A digital human-assisted service method based on video customer service applications

Abstract

The present invention discloses a digital human-assisted service method based on video customer service applications. The method includes: dynamically capturing and collecting video gesture and motion data of customer service behaviors at different levels, determining the acquisition time of gesture and motion data for all behaviors to be inspected and the abnormal manifestation forms of gesture and motion data when violations occur; classifying the video gesture and motion data of customer service behaviors at different levels using Naive Bayes, and using the behavior relevance corresponding to the confirmatory factor analysis algorithm and the specified set value of the behavior relevance to determine different inspection normal values obtained when using the confirmatory factor analysis algorithm to check the classified gesture and motion data of customer service behavior videos; using the obtained different inspection normal values; using the confirmatory factor analysis algorithm to check the classified gesture and motion data of customer service behavior videos. This method can effectively improve the transmission efficiency of video gesture and motion data of customer service behaviors and ensure the stability of the transmission of video gesture and motion data of customer service behaviors.

Description

Technical Field

[0001] The present invention belongs to the technical field of checking gesture and movement data of customer service violation behaviors, and particularly relates to a digital human assisted service method based on video customer service applications. Background Art

[0002] For existing digital human virtual anchors and virtual image products, the operation process includes anchor video acquisition, data processing, model training, and image output; in the production and output stage, based on the trained anchor image, according to the input text, voice, plus optional anchor emotions, background pictures, videos, standing postures, sitting postures, etc., video generation and output are performed; it can only achieve the driving from text and sound to expressions, without body movements; it can only use pictures or prefabricated videos as backgrounds for synthesis output, and cannot be synthesized with three-dimensional graphic packaging content; nor can it control the behavior norms of digital humans. Therefore, the existing video data control method based on digital human virtual anchors has problems of insufficient flexibility and low video data transmission efficiency. Summary of the Invention

[0003] The present invention provides a digital human assisted service method based on video customer service applications to overcome the problems of large amount of gesture and movement data transmitted for checking gesture and movement data and high power consumption of transmission devices currently.

[0004] Step J1: Dynamically capture and collect gesture and movement data of customer service behavior videos at different levels, and determine the acquisition time of gesture and movement data of all behaviors to be checked and the abnormal manifestation forms of gesture and movement data when violation behaviors occur.

[0005] Step J2: Classify gesture and movement data of customer service behavior videos at different levels using Naive Bayes, and perform behavior correlation analysis on the classified gesture and movement data using the confirmatory factor analysis algorithm. The behavior correlation is: the movement trajectory of gesture and movement data of customer service behavior videos within a set time.

[0006] Step J3: Use the behavior correlation corresponding to the confirmatory factor analysis algorithm and the specified set value of the behavior correlation to determine different normal inspection values obtained when using the confirmatory factor analysis algorithm to check the classified gesture and movement data of customer service behavior videos.

[0007] Step J4: Use the obtained different normal inspection values to determine the time for checking the gesture and movement data dynamically captured from customer service behavior videos.

[0008] Step J5: Use the confirmatory factor analysis algorithm to check the classified gesture and movement data of customer service behavior videos, and simultaneously generate the solutions of the gesture and movement data violation behavior inspection function and the solutions of the digital human abnormal behavior video function.

[0009] Further, the expression of the posture and motion data violation behavior checking function is:

[0010] g x =(R f +QP x )·U

[0011] where g x represents the posture and motion data violation behavior checking function, R f represents the standard value of the normal transmission of the posture and motion data, Q represents the transmission rate of the posture and motion data, P x represents the function of the transmission channel bandwidth of the posture and motion data, and U represents the factor of the posture and motion data violation behavior;

[0012] Further, the expression of the abnormal behavior video function of the digital human is:

[0013]

[0014] where represents the abnormal behavior video function of the customer service digital human, φ represents the abnormal factor of the digital human, L represents the set of abnormal positions of the digital human, S represents the inspection time, B represents the total number of abnormal reasons of the digital human, m represents the number of iterations, and F v represents the digital human setting standard function.

[0015] Further, after obtaining the solutions of the posture and motion data violation behavior checking function and the abnormal behavior video function of the digital human, it further includes:

[0016] Supplementing the behavior relevance, the specified set value of the behavior relevance, and the algorithm flag of the confirmatory factor analysis algorithm to the header of the posture and motion data to be transmitted; and supplementing the solutions of the posture and motion data violation behavior checking function and the abnormal behavior video function of the digital human to the control equation of the posture and motion data to be transmitted; and sending the posture and motion data to be transmitted to the customer service behavior video control platform.

