A multi-scene universal scoring system and scoring method

Abstract

The application discloses a multi-scene general scoring system and method, which comprises an activity information collecting module for collecting activity on-site program basic data and program theme data in real time, an activity information processing module for performing preliminary data conversion on the program basic data and program theme data collected by the activity information collecting module and identifying the program theme data, a music program standard scoring module for scoring singing or instrument performance programs, a dance action standard scoring module for scoring dance performance programs, a judge information management unit for collecting judge information to generate a judge scoring communication path, a contestant information management unit for collecting contestant entry information, and a data management unit for managing the writing, modification and deletion processes of the entire activity program basic data, program theme data, judge information and contestant entry information.

Description

Technical Field

[0001] This invention belongs to the field of event and competition scoring technology, specifically, it relates to a multi-scenario universal scoring system and scoring method. Background Technology

[0002] Enterprises, institutions, schools, and social organizations frequently hold various competitions and evaluations to enrich people's leisure time and showcase their talents, skills, and professional abilities. These activities all involve judges scoring the participants on-site. Currently, most of these on-site scoring methods are manual.

[0003] Chinese patent application CN201911116566.X discloses a competition scoring system based on a WeChat mini-program, comprising an application layer, a data resource management layer, and a physical layer. The application layer provides users with an interface to interact with the competition scoring system via the WeChat client, responsible for collecting user data, displaying information to users, and rendering pages. The application layer is connected to the data resource management layer; data generated by user operations in the application layer is directly written to the database server of the data resource management layer, i.e., the WeChat server's mini-program cloud database. The data displayed by the application layer directly originates from the database server of the data resource management layer. The data resource management layer provides a unified platform for data interaction between application layer users and the system; the functions of the data resource management layer are undertaken by the database server, which stores all data generated by user operations in the application layer. The physical layer consists of mobile terminal devices, computers, and networks, providing the most basic guarantee for the system's operation.

[0004] The shortcomings of existing patents are that, although the aforementioned patents have achieved a universal on-site scoring system that is not limited to a specific competition and can select different score calculation methods to adapt to different scenarios, the professional evaluation rules and scores of the performances or competitions by judges and audiences during the event are greatly influenced by the atmosphere on site. Furthermore, due to the special nature of the programs, judges may not be able to make professional scores for all performances that involve their own professional fields. Summary of the Invention

[0005] To address the problems that existing event competition systems cannot adapt to diverse program structures and that attending judges, with their limited professional expertise, cannot provide professional scores for all performances, this invention provides a multi-scenario universal scoring system and scoring method.

[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0007] A multi-scenario universal scoring system includes an activity information management unit, a scoring standard management unit, a judge information management unit, a participant information management unit, an activity process control unit, and a data management unit;

[0008] The event information management unit includes an event information acquisition module and an event information processing module. The event information acquisition module is used to collect basic program data and program theme data from the event site in real time. The event information processing module is used to perform preliminary data conversion and program theme data identification on the basic program data and program theme data collected by the event information acquisition module. The preliminary data conversion facilitates data identification and use in the entire scoring system, and also unifies the data values ​​collected from different ports.

[0009] The scoring criteria management unit includes a music program scoring module and a dance movement scoring module. The music program scoring module is used to score singing or instrumental performances, while the dance movement scoring module is used to score dance performances.

[0010] The judge information management unit is used to collect judge information, generate judge scoring communication paths (QR code scanning scoring), and judges' scores for the competition after each performance.

[0011] The contestant information management unit is used to collect contestant entry information and determine in real time during the competition whether the corresponding contestant is on stage for the entire program. It also matches the judges' scores after each performance to score the contestants; this prevents substitute performances and facilitates the system's statistics on performance scores and the scores of the corresponding performers.

[0012] The event flow control unit is used to control the flow of the entire event performance and to regulate the sound and lighting control systems of the on-site equipment.

[0013] The data management unit is used to manage the writing, modification, and deletion processes of basic event data, event theme data, judge information, and contestant entry information.

[0014] Furthermore, the basic program data collected by the event information management unit includes individual audience data upon entry, individual judge data, and real-time uploaded audio and video data during the live performances; the program theme data includes the list of all programs in the entire competition, program type data, and program performance duration.

