Network type exercise round extraction and collection generation method and system and storage medium

Abstract

The invention discloses a network type motion round extraction and collection generation method and system and a storage medium. The method comprises the following steps: extracting an audio and calculating a short-time energy sequence, after three-stage filtering smoothing, carrying out adaptive threshold binaryzation, detecting a state conversion edge identification round, carrying out duration filtering and boundary expansion, and extracting fragments to synthesize a collection. The whole process is independently completed in the system, third-party editing software or a complex model tool is not needed, deterministic processing is carried out only based on the one-dimensional audio signal without depending on a network transmission protocol and extra hardware support, the problems of calculation redundancy, noise interference, abnormal data transmission and the like in a traditional scheme can be effectively avoided, and the method is suitable for large-scale popularization and application. The timeliness of match round extraction, the accuracy of recognition and the reliability of collection generation are ensured, and the automatic processing requirements of various net sports are met.

Description

Technical Field

[0001] This invention belongs to the field of video processing technology, specifically a method, system, and storage medium for extracting motion rounds and generating highlights of net-like motion. Background Technology

[0002] With the increasing popularity of live sports broadcasts, viewers' demand for highlights of tennis matches is growing rapidly. Traditional highlight production relies on manual editing, which is time-consuming, labor-intensive, and costly. Automated video highlight generation technology has become a hot research topic in the industry.

[0003] Existing automation solutions are mainly divided into two categories: one is based on computer vision methods, which tracks the trajectory of moving targets, identifies hitting actions, or understands scenes to locate exciting segments; the other uses audio features to assist in analysis, such as detecting specific audio events such as cheers from the audience and whistles from the referee.

[0004] However, computer vision solutions suffer from high computational complexity, difficulty in algorithm implementation, weak generalization ability, and stringent hardware requirements; audio-assisted solutions can only identify isolated highlights, cannot extract complete rounds of the game, and are easily affected by on-site noise, making it difficult to meet the needs of practical applications. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the present invention aims to provide a method, system, and storage medium for extracting rounds and generating highlights of net sports, thereby solving the technical problems of existing computer vision-based motion video editing methods, such as high computational complexity, difficulty in algorithm implementation, weak generalization ability, and demanding hardware resource requirements, as well as the difficulty of extracting complete match rounds and poor anti-interference ability of traditional audio-assisted methods. The invention aims to achieve efficient, reliable, and automated round extraction and highlight generation that is adaptable to various net sports.

[0006] To achieve the above objectives, this invention discloses a method for extracting and generating highlights of net-like sports rallies, the method comprising the following steps:

[0007] Step S1: Acquire the audio signal from the video of a tennis match;

[0008] Step S2: Perform frame segmentation on the audio signal and calculate the short-time energy of each frame to obtain a short-time energy sequence;

[0009] Step S3: Perform three-stage filtering on the short-time energy sequence to sequentially achieve impulse noise suppression, energy envelope smoothing, and energy trend extraction, thereby obtaining a smoothed energy sequence;

[0010] Step S4: Based on the statistical characteristics of the smooth energy sequence and the motion type adjustment factor, determine the adaptive threshold, and mark the low energy period as the game in progress and the high energy period as the round interval through binarization processing to generate a time-domain binarized sequence;

[0011] Step S5: Detect the state transition edges of the time-domain binarized sequence to locate candidate rounds, and perform optimization processing such as duration constraint filtering, adjacent round merging and boundary expansion on the candidate rounds to obtain an optimized set of round time periods;

[0012] Step S6: Extract corresponding video segments from the original video according to the optimized set of round time periods, and generate match highlights in chronological order.

[0013] Furthermore, the standard format of the audio signal is a mono PCM format with a sampling rate greater than a set threshold and a quantization precision of 16 bits; when the competition video contains multiple audio tracks, the audio track containing the narration is selected as the audio signal; when it contains only a single audio track, that audio track is directly used as the audio signal.

[0014] Furthermore, the frame segmentation process employs a short-time analysis window, which slides forward in steps smaller than the window length to form overlapping analysis frames. The short-time energy is calculated by squaring the amplitude of all sampling points within a single frame, and then summing all the squaring results to obtain the corresponding short-time energy quantization value.