[0017] Further, using the confirmatory factor analysis algorithm to check the posture and motion data classified by the customer service behavior video, and simultaneously generating the solutions of the posture and motion data violation behavior checking function and the abnormal behavior video function of the digital human, includes:

[0018] Using the specified set value of the behavior relevance as the classification standard when checking the posture and motion data classified by the customer service behavior video;

[0019] For the posture and motion data classified by the customer service behavior video of all specified set values, when implementing the confirmatory factor analysis algorithm, setting the corresponding behavior to be checked in the solution of the posture and motion data violation behavior checking function as the inspection set position;

[0020] Append the gain value corresponding to the classified gesture action data of the customer service behavior video with the specified setting value determined by the confirmatory factor analysis algorithm to the end of the solution of the digital human abnormal behavior video function;

[0021] When the confirmatory factor analysis algorithm is not implemented, set the current position corresponding to the solution of the gesture action data violation behavior checking function to the unchecked position; and append the classified gesture action data of the customer service behavior video to the end of the solution of the digital human abnormal behavior video function.

[0022] Further, the confirmatory factor analysis algorithm is a full gesture action data check:

[0023] Take the behavior relevance as the first behavior to be checked in the classified gesture action data of the customer service behavior video;

[0024] For all behaviors to be checked in the classified gesture action data of the customer service behavior video, perform all operations on the values of the behavior to be checked and the values of the previous behavior to be checked. All the obtained values are the gain values corresponding to the behavior to be checked in the solution of the digital human abnormal behavior video function.

[0025] Further, the confirmatory factor analysis algorithm is a partial gesture action data check: Take the behavior relevance as the first behavior to be checked in the classified gesture action data of the customer service behavior video;

[0026] For all behaviors to be checked in the classified gesture action data of the customer service behavior video, perform partial operations on the values of the behavior to be checked and the values of the previous behavior to be checked. The solution of the partial operation is the gain value corresponding to the behavior to be checked in the solution of the digital human abnormal behavior video function.

[0027] Further, the confirmatory factor analysis algorithm is a time gesture action data check: Take the behavior relevance as all behaviors to be checked within the first complete sampling period in the classified gesture action data of the customer service behavior video;

[0028] Group by the number of behaviors to be checked within the sampling period, and divide all behaviors to be checked in the classified gesture action data of the customer service behavior video into multiple groups of gesture action data to be processed;

[0029] For each group of gesture action data to be processed, perform a point-by-point coincidence operation on the group of gesture action data to be processed and the previous group of gesture action data to be processed. The obtained coincidence value is the gain value corresponding to the group of gesture action data to be processed in the solution of the digital human abnormal behavior video function.

[0030] Further, the confirmatory factor analysis algorithm is variance gesture action data checking: taking the behavior relevance as all the behaviors to be checked within the first complete variance period of the gesture action data classified by the customer service behavior video;

[0031] Grouping according to the number of all the behaviors to be checked within the variance period, and dividing all the behaviors to be checked in the gesture action data classified by the customer service behavior video into multiple groups of gesture action data to be processed;

[0032] For each group of gesture action data to be processed, performing variance operation point by point between the group of gesture action data to be processed and its previous group of gesture action data to be processed, and the obtained variance value is the gain value corresponding to the group of gesture action data to be processed in the solution of the digital human abnormal behavior video function.

[0033] Further, the normal check value is: the sum of the standard value of the solution of the gesture action data violation check function and the standard value of the solution of the digital human abnormal behavior video function divided by the standard value of the gesture action data classified by the customer service behavior video.

[0034] Further, after sending the gesture action data to be transmitted to the customer service behavior video control platform, it further includes: analyzing the behavior relevance, the specified setting value of the behavior relevance, the check setting position, the solution of the gesture action data violation check function, and the solution of the digital human abnormal behavior video function from the received gesture action data, implementing the confirmatory factor analysis algorithm corresponding to the check setting position to obtain the gesture action data classified by the customer service behavior video before the check;

[0035] Among them, the confirmatory factor analysis algorithm corresponding to the check setting position and the confirmatory factor analysis algorithm corresponding to the check setting position are two-way operations.

[0036] Further, the customer service behavior video control platform uses a 5G wireless network for gesture action data transmission.

[0037] Compared with the prior art, a digital human assisted service method based on video customer service application provided by the present invention, aiming at the characteristic of high similarity of the checked gesture action data, checks the waveform gesture action data by using behavior relevance in the customer service behavior video control platform, reduces the amount of gesture action data, and reduces the transmission burden; reduces the bandwidth occupation time of the transmission channel in the customer service behavior video control platform, improves the transmission efficiency, and ensures the stability of the transmission of the gesture action data of the customer service behavior video. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the first process schematic diagram of the method of the present invention;

[0039] Figure 2Schematic diagram of the second process of the method of the present invention;

[0040] Figure 3 Schematic diagram of the third process of the method of the present invention. Detailed implementation manners

[0041] Now, exemplary embodiments of the present invention will be described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely, and to fully convey the scope of the present invention to those skilled in the art. The terms in the exemplary embodiments shown in the drawings are not intended to limit the present invention. In the drawings, the same unit / element is denoted by the same reference numeral.