[0015] Furthermore, the music program standard scoring module uses a comprehensive scoring method based on lyrics and pitch to score music programs. This comprehensive scoring method employs a factor-based evaluation formula:

[0016]

[0017] Wherein, k1 and k2 are the weights of lyrics and pitch in a singer's performance of a specific song, with a default setting of k1 + k2 = 1; d1 and d2 are the difficulty coefficients of the lyrics and pitch of the song, respectively, and we stipulate that d1 and d2 are both greater than 1; D(mfcc) and D(pitch) correspond to the differences in lyrics and pitch between the singer's performance and the standard original song, respectively; Vi_A and Vi_B represent the sound intensity of the standard sound template and the sound template to be matched.

[0018] Furthermore, the pitch of the program audio corresponds to the high and low changes of the fundamental frequency, and the feature extraction method for the fundamental frequency adopts the autocorrelation fundamental frequency detection algorithm:

[0019] Speech signal pre-emphasis;

[0020] Speech signal framing and windowing;

[0021] Substitute into the autocorrelation function;

[0022] Set the peak threshold;

[0023] The fundamental frequency is determined by the location of the peak.

[0024] The basic principle of the autocorrelation function method is to multiply the speech signal by an autocorrelation function based on the speech signal to obtain a new speech signal function. This function will produce a maximum value at a point where the time delay equals the pitch period. Based on this property, we estimate the pitch period of the speech signal by calculating the maximum point of the autocorrelation function of the speech signal.

[0025] Furthermore, the lyrics extraction method for the program audio employs the Mel-frequency cepstral coefficients (MFCC) extraction algorithm:

[0026] Preprocess the audio signal;

[0027] The time-domain waveform of the preprocessed audio signal is obtained by Fast Fourier Transform (FFT);

[0028] The sound signal is passed through a MEL filter bank to obtain the Mel spectrum coefficients;

[0029] Finally, the Mel cepstral coefficients are obtained through discrete cosine transform (DCT).

[0030] Furthermore, the dance movement standard scoring module uses a quality score decoupled movement quality assessment algorithm to score the dance movement program. The quality score decoupled movement quality assessment algorithm includes a feature extraction network combining dynamic and static flows, a quality score decoupling module, and a pairwise ranking prediction module.

[0031] The detailed process for scoring dance performances includes:

[0032] For a pair of input samples (v) i ,v j First, each video (V) is divided into T segments of equal length, and then fed into a Siamese neural network. Taking video V as an example... i For example, T video clips are fed into a dynamic stream network using I3D and a static stream network using ResNet to extract features F. Di and F Si The former refers to the extracted video V. i The dynamic flow features, the latter representing the extracted video V i The static flow characteristics.

[0033] Then F Di and F Si Input two quality score decoupling modules that do not share weights to obtain the dynamic flow feature F. Di High-quality score features F DPi Low quality score characteristic F DCi and the characteristic F of average aggregation DUi For static flow F Si Similarly, three features F can be obtained. SPi F SCi and F SUi The corresponding features are concatenated into F. Pi F Ci and F Ui Indicates video V i The final features are then input into the quality score predictor to obtain the final score.

[0034] For video V j The same method is used to obtain the final quality scores for the three feature inputs.

[0035] The pairwise ranking prediction module consists of a quality score predictor, which concatenates the three branch features of the obtained dynamic flow and static flow, regresses to obtain the predicted scores, and adds them together as the total score.

[0036] Furthermore, the detailed process for collecting judge information and generating judge scores through communication includes the following steps:

[0037] Collect judges' personal data, which includes their unique personal identification code at the event, their name, and their phone number.

[0038] The unique personal identification codes, judges' names, and judges' phone numbers at the event will be uniformly converted into strings.

[0039] Convert a string to binary code;

[0040] Fill all the binary codes into the number cells composed of 64 rectangles; (fill the number cells with white when the binary code is 0, and fill them with black when the binary code is 1).

[0041] Generate a QR code and return it to the individual judge via phone number or communication software connected to the phone number;

[0042] Judges scan a generated QR code after each program ends and before the next program begins to upload their scores for each program.

[0043] Furthermore, the scoring criteria management unit uses the following scoring formula for each performance:

[0044] F 总 =F 观众 ·W1+F 评委 • W2+ (ratings for music programs or dance moves) • W3;

[0045] F 观众 The average score given by the audience, W1 is the audience weighting coefficient, F 评委 W1 represents the average score given by the judges, W2 represents the judges' weighting coefficient, and W3 represents the intelligent system's weighting coefficient.

[0046] Furthermore, the scoring process for each performance by the scoring criteria management unit includes the following steps:

[0047] S001. The system scores programs based on their list and type, using the corresponding scoring method.