[0015] Furthermore, the three-stage filtering process specifically includes:

[0016] The first-stage filter uses nonlinear filtering for impulse noise suppression;

[0017] The second-stage filter uses a low-pass filter for energy envelope smoothing;

[0018] The third-level filter uses a long-time-window average filter for trend extraction;

[0019] Furthermore, the statistical features include at least the median and the first quartile of the smoothed energy sequence;

[0020] The adaptive threshold is expressed by a threshold formula based on the lower quartile distance as follows:

[0021]

[0022] Where T is the adaptive threshold; The central location of the energy distribution; Indicates the first quartile; Indicates the lower quartile distance; As a regulating factor;

[0023] Further, the optimization process includes: detecting state transition edges of the temporal binary sequence; marking the corresponding edge as the start of a round when the temporal binary sequence changes from a round interval state to a match in progress state; marking the corresponding edge as the end of a round when the temporal binary sequence changes from a match in progress state to a round interval state; extracting the time interval between the start and end of a round as a candidate round; filtering out abnormal segments based on the duration constraint of the candidate rounds, and merging adjacent candidate rounds with a time interval less than a preset threshold; and performing forward and backward expansion on the merged round start and end times to obtain a round time interval containing the complete action process.

[0024] Furthermore, the highlight generation modes include all round highlights, selected round highlights, and highlights of a specified duration, and the generated highlight formats include at least one of MP4, AVI, MKV, and MOV.

[0025] A system for extracting and generating highlights of net-like motion sequences, characterized in that the system comprises:

[0026] The audio extraction module is used to acquire the audio signal containing the commentary voice in tennis competition videos and convert it into a standard format;

[0027] The energy calculation module is used to perform frame-by-frame processing on the audio signal, calculate the short-time energy of each frame, and generate a time-synchronized short-time energy sequence.

[0028] The filtering module is used to perform three-stage filtering on the short-time energy sequence and output a smooth energy sequence.

[0029] The threshold processing module is used to extract the statistical features of the smoothed energy sequence, calculate the adaptive threshold by combining the motion type adjustment factor, and generate a time-domain binarized sequence.

[0030] The round-based localization module is used to detect state transition edges in the time-domain binarized sequence, locate candidate rounds, perform duration filtering, adjacent merging, and boundary expansion optimization to obtain an optimized set of round time periods.

[0031] The highlights generation module is used to extract video segments based on the optimized set of round time periods, generate match highlights according to a preset mode, and encode and output them.

[0032] Furthermore, the audio extraction module includes a video demultiplexer, an audio stream splitter, a sample rate converter, and an audio format decoder; the filtering module integrates a median filter, a Butterworth low-pass filter, and a moving average filter; and the round-based localization module includes an edge detector, a round-based verifier, a round-based merger, and a boundary expander.

[0033] A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the net-like motion round extraction and highlight generation method according to any one of claims 1-7.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] (1) This invention only analyzes and processes one-dimensional audio signals, avoiding the high computational overhead of video image processing, greatly reducing the overall computational complexity, and enabling efficient round extraction and highlight generation.

[0036] (2) The present invention adopts deterministic signal processing and threshold discrimination logic, with a clear technical path, no need to rely on complex model training, easy to reproduce, debug and engineer, and the processing effect is stable and controllable.

[0037] (3) Based on the common audio pattern of “round silence - interval activity” in the commentary voice of net sports, this invention can be flexibly adapted to various net sports such as table tennis, tennis, badminton and volleyball by configuring relevant parameters of sports type. It does not require separate development for a single project and its generalization ability is significantly better than the existing technology.

[0038] (4) The core algorithm of this invention can run in real time or near real time on a general CPU platform without special hardware support, which reduces deployment costs and can be widely used in various scenarios such as personal devices, live streaming terminals, training and analysis equipment.

[0039] (5) This invention can accurately extract complete game rounds, which not only meets the core needs of post-match highlights production, but can also be extended to multiple fields such as live real-time editing assistance, athlete training video analysis, and match rhythm and round data statistics, making it more practical and expandable.