[0042] Unless otherwise specified, the terms (including scientific and technical terms) used herein have the ordinary meaning understood by those skilled in the art. In addition, it can be understood that the terms defined in a commonly used dictionary should be understood as having a meaning consistent with the context of the relevant field, and should not be understood as having an idealized or overly formal meaning.

[0043] Currently, in various scenarios, inspection gesture and motion data is transmitted between different devices. When transmitting inspection gesture and motion data between these devices, a local interface is usually utilized and a low-rate communication medium is adopted, such as a USB gesture and motion data cable. Currently, during the transmission of inspection gesture and motion data, a large amount of similar waveform inspection gesture and motion data needs to occupy the bandwidth for a long time. Therefore, the transmission efficiency of inspection gesture and motion data is low and the device power consumption is large.

[0044] Therefore, when transmitting inspection gesture and motion data to the customer service behavior video control platform side using a pre-agreed communication protocol at both ends of the transmission channel, it is a feasible method to improve the transmission efficiency of inspection gesture and motion data by processing the inspection gesture and motion data and then supplementing it into the inspection gesture and motion data as the inspection gesture and motion data payload for transmission.

[0045] In an embodiment of the present invention, when the customer service behavior video control platform at one end of the transmission channel sends inspection gesture and motion data using a pre-set communication protocol, the original inspection gesture and motion data dynamically captured by the metering chip is processed and then transmitted as the payload part in the inspection gesture and motion data; when the customer service behavior video control platform at the other end of the transmission channel receives inspection gesture and motion data using a pre-set communication protocol, the payload part is analyzed from the inspection gesture and motion data, and after de-inspection processing, the original inspection gesture and motion data dynamically captured by the metering chip is obtained.

[0046] Processing the original inspection gesture and motion data using the gesture and motion data inspection method according to an embodiment of the present invention significantly reduces the scale of the gesture and motion data, thereby improving the transmission efficiency of the gesture and motion data.

[0047] A digital human assisted service method based on video customer service applications provided by the present invention, aiming at the characteristic of high similarity of inspection gesture and motion data, before transmission, uses behavioral relevance to inspect the waveform gesture and motion data, reduces the amount of gesture and motion data, and alleviates the transmission burden; reduces the bandwidth occupation time of the transmission channel in the customer service behavior video control platform, improves the transmission efficiency; is conducive to reducing the transmission speed of the gesture and motion data in the customer service behavior video control platform and achieving low-speed and high-efficiency transmission.

[0048] A digital human assisted service method based on video customer service applications provided by the present invention is particularly suitable for application in scenarios where the local interface communication rate is limited.

[0049] The following gives the definitions of each term:

[0050] The standard value of the curve gesture and motion data is the storage space occupied by the gesture and motion data to be processed, that is, the number of bytes.

[0051] Behavioral relevance is the classified gesture and motion data of the customer service behavior video of one or several consecutive behaviors to be inspected.

[0052] The standard value of behavioral relevance, that is, the specified set value of behavioral relevance, is the classification standard for all inspection operations during gesture and motion data processing; the original curve gesture and motion data are grouped according to this classification standard and processed one by one to form the solution of the gesture and motion data violation inspection function and the solution of the digital human abnormal behavior video function.

[0053] Among them, the solution of the gesture and motion data violation inspection function is a sequence constructed in units of bytes, and all bytes are used to describe whether the inspected gesture and motion data is the same as the classified gesture and motion data of the customer service behavior video according to the specified set value or classification standard.

[0054] The standard value (in units of bytes) of the solution of the gesture and motion data violation inspection function is:

[0055] Curve gesture and motion data standard value / specified set value / 8;

[0056] As a qualitative description part, the storage space occupied by the solution of the gesture and motion data violation inspection function is small, and it can also be considered as the summary part of the gesture and motion data.

[0057] The solution of the digital human abnormal behavior video function is used to store the gain value corresponding to the original behavior to be inspected that is the same as the specified set value of behavioral relevance or the original gesture and motion data of the behavior to be inspected.

[0058] The solution of the digital human abnormal behavior video function and the feature series are interdependent;

[0059] For example, within the solution of the digital human abnormal behavior video function, if the byte bit of the solution of the corresponding gesture action data violation check function is 1, then at the corresponding specified setting value, it is the gain value corresponding to the original behavior gesture action data to be inspected. At this time, the specified setting value is the specified setting value after inspection; if the byte bit of the solution of the corresponding gesture action data violation check function is 0, then at the corresponding specified setting value, it is the original behavior gesture action data to be inspected. At this time, the specified setting value is the specified setting value before inspection.