[0048] S002. The total score given by the audience is divided by the number of audience members who gave the score to obtain the average score given by the audience; the audience score generation communication path and detailed scoring process are the same as those of the judges.

[0049] S003. Collect the program score given by each judge, remove the highest and lowest scores, divide by (total number of judges - 2) to get the average score given by the judges.

[0050] S004. Substitute all the scores from steps S001, S002, and S003 into the scoring formula for each performance to calculate the total score (F) of the performance. 总 W1, W2, and W3 can be preset to 1 / 3, or their weight values ​​can be set by the official administrator of the competition.

[0051] A general scoring method for multiple scenarios includes the following steps:

[0052] S1. Collect personal data of the audience, judges and performers at the entrance of the event venue;

[0053] S2. Real-time acquisition of audio and video data of the program on the performance stage at the event site using audio and video acquisition equipment;

[0054] S3. Determine the current program type based on the program theme data collected before the performance begins; the current program type is matched and searched in the database according to the input or collected program type data; if it is a music program or a crosstalk program, proceed to step S4; if it is a dance program or an acrobatic program, proceed to step S5; if it is a music and dance program, proceed to step S6.

[0055] S4. Use a combined scoring method of lyrics and pitch to score music programs;

[0056] S5. A quality score-decoupled motion quality assessment algorithm is used to score dance moves.

[0057] S6. Classify the program's audio and video data and use the scoring methods in S4 and S5 to score the program's music and dance moves respectively.

[0058] S7. Multiply the program score calculated in steps S4 and S5 by the intelligent system weight coefficient, add the average score given by the audience by the audience weight coefficient, and add the average score given by the judges by the judges weight coefficient to obtain the total score of a single type of program.

[0059] S8. Sum the music score and dance score calculated in step S6, divide by 2, multiply the resulting program score by the intelligent system weight coefficient, add the average score given by the audience multiplied by the audience weight coefficient, and add the average score given by the judges multiplied by the judges weight coefficient to obtain the total score of the multiple program types.

[0060] Compared with the prior art, the present invention has the following advantages:

[0061] The program composition of the event competition is classified by collecting basic data and program theme data from the event site, and the diversified program composition structure is decomposed. Then, the scoring standard management unit can realize the scoring of single-type programs, as well as the scoring of multiple types of programs.

[0062] By collecting judge / audience information, a corresponding scoring communication path is generated between judges and audience members, ensuring the privacy of each person's scoring at the event and preventing them from being influenced by others' scores.

[0063] Each program and performer is comprehensively evaluated and scored through a combination of scores from the audience, judges, and an intelligent system. This prevents situations where judges and audience members are unable to provide objective and professional scores for individual art programs. Attached Figure Description

[0064] Figure 1This is a block diagram of the overall structure of a multi-scenario general scoring system according to an embodiment of the present invention;

[0065] Figure 2 This is a detailed flowchart of the feature extraction method for fundamental frequency in an embodiment of the present invention;

[0066] Figure 3 This is a detailed flowchart of the method for extracting lyrics from program audio in an embodiment of the present invention;

[0067] Figure 4 This is a detailed flowchart of the communication path for collecting judge information and generating judge scores in an embodiment of the present invention;

[0068] Figure 5 This is a flowchart illustrating the scoring process of each performance by the scoring standard management unit in this embodiment of the invention.

[0069] Figure 6 This is an overall flowchart of a multi-scenario general scoring method in an embodiment of the present invention. Detailed Implementation

[0070] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0071] like Figure 1 As shown, this embodiment provides a multi-scenario general scoring system, including an activity information management unit, a scoring standard management unit, a judge information management unit, a contestant information management unit, an activity process control unit, and a data management unit;

[0072] The event information management unit includes an event information acquisition module and an event information processing module. The event information acquisition module is used to collect basic program data and program theme data from the event site in real time. The event information processing module is used to perform preliminary data conversion and program theme data identification on the basic program data and program theme data collected by the event information acquisition module. The preliminary data conversion facilitates data identification and use in the entire scoring system, and also unifies the data values ​​collected from different ports.

[0073] The scoring criteria management unit includes a music program scoring module and a dance movement scoring module. The music program scoring module is used to score singing or instrumental performances, while the dance movement scoring module is used to score dance performances.

[0074] The judge information management unit is used to collect judge information, generate judge scoring communication paths (QR code scanning scoring), and judges' scores for the competition after each performance.