[0040] (6) This invention focuses on the periodic energy change pattern of the narration speech, effectively suppresses impulse noise and high frequency fluctuations through three-level filtering, and combines adaptive threshold to adapt to different motion scenarios, with more outstanding anti-interference ability and higher round recognition accuracy. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method steps of the present invention.

[0042] Figure 2 This is a schematic diagram illustrating the short-time energy sequence calculation of audio signals using the method of this invention.

[0043] Figure 3 This is a schematic diagram of the multi-stage filtering process for short-time energy sequences in the method of the present invention.

[0044] Figure 4This is a schematic diagram of the smooth energy sequence binarization process based on adaptive thresholding according to the method of the present invention.

[0045] Figure 5 This is a schematic diagram of round-based localization and boundary optimization based on time-domain binarized sequences in the method of the present invention.

[0046] Figure 6 This is a block diagram of the system structure for round extraction and compilation generation in the method of the present invention. Detailed Implementation

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 This application provides a detailed description of the technical solutions provided in each embodiment.

[0049] like Figure 1 As shown in the figure, the specific steps of the method for extracting and generating highlights of net-like motion rounds provided in this application embodiment include:

[0050] Step S1: Acquire the audio signal from the video of a tennis match;

[0051] This step is used to acquire the audio data to be processed. The input video is a match video of a tennis, badminton, volleyball, or other tennis sport, and its audio portion includes commentary. The video format includes at least one of MP4, AVI, MKV, and MOV.

[0052] When the competition video contains multiple audio tracks (or audio streams), the audio track (or audio stream) containing the narration is extracted; when the competition video contains only a single audio track (or audio stream), the single audio track (or audio stream) is used as the audio signal. In one embodiment, audio and video processing tools can be used to demultiplex and decode the video, and the extracted audio is uniformly converted to a mono PCM format with a sampling rate of 16kHz or higher and 16-bit quantization precision to ensure consistency in subsequent processing.

[0053] Step S2: Perform frame segmentation on the audio signal and calculate the short-time energy of each frame to obtain a short-time energy sequence;

[0054] This step obtains the short-time energy sequence of the audio signal through short-time energy analysis, such as... Figure 2 As shown, it specifically includes:

[0055] Step S21: Frame segmentation processing, dividing the continuous audio signal into short frames; the frame length is 20–50ms (corresponding to approximately 320–800 sampling points at a 16kHz sampling rate), and the frame shift is 40%–50% of the frame length to achieve inter-frame overlap;

[0056] Step S22: Energy calculation. Calculate the sum of squares for the sampling points within the i-th frame to obtain the short-time energy, expressed as:

[0057]

[0058] in, The short-time energy value of the i-th frame of audio is a quantitative indicator reflecting the signal strength of that frame; the amplitude of the n-th audio sampling point within the i-th frame is also represented. If it belongs to the i-th frame, it means that the value of n covers all sampling points in the i-th frame.

[0059] Step S23: Time synchronization for each short-time energy value Assign a timestamp t(i), and in one embodiment, take the center time of the frame to obtain a time-synchronized short-time energy sequence {E(i), t(i)}.

[0060] Step S3: Perform three-stage filtering on the short-time energy sequence to sequentially achieve impulse noise suppression, energy envelope smoothing, and energy trend extraction, thereby obtaining a smoothed energy sequence;

[0061] The short-time energy sequence contains impulse interference such as ball-hitting sounds, collision sounds, and audience cheers, and needs to be smoothed to obtain a smooth energy sequence (also known as a smooth energy envelope). Figure 3 As shown, the multi-stage filtering includes:

[0062] (1) First-level filtering: Nonlinear filtering is used for impulse noise suppression; in one embodiment, median filtering is used, with a window size of 5–10 points, corresponding to approximately 100–200ms, and the output is denoted as { }

[0063] (2) Second-stage filtering: Low-pass filtering is used for energy envelope smoothing; in one embodiment, a Butterworth low-pass filter is used with a cutoff frequency of 0.1–1Hz, and the output is denoted as { )}.

