined voice attribute from the divided commentary data, and generates a highlight candidate list by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 8 In claim 7, a user-customized highlight generation device wherein the specified voice attribute includes at least one of intensity, pitch, and speed. Claim 9 In claim 1, the program is a user-customized highlight generation device that extracts a user highlight list with user preferences applied from a highlight list, generates video clips for each of the extracted user highlight lists, and integrates the generated video clips to generate a user highlight video. Claim 10 In claim 9, the video clip and highlight video are a user-customized highlight generation device in which at least one of the length and subtitles is edited through a previewer. Claim 11 A method for generating user-customized highlights, comprising: a step of separating video data, background sound data, and commentary data from game data; a step of generating highlight candidate lists corresponding to each of the separated video data, background sound data, and commentary data based on a pre-trained artificial intelligence model corresponding to each of the separated video data, background sound data, and commentary data; a step of generating a highlight list by integrating the generated highlight candidate lists; and a step of generating highlights selected based on user preferences from the generated highlight list, wherein each of the highlight candidate lists is generated by combining a list of major event occurrences generated by a learning model corresponding to each of the video data, background sound data, and commentary data, and a list of major event occurrence times based on the results of analyzing each of the video data, background sound data, and commentary data. Claim 12 A method for generating user-customized highlights according to claim 11, wherein the step of generating highlight candidate lists comprises: a step of extracting frames at a predetermined period from video data; a step of generating a list of major event occurrences based on an image learning model from the extracted frames; a step of generating a list of major event occurrence times based on analysis results of at least one predetermined image attribute change and character attribute change from the extracted frames; and a step of generating highlight candidate lists by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 13 In claim 12, the video learning model is a user-customized highlight generation method in which the video learning model is pre-learned with video data including at least one of a character, an indicator, a skill, a function, an ability, and a movement path. Claim 14 A method for generating user-customized highlights according to claim 12, wherein the image attribute includes at least one of color, brightness, and saturation, and the character attribute includes at least one of movement, movement path distance, and number. Claim 15 A method for generating user-customized highlights according to claim 12, wherein the step of generating highlight candidate lists comprises: dividing background sound data into intervals; generating a list of major event occurrences based on a background sound learning model from the divided background sound data; generating a list of major event occurrence times based on analysis results of background sound frequency features from the divided background sound data; and generating a highlight candidate list by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 16 In claim 15, a user-customized highlight generation method wherein the background sound learning model is pre-trained with acoustic data including at least one of game background sound and combat sound effect. Claim 17 A method for generating user-customized highlights according to claim 11, wherein the step of generating a list of highlight candidates comprises: dividing commentary data into intervals; generating a list of major event occurrences based on a commentary learning model from the divided commentary data; generating a list of major event occurrence times based on an analysis result for at least one predetermined voice attribute from the divided commentary data; and generating a list of highlight candidates by combining the list of major event occurrences and the list of major event occurrence intervals. Claim 18 In claim 17, a user-customized highlight generation method wherein a predetermined voice attribute includes at least one of intensity, pitch, and speed. Claim 19 A method for generating user-customized highlights according to claim 11, wherein the step of generating highlights comprises: extracting a user highlight list to which user preferences are applied from a highlight list; generating video clips for each of the extracted user highlight lists; and integrating the generated video clips to generate a user highlight video. Claim 20 In claim 19, a method for generating user-customized highlights, wherein video clips and highlight videos are edited in length and subtitles through a previewer.

Images