[0075] The contestant information management unit is used to collect contestant entry information and determine in real time during the competition whether the corresponding contestant is on stage for the entire program. It also matches the judges' scores after each performance to score the contestants; this prevents substitute performances and facilitates the system's statistics on performance scores and the scores of the corresponding performers.

[0076] The event flow control unit is used to control the flow of the entire event performance and to regulate the sound and lighting control systems of the on-site equipment.

[0077] The data management unit is used to manage the writing, modification, and deletion processes of basic event data, event theme data, judge information, and contestant entry information.

[0078] The basic program data collected by the event information management unit includes individual audience data upon entry, individual judge data, and real-time uploaded audio and video data during the live performances; the program theme data includes the list of all programs in the entire competition, program type data, and program performance duration.

[0079] The standard scoring module for music programs uses a combined scoring method based on lyrics and pitch to evaluate the programs. This combined scoring method employs a factor-based evaluation formula:

[0080]

[0081] Wherein, k1 and k2 are the weights of lyrics and pitch in a singer's performance of a specific song, with a default setting of k1 + k2 = 1; d1 and d2 are the difficulty coefficients of the lyrics and pitch of the song, respectively, and we stipulate that d1 and d2 are both greater than 1; D(mfcc) and D(pitch) correspond to the differences in lyrics and pitch between the singer's performance and the standard original song, respectively; Vi_A and Vi_B represent the sound intensity of the standard sound template and the sound template to be matched.

[0082] In this embodiment, k1 = k2 = 0.5 and d1 = d2 = 1 are preset. Therefore, the above formula can be simplified to:

[0083]

[0084] From the above formula, we can derive:

[0085] When D(mfcc) = D(pitch) = 0, meaning the singer's lyrics and pitch are exactly the same as the original, the highest score is:

[0086]

[0087] When D(mfcc) = D(pitch) = ∞, meaning the singer's lyrics and pitch differ greatly from the original, the score will be lower than the original, even approaching 0.

[0088] like Figure 2 As shown, the pitch of the program audio corresponds to the high and low changes of the fundamental frequency. The feature extraction method for the fundamental frequency adopts the autocorrelation fundamental frequency detection algorithm:

[0089] Speech signal pre-emphasis;

[0090] Speech signal framing and windowing;

[0091] Substitute into the autocorrelation function;

[0092] Set the peak threshold;

[0093] The fundamental frequency is determined by the location of the peak.

[0094] Furthermore, speech signals are divided into two types: discrete digital speech signals and random signal sequences or periodic signals;

[0095] The autocorrelation function of a discrete digital speech signal sequence is defined as follows:

[0096]

[0097] Where x(m) is the digital speech signal function, and k is the number of signal delay points;

[0098] The autocorrelation function of a random signal sequence or a periodic signal sequence is defined as:

[0099]

[0100] N represents the periodic data value.

[0101] The basic principle of the autocorrelation function method is to multiply the speech signal by an autocorrelation function based on the speech signal to obtain a new speech signal function. This function will produce a maximum value at a point where the time delay equals the pitch period. Based on this property, we estimate the pitch period of the speech signal by calculating the maximum point of the autocorrelation function of the speech signal.

[0102] like Figure 3 As shown, the lyrics extraction method for the program audio uses the Mel-frequency cepstral coefficients (MFCC) extraction algorithm:

[0103] Preprocess the audio signal;

[0104] The time-domain waveform of the preprocessed audio signal is obtained by Fast Fourier Transform (FFT);

[0105] Function code for extracting MFCC parameters in MATLAB:

[0106] fumetion cee=lEcc(x,Fs,n,nc,nfft,fl,fh,ovlp)

[0107] ifnargin<2 Fs=8000;end

[0108] if nargin < 3 n = 12; end

[0109] ifnargin<4 nC=24;end

[0110] ifnargin<5 nfft=256; end

[0111] ifnargin<6 f1=0; end

[0112] ifnargin<7 fh=0.5; end

[0113] ifnargin<8 ovlp=0.5; end.

[0114] Where x: is the matrix data of the input audio file;

[0115] Fs: Sampling frequency, default value is 8000Hz;

[0116] nM: The dimension of MFCC, the default value is 12;

[0117] nC: The number of filters, the default value is 24;

[0118] nfft: The number of sampling points per frame, with a default value of 256;

[0119] fl: This is the percentage of the sampling frequency occupied by the lowest sampling frequency in each frame shift, i.e., the lowest sampling frequency is fl*Fs, and the default value is 0;

[0120] fh: The percentage of the highest sampling frequency in each frame relative to the frame shift frequency, i.e., the highest sampling frequency is fh*Fs, with a default value of 0.5;

[0121] ovlp: The percentage of the frame size occupied by the frame shift in each frame, with a default value of 0.5.