[0064] (3) Three-stage filtering: Long-term window averaging filter is used for trend extraction; in one embodiment, a moving average window filter is used, with a window duration of 1–5 seconds, and the output is denoted as { )}, that is, the smooth energy sequence.

[0065] Step S4: Based on the statistical characteristics of the smooth energy sequence and the motion type adjustment factor, determine the adaptive threshold, and mark the low energy period as the game in progress and the high energy period as the round interval through binarization processing to generate a time-domain binarized sequence;

[0066] like Figure 4 As shown, for the smooth energy sequence { Thresholding is performed to generate a time-domain binarized sequence, including:

[0067] Step S41: Statistical feature extraction, calculate the median and the first quartile;

[0068] Step S42: Adaptive threshold calculation, using a threshold formula based on the lower quartile distance, expressed as follows:

[0069]

[0070] Where T is the adaptive threshold; The central location of the energy distribution; Indicates the lower quartile; This represents the lower quartile distance, which characterizes the width of the low-energy region; As an adjustment factor, it controls the relative position of the threshold;

[0071] Recommended values ​​for adjustment factor k:

[0072] (1) When the sport is table tennis, the value of k is 0.8-1.0 because the rhythm is fast and the energy contrast is obvious.

[0073] (2) When the sport is tennis, the value of k is 1.0-1.2 because the length of the rally varies greatly and a more tolerant threshold is needed.

[0074] (3) When the sport is badminton, the value of k is 0.9-1.1, because its characteristics are between those of table tennis and tennis.

[0075] (4) When the sport is volleyball, the value of k is 0.7-0.9 because there is a lot of background noise in team sports.

[0076] Step S43: Binarization processing to generate a time-domain binarized sequence {B(i)} with timestamp synchronization. The specific determination rule is as follows:

[0077] When smooth energy value Less than the adaptive threshold When, the binary state at that moment is marked as B(i)=0, corresponding to the state of the game in progress; when When the value is greater than or equal to the threshold T, the state is marked as B(i)=1, corresponding to the round interval state. This mapping relationship is highly consistent with the commentator's commentary behavior pattern; during the game, the commentator usually remains relatively silent to avoid disturbing the audience, at which time the audio signal exhibits low energy characteristics, so it is represented by 0; during the round interval, the commentator will conduct tactical commentary, score announcement, or player status analysis, and the audio signal exhibits high energy characteristics, so it is represented by 1.

[0078] Step S5: Detect the state transition edges of the time-domain binarized sequence to locate candidate rounds, and perform optimization processing such as duration constraint filtering, adjacent round merging and boundary expansion on the candidate rounds to obtain an optimized set of round time periods;

[0079] like Figure 5 As shown, the transition edge location round boundary based on the time-domain binarized sequence {B(i)} includes:

[0080] Step S51: Round positioning, calculate the first difference for B(i):

[0081]

[0082] Where D(i)=-1 represents a falling edge (marking the start of a round), and D(i)=1 represents a rising edge (marking the end of a round);

[0083] Step S52: Candidate round extraction. The time interval between each falling edge and its most recent rising edge is taken as a candidate round, and the start time of each candidate round is recorded. and end time .

[0084] Step S53: Round validity verification. Based on the statistical characteristics of the sport, set round duration constraints. , Minimum duration This indicates the shortest duration of a valid round. Low-energy segments shorter than this value are usually false alarms (such as momentary pauses by the commentator, brief fluctuations in the audio signal, etc.) and do not constitute a true round. Maximum Duration This indicates the longest duration of a valid round. Low-energy segments exceeding this value typically indicate abnormal situations (such as match interruptions, equipment malfunctions, long pauses, etc.) rather than normal match rounds. Only rounds with durations within [[...] , Only silent segments within the specified interval are recognized as valid rounds.

[0085] For the following four ball sports, the threshold range for the round duration constraint is:

[0086] table tennis[ =3 seconds, =60 seconds], tennis =5 seconds, =180 seconds]

[0087] badminton[ =4 seconds, =120 seconds] and tennis [ =6 seconds, =90 seconds].