[0060] That is to say, in the solution of the digital human abnormal behavior video function, there are two types of specified setting values; and the specified setting value before inspection is not less than the specified setting value after inspection.

[0061] After the above inspection, the inspection normal value is:

[0062] (Standard value of the solution of the gesture action data violation check function + Standard value of the solution of the digital human abnormal behavior video function) / Standard value of the original curve gesture action data

[0063] It should be understood that the above definition of the inspection normal value ignores the storage space occupied by the specified setting value of behavior relevance, the behavior relevance, and the identifier of the descriptor inspection algorithm.

[0064] Such as Figure 1 As shown, a digital human assisted service method based on video customer service application in an embodiment includes the following steps:

[0065] Step J1: Dynamically capture and collect the gesture action data of customer service behavior videos at different levels, and determine the acquisition time of the gesture action data of all behaviors to be inspected and the abnormal manifestation forms of the gesture action data when violations occur;

[0066] Step J2: Classify the gesture action data of customer service behavior videos at different levels using Naive Bayes, and use the confirmatory factor analysis algorithm to analyze the behavior relevance of the classified gesture action data. The behavior relevance is: the movement trajectory of the gesture action data of the customer service behavior video within a set time;

[0067] Step J3: Use the behavior relevance corresponding to the confirmatory factor analysis algorithm and the specified setting value of the behavior relevance to determine different inspection normal values obtained when using the confirmatory factor analysis algorithm to process the classified gesture action data of the customer service behavior video;

[0068] Step J4: Determine the time for dynamically capturing and inspecting the posture and motion data of the customer service behavior video using the obtained different inspection normal values.

[0069] Step J5: Use the confirmatory factor analysis algorithm to process the classified posture and motion data of the customer service behavior video, and simultaneously generate the solutions of the posture and motion data violation behavior inspection function and the digital human abnormal behavior video function.

[0070] The posture and motion data violation behavior inspection function has the expression:

[0071] g x =(R f +QP x )·U

[0072] where g x represents the posture and motion data violation behavior inspection function, R f represents the standard value of normal transmission of the posture and motion data, Q represents the transmission rate of the posture and motion data, P x represents the posture and motion data transmission channel bandwidth function, and U represents the posture and motion data violation behavior factor;

[0073] The digital human abnormal behavior video function has the expression:

[0074]

[0075] where represents the customer service digital human abnormal behavior video function, φ represents the digital human abnormal factor, L represents the set of abnormal positions of the digital human, S represents the inspection time, B represents the total number of abnormal reasons of the digital human, m represents the number of iterations, and F v represents the digital human setting standard function.

[0076] As Figure 2 shown, specifically, after generating the solutions of the posture and motion data violation behavior inspection function and the digital human abnormal behavior video function, it further includes:

[0077] Step N1: Supplement the behavior relevance, the specified set value of the behavior relevance, and the algorithm flag of the confirmatory factor analysis algorithm to the header of the posture and motion data to be transmitted;

[0078] Step N2: Supplement the solutions of the posture and motion data violation behavior inspection function and the digital human abnormal behavior video function to the control equation of the posture and motion data to be transmitted;

[0079] Step N3: Send the posture and motion data to be transmitted to the customer service behavior video control platform.

[0080] As Figure 3As shown in the figure, specifically, using the confirmatory factor analysis algorithm, process the gesture action data for classifying the customer service behavior videos, and simultaneously generate the solutions of the gesture action data violation behavior inspection function and the digital human abnormal behavior video function, including:

[0081] Step G1: Set the classification standard when processing the gesture action data for classifying the customer service behavior videos with the specified set value of behavior relevance.

[0082] Step G2: For the gesture action data of the customer service behavior videos with all specified set values, when implementing the confirmatory factor analysis algorithm, set the corresponding behavior to be inspected in the solution of the gesture action data violation behavior inspection function as the inspection set position; and append the gain value corresponding to the gesture action data of the customer service behavior videos with this specified set value determined by the confirmatory factor analysis algorithm to the end of the solution of the digital human abnormal behavior video function.

[0083] Step G3: When not implementing the confirmatory factor analysis algorithm, set the current position corresponding to the solution of the gesture action data violation behavior inspection function as the inspection non-set position; and append the gesture action data of the customer service behavior videos to the end of the solution of the digital human abnormal behavior video function.

[0084] Specifically, the confirmatory factor analysis algorithm is the gesture action data coincidence check:

[0085] Take the behavior relevance as the first behavior to be inspected in the gesture action data of the customer service behavior videos.

[0086] For all behaviors to be inspected in the gesture action data of the customer service behavior videos,

[0087] Compare the value of the behavior to be inspected with the value of the previous behavior to be inspected, and the obtained ratio is the gain value corresponding to the behavior to be inspected in the solution of the digital human abnormal behavior video function.