[0122] The sound signal is passed through a MEL filter bank to obtain the Mel spectrum coefficients;

[0123] The spectrum obtained by Fast Fourier Transform (FFT) is filtered by power spectrum of the signal using a triangular filter bank

[21] . The range covered by each triangular window filter is approximately the critical bandwidth of the human ear, thereby simulating the masking effect of the human ear. A set of coefficients m1, m2, ... can be obtained. The number of filter banks S is determined by the frequency range of the signal.

[0124] Where, calculate m i The formula is shown below:

[0125]

[0126] Where m i Let X(k) represent the i-th dimension of the MEL spectrum coefficients, N be the number of sampling points, X(k) be the sound signal function, S be the number of filter banks, and H be the number of filter banks. i (k) is a function of the i-th filter, H i (k):

[0127]

[0128] f[i] is the center frequency of the triangular filter bank, which satisfies:

[0129] Mel(f[i+1])-Mel(f[i])=Mel(f[i])-Mel(f[i-1])

[0130] Finally, the Mel cepstral coefficients are obtained through discrete cosine transform (DCT).

[0131] The formula for calculating the discrete cosine transform is as follows:

[0132]

[0133] Where, m j Given a time-domain sequence of length 1-P, where i ranges from 1 to P; C(i) is the frequency-domain coefficient sequence of length 1-P.

[0134] Furthermore, the formula for defining the sound intensity curve is:

[0135]

[0136] Among them, V i S represents the average sound intensity of the entire audio file, |S(m)| represents the signal amplitude at the m-th sampling point, and M is the number of sampling points.

[0137] Furthermore, the dance movement standard scoring module uses a quality score decoupled movement quality assessment algorithm to score the dance movement program. The quality score decoupled movement quality assessment algorithm includes a feature extraction network combining dynamic and static flows, a quality score decoupling module, and a pairwise ranking prediction module.

[0138] For a pair of input samples (vi, vj) ∈ V, each video is first divided into T segments of equal length, and then fed into a Siamese neural network. Taking video V as an example... i For example, T video clips are fed into a dynamic stream network using I3D and a static stream network using ResNet to extract features F. Di and F Si The former refers to the extracted video V. i The dynamic flow features, the latter representing the extracted video V i The static flow characteristics.

[0139] Then F Di and F Si Input two quality score decoupling modules that do not share weights to obtain the dynamic flow feature F. Di High-quality score feature F DPi Low quality score characteristic F DCi and the characteristic F of average aggregation DUi For static flow F Si Similarly, three features F can be obtained. SPi F SCi and F SUi Concatenate the corresponding features into F Pi F Ci and F Ui Indicates video V i The final features are then input into the quality score predictor to obtain the final score.

[0140] For video V j The same method is used to obtain the final quality scores for the three feature inputs.

[0141] The pairwise ranking prediction module consists of a quality score predictor. It concatenates the three branches of features obtained from the dynamic flow and static flow, regresses them to obtain predicted scores, and adds them together to obtain the total score.

[0142] s = S r ([F D ;F S ]);

[0143] The quality score predictor Sr(·) consists of three fully connected layers (FC) and a sigmoid activation function;

[0144] The Sigmoid activation function is used to normalize the estimated scores to the range [0, 1]. After obtaining the predicted scores of two input samples, the output is compared and ranked in pairs.

[0145] Four different loss functions were defined:

[0146] Pairwise sorting loss maximizes the difference between sorted video pairs;

[0147] The pairwise ranking loss is used to constrain the predicted scores of all three branches of features for video vi to be greater than the predicted score for video vj; the three components are as follows:

[0148]

[0149] The final loss function is the sum of these three parts:

[0150]

[0151] m1 is a hyperparameter representing the difference in scores between two videos. If the difference is greater than m1, then v i and v j The pairwise ranking loss for the corresponding features is 0.

[0152] The difference between feature pairs obtained by difference loss decoupling is greater than the difference between average feature pairs.

[0153] Differential loss formula:

[0154]

[0155] The final loss function is the sum of these two parts:

[0156]

[0157] m2 is the difference between the two hyperparameter constraints in this loss; this loss encourages the fractional difference Sr(F) between the decoupled high-quality fractional features. Pi )-Sr(F Pj The score difference Sr(F) between the low-quality score feature and the low-quality score feature Ci )-Sr(F Cj All are greater than the difference between the average polymer mass fractions, Sr(F) Ui )-Sr(F Uj This encourages the quality score decoupling module to obtain video feature representations that better distinguish the action quality of two videos than average aggregation.