[0088] Step S54: Merge adjacent rounds. Check the interval between adjacent candidate rounds. If it is less than a threshold (e.g., 2 seconds), merge them into one round. This process takes into account the short pauses that the commentator may make due to breathing, thinking, etc., and avoids incorrectly splitting complete rounds.

[0089] Step S55: Round Boundary Expansion. There is a systematic deviation between the boundary moments obtained from audio energy analysis and the actual combat actions. Commentators typically begin commentary after the action occurs and stop after the score is determined. Directly using these boundaries will lead to:

[0090] (1) Front-end missing: loss of key actions such as serving and preparation;

[0091] (2) Backend truncation: Loss of scoring reactions, celebrations, and other visuals;

[0092] (3) Abrupt visual impact: lack of visual buffering affects smoothness.

[0093] To ensure the generated highlights are complete and visually appealing, fine-tuning of the recognized round boundaries is necessary. Therefore, forward expansion of the recognized round boundaries is required. and backward expansion The forward extension parameters are set according to different sports characteristics: 1-2 seconds for table tennis, 3-4 seconds for tennis, 1-1.5 seconds for badminton, and 3-4 seconds for volleyball, to cover the complete process from preparation to hitting the ball; the backward extension parameters are uniformly set to 0.5-1 seconds to preserve the crucial reaction time after scoring. This results in an optimized extended rally time. Based on this, the corresponding segments can be accurately extracted from the original video.

[0094] Step S6: Extract corresponding video segments from the original video according to the optimized set of round time periods, and generate match highlights in chronological order.

[0095] The extended round time period The compilation is arranged in chronological order, and transition effects can be selectively inserted, round numbers and score information can be added, and the intervals between segments can be adjusted to control the pace of the compilation. The final output is in mainstream formats such as MP4 and AVI. The encoding parameters can be optimized and configured according to the target platform (TV, network, mobile device) to generate exciting compilations adapted to different playback scenarios.

[0096] like Figure 6 As shown, this embodiment of the invention also provides a system implementation, which includes an audio extraction module, an energy calculation module, a filtering module, a threshold processing module, a round positioning module, and a highlight generation module; each module is used to implement the corresponding functions of steps S1 to S6, and the modules are connected through an interface to support parameter configuration and expansion. The audio extraction module includes a video demultiplexer, an audio stream separator, a sample rate converter, and an audio format decoder, responsible for separating the audio stream from the input video and converting it to the standard PCM format with a 16kHz sample rate. The energy calculation module extracts the short-time energy sequence of the audio signal through a frame processing unit, a short-time energy calculator, an energy sequence generator, and a timestamp synchronization unit. The filtering module integrates a three-stage filter to suppress noise and smooth the energy sequence. The thresholding module includes a statistical feature extractor, an adaptive threshold calculator, and a binarization processor to accurately determine the commentary status. The round positioning module identifies and optimizes the round boundaries of the match through an edge detector, an edge matcher, a round verifier, a round merger, and a boundary expander. The highlight generation module integrates a video clip extractor, a clip sorter, and a video encoder outputter to complete all the functions of accurately extracting round clips from the original video and generating high-quality highlights.

[0097] Beyond basic highlight generation, this invention can be extended to multiple fields: Real-time processing mode can be combined with streaming media technology to achieve real-time round marking and highlight clip push during live broadcasts; data statistical analysis function can automatically calculate key data such as match rhythm, round duration distribution, and match intensity based on identified round information; personalized highlight function supports generating customized content according to user preferences; multi-camera synchronization function can use audio analysis results as a time reference to achieve automatic synchronized editing of multi-camera videos. In summary, this invention, by innovatively utilizing audio energy characteristics, achieves efficient automatic identification of netball match rounds, providing a new technical path for intelligent processing of sports videos.