[0088] Specifically, the confirmatory factor analysis algorithm is the gesture action data coincidence check:

[0089] Take the behavior relevance as the first behavior to be inspected in the gesture action data of the customer service behavior videos.

[0090] For all behaviors to be inspected in the gesture action data of the customer service behavior videos,

[0091] Perform a coincidence function operation on the value of the behavior to be inspected and the value of the previous behavior to be inspected, and the obtained coincidence value is the gain value corresponding to the behavior to be inspected in the solution of the digital human abnormal behavior video function.

[0092] Specifically, for the digital human assisted service method based on video customer service applications, the confirmatory factor analysis algorithm is the time gesture action data check:

[0093] Take the behavior relevance as all the to-be-verified behaviors within the first complete sampling period of the gesture action data for classifying customer service behavior videos.

[0094] Group by the number of to-be-verified behaviors within the sampling period, and divide all the to-be-verified behaviors in the gesture action data for classifying customer service behavior videos into multiple to-be-processed gesture action data groups.

[0095] For each to-be-processed gesture action data group, perform a point-by-point ratio operation or coincidence operation between the to-be-processed gesture action data group and its previous to-be-processed gesture action data group. The obtained ratio or coincidence value is the gain value corresponding to the to-be-processed gesture action data in the solution of the digital human abnormal behavior video function.

[0096] Specifically, the confirmatory factor analysis algorithm is the coincidence gesture action data check:

[0097] Take the behavior relevance as all the to-be-verified behaviors within the first complete coincidence period of the gesture action data for classifying customer service behavior videos.

[0098] Group by the number of all the to-be-verified behaviors within the coincidence period, and divide all the to-be-verified behaviors in the gesture action data for classifying customer service behavior videos into multiple to-be-processed gesture action data groups.

[0099] For each to-be-processed gesture action data group, perform a point-by-point ratio operation or coincidence operation between the to-be-processed gesture action data group and its previous to-be-processed gesture action data group. The obtained ratio or coincidence value is the gain value corresponding to the to-be-processed gesture action data in the solution of the digital human abnormal behavior video function.

[0100] Specifically, the check normal value is: the sum of the standard value of the solution of the gesture action data violation check function and the standard value of the solution of the digital human abnormal behavior video function, divided by the standard value of the gesture action data for classifying customer service behavior videos.

[0101] Specifically, after sending the to-be-transmitted gesture action data to the customer service behavior video control platform, it further includes:

[0102] Analyze the behavior relevance, the specified set value of the behavior relevance, the check set position, the solution of the gesture action data violation check function, and the solution of the digital human abnormal behavior video function from the received gesture action data.

[0103] Implement the confirmatory factor analysis algorithm corresponding to the solution at the check set position to obtain the gesture action data for classifying customer service behavior videos before the check; among them, the confirmatory factor analysis algorithm corresponding to the solution at the check set position and the confirmatory factor analysis algorithm corresponding to the check set position are two-way operations.

[0104] It should be understood that the inspection gesture action data is usually transmitted between the customer service behavior video control platforms according to a certain protocol. Among all the gesture action data of this protocol, it includes a frame header part and a frame gesture action data area part. Among them, the frame gesture action data area part is the payload of the gesture action data. This inspection method only involves the content added to the gesture action data header part or the gesture action data body part in the gesture action data area part; the processing steps for the frame header part can be carried out in the manner disclosed in the prior art and will not be elaborated here.

[0105] Generally, the inspection frequency is a fixed value, and the duration of each sampling period is the reciprocal of the inspection frequency.

[0106] The gesture action data volume corresponding to the sampling periods of multiple channels can be determined by the following formula:

[0107] Sampling frequency × acquisition accuracy of a single behavior to be inspected × number of channels.

[0108] Therefore, among the gesture action data returned by the metering chip, within all seconds, the gesture action data volume corresponding to the sampling period of a single channel can be determined by the following formula:

[0109] Inspection frequency × sampling frequency × acquisition accuracy of a single behavior to be inspected.

[0110] As above, using the number of channels, the specified set value, and the sampling accuracy, the gesture action data scale of the gesture action data for classifying the customer service behavior video to be inspected can be determined, that is, the standard value of the original sampled gesture action data.

[0111] It should be understood that the sampled gesture action data is transmitted sequentially according to a single channel, that is, within each gesture action data, only the sampled gesture action data from a single sampling channel is transmitted.

[0112] On the other hand, the sampling accuracies and gesture action data accuracies of multiple channels may vary. Therefore, each channel independently uses its own inspection algorithm, including the behavior relevance of its own specified set value, the solution of the gesture action data violation behavior inspection function, and the solution of the digital human abnormal behavior video function.