[0158] The ranking-aware loss optimizes the decoupling module to decouple high and low score features;

[0159] The formula for order-aware loss is:

[0160]

[0161] By setting a hyperparameter m3 that is larger than m2, the video v is constrained. i High-quality score features predict scores and video v j The difference in predicted scores for low-quality features is much greater than the difference in quality scores for average aggregate features. Thus, the two features obtained by constraint decoupling represent high-quality features and low-quality features, respectively.

[0162] The diversity loss constraint focuses on different segments with high / low quality score features.

[0163] A diversity loss is added to directly constrain the attention matrix of the quality score decoupling module:

[0164]

[0165] Here, K×T represents the attention matrix, where K represents the number of decoupled feature branches (set to 2 in the model), represent high / low quality score feature branches respectively, T represents the number of video segments, and ak(vi1) represents the video segment v in the Kth branch. i1 The corresponding attention.

[0166] Formula for diversity loss in dynamic flow:

[0167]

[0168] Formula for diversity loss in static flow:

[0169]

[0170] Where I represents the identity matrix, ||·|| 2 F Represents the Frobenius norm;

[0171] The formula for overall diversity loss is:

[0172]

[0173] During testing, the three scores are added together to obtain the final predicted score, and the pairwise sorting accuracy is calculated.

[0174] The total loss function consists of four parts: pairwise ranking loss, difference loss, ranking-aware loss, and diversity loss.

[0175] L all =L rank +L dist +L aware +λL div

[0176] The formula for calculating the total score is:

[0177] S(v i ) = S r (F Pi )+S r (F Ci )+S r (F Ui )

[0178] like Figure 4 As shown, the detailed process of collecting judge information and generating judge scores through communication includes the following steps:

[0179] Collect judges' personal data, which includes their unique personal identification code at the event, their name, and their phone number.

[0180] The unique personal identification codes, judges' names, and judges' phone numbers at the event will be uniformly converted into strings.

[0181] Convert a string to binary code;

[0182] Fill all the binary codes into the number cells composed of 64 rectangles; (fill the number cells with white when the binary code is 0, and fill them with black when the binary code is 1).

[0183] Generate a QR code and return it to the individual judge via phone number or communication software connected to the phone number;

[0184] Judges scan a generated QR code after each program ends and before the next program begins to upload their scores for each program.

[0185] Furthermore, the scoring criteria management unit uses the following scoring formula for each performance:

[0186] F 总 =F 观众 ·W1+F 评委 • W2+ (ratings for music programs or dance moves) • W3;

[0187] F 观众 The average score given by the audience, W1 is the audience weighting coefficient, F 评委 W1 represents the average score given by the judges, W2 represents the judges' weighting coefficient, and W3 represents the intelligent system's weighting coefficient.

[0188] like Figure 5 As shown, the scoring process for each performance by the scoring criteria management unit includes the following steps:

[0189] S001. The system scores programs based on their list and type, using the corresponding scoring method.

[0190] S002. The total score given by the audience is divided by the number of audience members who gave the score to obtain the average score given by the audience; the audience score generation communication path and detailed scoring process are the same as those of the judges.

[0191] S003. Collect the program score given by each judge, remove the highest and lowest scores, divide by (total number of judges - 2) to get the average score given by the judges.

[0192] S004. Substitute all the scores from steps S001, S002, and S003 into the scoring formula for each performance to calculate the total score (F) of the performance. 总 W1, W2, and W3 can be preset to 1 / 3, or their weight values ​​can be set by the official administrator of the competition.

[0193] like Figure 6 As shown, a general scoring method for multiple scenarios includes the following steps:

[0194] S1. Collect personal data of the audience, judges and performers at the entrance of the event venue;

[0195] S2. Real-time acquisition of audio and video data of the program on the performance stage at the event site using audio and video acquisition equipment;

[0196] S3. Based on the program theme data collected before the performance, determine the current program type; the current program type is matched and searched in the database according to the input or collected program type data; if it is a music program or a crosstalk program, proceed to step S4; if it is a dance program or an acrobatic program, proceed to step S5; if it is a music and dance program, proceed to step S6; for music programs or crosstalk programs, the collected audio data is compared and scored with the original audio data obtained by the system from a large amount of web crawling; for dance programs or acrobatic programs, the collected video data is compared and scored with the original video data obtained by the system from a large amount of web crawling.