[0098] In addition, this application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the above-described method for extracting and generating highlights of net-like motion rounds.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for extracting and generating highlights of net-like motion sequences, characterized in that, The method steps include: Step S1: Acquire the audio signal from the video of a tennis match; Step S2: Perform frame segmentation on the audio signal and calculate the short-time energy of each frame to obtain a short-time energy sequence; Step S3: Perform three-stage filtering on the short-time energy sequence to sequentially achieve impulse noise suppression, energy envelope smoothing, and energy trend extraction, thereby obtaining a smoothed energy sequence; Step S4: Based on the statistical characteristics of the smooth energy sequence and the motion type adjustment factor, determine the adaptive threshold, and mark the low energy period as the game in progress and the high energy period as the round interval through binarization processing to generate a time-domain binarized sequence; Step S5: Detect the state transition edges of the time-domain binarized sequence to locate candidate rounds, and perform optimization processing such as duration constraint filtering, adjacent round merging and boundary expansion on the candidate rounds to obtain an optimized set of round time periods; Step S6: Extract corresponding video segments from the original video according to the optimized set of round time periods, and generate match highlights in chronological order.

2. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The standard format of the audio signal is mono, PCM format with a sampling rate greater than a set threshold and quantization precision of 16 bits; when the competition video contains multiple audio tracks, the audio track containing the narration is selected as the audio signal; when it contains only a single audio track, that audio track is directly used as the audio signal.

3. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The frame segmentation process uses a short-time analysis window and slides forward in steps smaller than the window length to form overlapping analysis frames. The short-time energy is calculated by squaring the amplitude of all sampling points in a single frame and then summing all the squaring results to obtain the corresponding short-time energy quantization value.

4. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The three-stage filtering process specifically includes: The first-stage filter uses nonlinear filtering for impulse noise suppression; The second-stage filter uses a low-pass filter for energy envelope smoothing; The third-level filter uses a long-time-window average filter for trend extraction.

5. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The statistical features include at least the median and the first quartile of the smoothed energy sequence; The adaptive threshold is expressed by a threshold formula based on the lower quartile distance as follows: Where T is the adaptive threshold; The central location of the energy distribution; Indicates the first quartile; Indicates the lower quartile distance; It is a regulating factor.

6. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The optimization process includes: detecting state transition edges in the temporal binary sequence; marking the corresponding edge as the start of a round when the temporal binary sequence changes from a round interval state to a match in progress state; marking the corresponding edge as the end of a round when the temporal binary sequence changes from a match in progress state to a round interval state; extracting the time interval between the start and end of a round as candidate rounds; filtering out abnormal segments based on the duration constraints of the candidate rounds, and merging adjacent candidate rounds with a time interval less than a preset threshold; and performing forward and backward expansion on the merged round start and end times to obtain a round time interval containing the complete action process.

7. The method for extracting and generating highlights of net-like motion rounds according to claim 1, characterized in that, The highlight generation modes include all round highlights, selected round highlights, and highlights of a specified duration. The generated highlight formats include at least one of MP4, AVI, MKV, and MOV.

8. A system for extracting and generating highlights of net-like motion sequences, characterized in that, The system includes: The audio extraction module is used to acquire the audio signal containing the commentary voice in tennis competition videos and convert it into a standard format; The energy calculation module is used to perform frame-by-frame processing on the audio signal, calculate the short-time energy of each frame, and generate a time-synchronized short-time energy sequence. The filtering module is used to perform three-stage filtering on the short-time energy sequence and output a smooth energy sequence. The threshold processing module is used to extract the statistical features of the smoothed energy sequence, calculate the adaptive threshold by combining the motion type adjustment factor, and generate a time-domain binarized sequence. The round-based localization module is used to detect state transition edges in the time-domain binarized sequence, locate candidate rounds, perform duration filtering, adjacent merging, and boundary expansion optimization to obtain an optimized set of round time periods. The highlights generation module is used to extract video segments based on the optimized set of round time periods, generate match highlights according to a preset mode, and encode and output them.

9. A system for extracting and generating highlights of net-like motion rounds according to claim 8, characterized in that, The audio extraction module includes a video demultiplexer, an audio stream splitter, a sample rate converter, and an audio format decoder; The filtering module integrates a median filter, a Butterworth low-pass filter, and a moving average filter; the round-based localization module includes an edge detector, a round-based verifier, a round-based merger, and a boundary expander.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the net-like motion round extraction and highlight generation method according to any one of claims 1-7.

Images