[0113] Dynamically capture and collect the gesture action data of different levels of customer service behavior videos, and determine the gesture action data acquisition time of all behaviors to be inspected and the abnormal manifestation forms of the gesture action data when violations occur, specifically including:

[0114] Traverse the waveform curve pose action data within the classified pose action data of the customer service behavior videos obtained, analyze the correlation between the pose action data of the behavior to be inspected, and determine the acquisition time of the pose action data of all behaviors to be inspected and the abnormal manifestation form of the pose action data when a violation occurs.

[0115] The correlation between the pose action data of the behavior to be inspected reflects the pose action data pattern. The entire classified pose action data of the customer service behavior videos can be described using behavior relevance and pose action data pattern:

[0116] For example, use a certain behavior to be inspected as the behavior relevance, and use the ratio or coincidence value between the other behaviors to be inspected and this behavior relevance to describe the other behaviors to be inspected;

[0117] For example, use a certain period as the behavior relevance, and use the ratio or coincidence value between each point of other periods and each point of this behavior relevance to describe other periods.

[0118] When different pose action data inspection algorithms are used for all batches of pose action data to be transmitted, the normal inspection values obtained can be different.

[0119] Define the normal inspection value as the ratio of the storage space occupied by the pose action data before and after inspection; the larger the normal inspection value, the less transmission bandwidth the pose action data after inspection occupies, and the higher the transmission efficiency.

[0120] For the curve waveform pose action data of a single channel, the same inspection algorithm can always be used; or as time progresses, it may be changed to other different inspection algorithms.

[0121] For inspecting the pose action data, the following different-level sub-inspection algorithms can be used, including: coincidence of pose action data inspection, coincidence of pose action data inspection, time pose action data inspection, coincidence of curve inspection, etc. When assembling the pose action data, the inspection setting bit can be used to identify the target inspection algorithm adopted.

[0122] For any single-channel pose action data obtained from the sampling chip, the sub-inspection algorithm with the optimal normal inspection value can be selected as the inspection algorithm for the current classified pose action data of the customer service behavior videos.

[0123] The coincidence of pose action data inspection is a single-point inspection method. Specifically, take the behavior relevance as the behavior to be inspected with the largest or smallest value or the first behavior to be inspected in the waveform curve;

[0124] For all behaviors to be inspected in the classified pose action data of the customer service behavior videos,

[0125] Compare the value of the behavior to be inspected with the value of its previous behavior to be inspected. The obtained ratio is the gain value corresponding to the behavior to be inspected in the solution of the digital human abnormal behavior video function.

[0126] For example, the acquisition precision of all behaviors to be inspected in the waveform curve is 4 bytes; the specified set value corresponding to the numerical difference between any two behaviors to be inspected in the waveform curve is 2 bytes. Then, after the coincidence attitude action data check, the normal check value is approximately 2. Here, "approximate" means ignoring the attitude action data standard value in the solution of the attitude action data violation check function.

[0127] Use the ratio between the behavior correlation and the first behavior to be inspected in the classified attitude action data of the customer service behavior video as the gain value of the first behavior to be inspected. Starting from the second behavior to be inspected in the classified attitude action data of the customer service behavior video, make an incremental comparison with the previous behavior to be inspected point by point:

[0128] Take the behavior correlation as the first behavior to be inspected in the classified attitude action data of the customer service behavior video;

[0129] For all behaviors to be inspected in the classified attitude action data of the customer service behavior video,

[0130] Perform a coincidence function operation on the value of the behavior to be inspected and the value of its previous behavior to be inspected. The obtained coincidence value is the gain value corresponding to the behavior to be inspected in the solution of the digital human abnormal behavior video function.

[0131] For example, the acquisition precision of all behaviors to be inspected in the waveform curve is 4 bytes; the specified set value corresponding to the quotient of the numerical values of any two behaviors to be inspected in the waveform curve is 2 bytes. Then, after the coincidence attitude action data check, the normal check value is approximately 2.

[0132] That is, by performing the coincidence function operation, approximately check the specified set value of the gain value corresponding to the behavior to be inspected as one byte, which is less than the specified set value of the original behavior to be inspected.

[0133] Specifically, the time attitude action data verification factor analysis algorithm is as follows:

[0134] Take the behavior correlation as all behaviors to be inspected within the first complete sampling period in the classified attitude action data of the customer service behavior video;

[0135] Group according to the number of behaviors to be inspected within the sampling period, and divide all behaviors to be inspected in the classified attitude action data of the customer service behavior video into multiple groups of attitude action data to be processed;

[0136] For each group of attitude action data to be processed,

[0137] Perform a point-by-point ratio operation or coincidence operation on the to-be-processed posture action data group and its previous to-be-processed posture action data group. The obtained ratio or coincidence value is the gain value corresponding to the to-be-processed posture action data in the solution of the digital human abnormal behavior video function.