[0197] S4. Use a combined scoring method of lyrics and pitch to score music programs;

[0198] S5. A quality score-decoupled motion quality assessment algorithm is used to score dance moves.

[0199] S6. Classify the program's audio and video data and use the scoring methods in S4 and S5 to score the program's music and dance moves respectively.

[0200] S7. Multiply the program score calculated in steps S4 and S5 by the intelligent system weight coefficient, add the average score given by the audience by the audience weight coefficient, and add the average score given by the judges by the judges weight coefficient to obtain the total score of a single type of program.

[0201] S8. Sum the music score and dance score calculated in step S6, divide by 2, multiply the resulting program score by the intelligent system weight coefficient, add the average score given by the audience multiplied by the audience weight coefficient, and add the average score given by the judges multiplied by the judges weight coefficient to obtain the total score of the multiple program types.

[0202] Compared with the prior art, the present invention has the following advantages:

[0203] The program composition of the event competition is classified by collecting basic data and program theme data from the event site, and the diversified program composition structure is decomposed. Then, the scoring standard management unit can realize the scoring of single-type programs, as well as the scoring of multiple types of programs.

[0204] By collecting judge / audience information, a corresponding scoring communication path is generated between judges and audience members, ensuring the privacy of each person's scoring at the event and preventing them from being influenced by others' scores.

[0205] Each program and performer is comprehensively evaluated and scored through a combination of scores from the audience, judges, and an intelligent system. This prevents situations where judges and audience members are unable to provide objective and professional scores for individual art programs.

[0206] The foregoing has provided a detailed description of a multi-scenario universal scoring system and scoring method provided in this application. The specific embodiments described are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A multi-scenario universal scoring system, characterized in that, It includes an event information management unit, a scoring criteria management unit, a judge information management unit, a participant information management unit, an event process control unit, and a data management unit; The event information management unit includes an event information acquisition module and an event information processing module. The event information acquisition module is used to collect basic event data and event theme data in real time. The event information processing module is used to perform preliminary data conversion and program theme data identification on the basic event data and event theme data collected by the event information acquisition module. The scoring criteria management unit includes a music program scoring module and a dance movement scoring module. The music program scoring module is used to score singing or instrumental performances, while the dance movement scoring module is used to score dance performances. The judge information management unit is used to collect judge information, generate judge scoring communication paths, and record judges' scores for the competition after each performance. The contestant information management unit is used to collect contestant entry information and determine in real time during the competition whether the corresponding contestant is on the stage for the entire program. It also matches the judges' scores for the competition after each performance and scores the contestants accordingly. The event flow control unit is used to control the flow of the entire event performance and to regulate the sound and lighting control systems of the on-site equipment. The data management unit is used to manage the writing, modification, and deletion processes of basic event data, event theme data, judge information, and contestant entry information.

2. The multi-scenario universal scoring system according to claim 1, characterized in that, The basic program data collected by the event information management unit includes individual audience data upon entry, individual judge data, and real-time uploaded audio and video data during the live performances; the program theme data includes the list of all programs in the entire competition, program type data, and program performance duration.

3. The multi-scenario universal scoring system according to claim 2, characterized in that, The standard scoring module for music programs uses a combined scoring method based on lyrics and pitch to evaluate the programs. This combined scoring method employs a factor-based evaluation formula: Wherein, k1 and k2 are the weights of lyrics and pitch in a singer's performance of a specific song, with a default setting of k1 + k2 = 1; d1 and d2 are the difficulty coefficients of the lyrics and pitch of the song, respectively, and we stipulate that d1 and d2 are both greater than 1; D(mfcc) and D(pitch) correspond to the differences in lyrics and pitch between the singer's performance and the standard original song, respectively; Vi_A and Vi_B represent the sound intensity of the standard sound template and the sound template to be matched.

4. The multi-scenario universal scoring system according to claim 3, characterized in that, The pitch of the program audio corresponds to the high and low changes of the fundamental frequency. The feature extraction method for the fundamental frequency adopts the autocorrelation fundamental frequency detection algorithm. Speech signal pre-emphasis; Speech signal framing and windowing; Substitute into the autocorrelation function; Set the peak threshold; The fundamental frequency is determined by the location of the peak.