[0138] The time posture action data check is a continuous multi-point check method. Specifically, take the behavior relevance as the behavior to be inspected in a complete sampling period of the waveform curve (at this time, the sampling period is other time functions except coincidence or all; and the initial phase of this time function is not necessarily zero);

[0139] The method of performing a point-by-point ratio operation or coincidence operation on the to-be-processed posture action data group and its previous to-be-processed posture action data group is the same as the coincidence check or non-coincidence check, which will not be elaborated here.

[0140] Specifically, the coincidence posture action data verification factor analysis algorithm is as follows:

[0141] Take the behavior relevance as all the behaviors to be inspected within the first complete coincidence period of the customer service behavior video classification posture action data;

[0142] Group according to the number of all the behaviors to be inspected within the coincidence period, and divide all the behaviors to be inspected in the customer service behavior video classification posture action data into multiple to-be-processed posture action data groups;

[0143] For each to-be-processed posture action data group,

[0144] Perform a point-by-point ratio operation or coincidence operation on the to-be-processed posture action data group and its previous to-be-processed posture action data group. The obtained ratio or coincidence value is the gain value corresponding to the to-be-processed posture action data in the solution of the digital human abnormal behavior video function.

[0145] The method of performing a point-by-point ratio operation or coincidence operation on the to-be-processed posture action data group and its previous to-be-processed posture action data group is the same as the coincidence check or non-coincidence check, which will not be elaborated here.

[0146] The coincidence posture action data check is a continuous multi-point check method. Specifically, take the behavior relevance as the behavior to be inspected in a complete sampling period of the coincidence function in the waveform curve, or the peak or valley value in the coincidence function;

[0147] Group according to the number of behaviors to be inspected, and perform a point-by-point ratio operation or coincidence operation on the values of other behaviors to be inspected in the waveform curve and this behavior relevance. The obtained ratio or coincidence value is the same number of posture action data points in the solution of the digital human abnormal behavior video function as the number of behaviors to be inspected in the sampling period.

[0148] Furthermore, the sub-check algorithm can be optionally adopted or not to describe other behaviors to be inspected that belong to the same batch of customer service behavior video classification gesture action data related to the behavior association, and they are identified using the inspection setting position sequence. For example, if the current behavior association description is used, the position is set to 1; if the current behavior association description is not used, the position is set to 0.

[0149] After the gesture action data inspection and before starting the gesture action data transmission, the following steps are used to assemble the gesture action data:

[0150] Supplement the gesture action data header with the behavior association;

[0151] Supplement the gesture action data header with the specified set value of the behavior association;

[0152] Supplement the gesture action data body with the solution of the gesture action data violation inspection function;

[0153] Supplement the gesture action data body with the solution of the digital human abnormal behavior video function.

[0154] Among them, the behavior association can be the behavior association determined by dynamic capture during inspection or the pre-set behavior association;

[0155] Among them, in the solution of the gesture action data violation inspection function, if the waveform curve behavior to be inspected uses the current behavior association description, then the gesture action data point at the corresponding position in the inspection setting position sequence is set to 1; if the current behavior association description is not used, then the gesture action data point at the corresponding position in the inspection setting position sequence is set to 0.

[0156] Among them, the solution of the digital human abnormal behavior video function is generated by successively describing the subsequent waveform curve behaviors to be inspected using the behavior association. Usually, the standard value or width of all gesture action data points in the solution of the gesture action data violation inspection function is smaller than the standard value or width of the behavior association.

[0157] In addition, for the dynamic capture and collection of gesture action data of customer service behavior videos at different levels, it also includes:

[0158] Determine the standard value of the customer service behavior video classification gesture action data obtained and send it to the customer service behavior video control platform side, so that when the customer service behavior video control platform uses the received gesture action data for solution inspection to restore the inspection gesture action data, comparison or verification can be performed.

[0159] On the side of the customer service behavior video control platform, for the posture action data received from the transmission channel, according to the agreed protocol, analyze the behavior relevance, the specified set value of the behavior relevance, the inspection set position, the solution of the posture action data violation inspection function, and the solution of the digital human abnormal behavior video function from the posture action data, and obtain the inspected posture action data after inspection; or

[0160] Store the behavior relevance, the specified set value of the behavior relevance, the inspection set position, the solution of the posture action data violation inspection function, and the solution of the digital human abnormal behavior video function, and start the inspection operation after receiving the inspection request.

[0161] Traversingly search for the behavior relevance and the specified set value of the behavior relevance in the curve posture action data to be inspected;

[0162] For any original sampled posture action data, the specified set value is used to determine the classification standard when performing posture action data inspection, that is, the width of the posture action data for all inspection operations.

[0163] Using the behavior relevance, select the corresponding confirmatory factor analysis algorithms to inspect the curve posture action data respectively, and obtain the inspected posture action data;

[0164] When this method is specifically implemented, an inspection algorithm library is developed. This inspection algorithm library contains several confirmatory factor analysis algorithms for posture action data, such as: whether the posture action data coincides, time posture action data inspection, partial posture action data inspection, etc.