5. A multi-scenario universal scoring system according to claim 4, characterized in that, The lyrics extraction method for the program audio uses the Mel-Cepstral Coefficient Extraction Algorithm: Preprocess the audio signal; The time-domain waveform of the preprocessed audio signal is obtained by fast Fourier transform; The sound signal is passed through a MEL filter bank to obtain the Mel spectrum coefficients; Finally, the Mel cepstral coefficients are obtained through discrete cosine transform.

6. The multi-scenario universal scoring system according to claim 5, characterized in that, The dance movement standard scoring module uses a quality score decoupled movement quality assessment algorithm to score dance movement programs. The quality score decoupled movement quality assessment algorithm includes a feature extraction network combining dynamic and static flows, a quality score decoupling module, and a pairwise ranking prediction module. The detailed process for scoring dance performances includes: First, each video is divided into T segments of equal length, and then fed into a Siamese neural network; Dynamic and static flow features are extracted using the ResNet static flow network. Then the dynamic flow feature F Di and static flow characteristics F Si Input two quality score decoupling modules that do not share weights to obtain the dynamic flow feature F. Di High-quality score features F DPi Low quality score characteristic F DCi and the characteristic F of average aggregation DUi For static flow F Si Similarly, three features F are obtained. SPi F SCi and F SUi ; Concatenate the corresponding features into F Pi F Ci and F Ui Indicates video V i The final feature is to input these three features into the pairwise ranking prediction module, which consists of a quality score predictor. The obtained dynamic flow and static flow three branch features are concatenated together, and the predicted scores are obtained by regression. The sum of these scores is the total score.

7. A multi-scenario universal scoring system according to claim 6, characterized in that, The detailed process of collecting judge information and generating judge scores includes the following steps: Collect judges' personal data, which includes their unique personal identification code at the event, their name, and their phone number. The unique personal identification codes, judges' names, and judges' phone numbers at the event will be uniformly converted into strings. Convert a string to binary code; Fill all the binary codes into the number cells composed of 64 rectangular boxes; Generate a QR code and return it to the individual judge via phone number or communication software connected to the phone number; Judges scan a generated QR code after each program ends and before the next program begins to upload their scores for each program.

8. A multi-scenario universal scoring system according to claim 7, characterized in that, The scoring formula for each performance is as follows: F 总 =F 观众 ·W1+F 评委 • W2+ (ratings for music programs or dance moves) • W3; F 观众 The average score given by the audience, W1 is the audience weighting coefficient, F 评委 W1 represents the average score given by the judges, W2 represents the judges' weighting coefficient, and W3 represents the intelligent system's weighting coefficient.

9. A multi-scenario universal scoring system according to claim 8, characterized in that, The scoring process for each performance by the scoring criteria management unit includes the following steps: S001. The system scores programs based on their list and type, using the corresponding scoring method. S002. The total score given by the audience is divided by the number of audience members who gave the score to obtain the average score given by the audience; the audience score generation communication path and detailed scoring process are the same as those of the judges. S003. Collect the program score given by each judge, remove the highest and lowest scores, divide by (total number of judges - 2) to get the average score given by the judges. S004. Substitute all the scores from steps S001, S002, and S003 into the scoring formula for each performance program to calculate the total score for the performance program.

10. A universal scoring method for multiple scenarios, characterized in that, Including the following steps: S1. Collect personal data of the audience, judges and performers at the entrance of the event venue; S2. Real-time acquisition of audio and video data of the program on the performance stage at the event site using audio and video acquisition equipment; S3. Determine the current program type based on the program theme data collected before the performance begins; The current program type is matched and searched in the database based on the input or collected program type data; If it is a music program or a crosstalk program, proceed to step S4; if it is a dance program or an acrobatic program, proceed to step S5; if it is a music and dance program, proceed to step S6. S4. Use a combined scoring method of lyrics and pitch to score music programs; S5. A quality score-decoupled motion quality assessment algorithm is used to score dance moves. S6. Classify the program's audio and video data and use the scoring methods in S4 and S5 to score the program's music and dance moves respectively. S7. Multiply the program score calculated in steps S4 and S5 by the intelligent system weight coefficient, add the average score given by the audience by the audience weight coefficient, and add the average score given by the judges by the judges weight coefficient to obtain the total score of a single type of program. S8. Sum the music score and dance score calculated in step S6, divide by 2, multiply the resulting program score by the intelligent system weight coefficient, add the average score given by the audience multiplied by the audience weight coefficient, and add the average score given by the judges multiplied by the judges weight coefficient to obtain the total score of the multiple program types.

Images