[0165] Calculate the ratio of the standard value of the original curve posture action data to the standard value of the inspected posture action data when using different confirmatory factor analysis algorithms respectively, and use it as the inspection normal value;

[0166] Select the sub-inspection algorithm with the largest inspection normal value as the confirmatory factor analysis algorithm for the current curve posture action data.

[0167] Subsequently, use each bit in the solution of the posture action data violation inspection function to identify whether to replace the posture action data before inspection with a gain value or retain the posture action data before inspection bit by bit, and append the gain value to the end of the solution of the digital human abnormal behavior video function;

[0168] The behavior relevance, the specified set value of the behavior relevance, the inspection set position, the solution of the posture action data violation inspection function, and the solution of the digital human abnormal behavior video function together constitute the final inspected posture action data to be transmitted.

[0169] The lossless inspection algorithm has the following characteristics:

[0170] 1) Traverse the posture action data to find behavior correlations;

[0171] 2) For the waveform posture action data with good consistency in describing behavior correlations, the inspection effect is better;

[0172] 3) An inspection algorithm for best checking normal values can be selected to perform curve posture action data inspection;

[0173] 4) In the assembled posture action data, the behavior correlation, inspection setting position, the solution of the posture action data violation inspection function, and the solution of the digital human abnormal behavior video function together form the payload. The storage space occupied by this payload is adapted to the payload formed by the behavior correlation, inspection setting position, the solution of the posture action data violation inspection function, and the solution of the digital human abnormal behavior video function together.

[0174] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that without departing from the principles and spirit of the present invention, these embodiments can be subjected to equivalent changes, modifications, substitutions, and variations at different levels. The scope of the present invention is defined by the appended claims and their equivalent scope.

Claims

1. A digital human-assisted service method based on video customer service applications, characterized in that, Including the following steps: Step J1: Dynamically capture and collect the video gesture action data of customer service behaviors at different levels, determine the acquisition time of the gesture action data of all behaviors to be inspected and the abnormal manifestation forms of the gesture action data when violations occur; Step J2: Classify the video gesture action data of customer service behaviors at different levels using Naive Bayes, and perform behavior correlation analysis on the classified gesture action data using the confirmatory factor analysis algorithm. The behavior correlation is: the movement trajectory of the video gesture action data of customer service behaviors within a set time; Step J3: Use the behavior correlation corresponding to the confirmatory factor analysis algorithm and the specified set value of the behavior correlation to determine different inspection normal values obtained when using the confirmatory factor analysis algorithm to inspect the classified gesture action data of customer service behavior videos; Step J4: Use the obtained different inspection normal values to determine the time for inspecting the dynamically captured gesture action data of customer service behavior videos; Step J5: Use the confirmatory factor analysis algorithm to inspect the classified gesture action data of customer service behavior videos, and simultaneously generate the solution of the gesture action data violation behavior inspection function and the solution of the digital human abnormal behavior video function; The expression of the gesture action data violation behavior inspection function is: g x = (R f + QP x ) · U Among them, g x represents the posture action data violation check function, R f represents the standard value for the normal transmission of posture action data, Q represents the transmission rate of posture action data, P x represents the posture action data transmission channel bandwidth function, and U represents the posture action data violation factor; The expression of the digital human abnormal behavior video function is: Among them, represents the video function of the abnormal behavior of the customer service digital human, φ represents the abnormal factor of the digital human, L represents the set of abnormal positions of the digital human, S represents the inspection time, B represents the total number of abnormal reasons of the digital human, m represents the number of iterations, F v represents the digital human setting standard function.

2. The digital human-assisted service method based on video customer service application according to claim 1, wherein, After generating the solution of the gesture action data violation behavior inspection function and the solution of the digital human abnormal behavior video function, it further includes: Supplement the behavior correlation, the specified set value of the behavior correlation, and the algorithm flag of the confirmatory factor analysis algorithm to the header of the gesture action data to be transmitted; and supplement the solution of the gesture action data violation behavior inspection function and the solution of the digital human abnormal behavior video function to the control equation of the gesture action data to be transmitted; send the gesture action data to be transmitted to the customer service behavior video control platform.

3. The digital human-assisted service method based on video customer service application according to claim 1, wherein The using the confirmatory factor analysis algorithm to inspect the classified gesture action data of customer service behavior videos and simultaneously generate the solution of the gesture action data violation behavior inspection function and the solution of the digital human abnormal behavior video function includes: Taking the specified set value of the behavior correlation as the classification standard when inspecting the classified gesture action data of customer service behavior videos; For the classified gesture action data of customer service behavior videos with all specified set values, when implementing the confirmatory factor analysis algorithm, set the corresponding behavior to be inspected in the solution of the gesture action data violation behavior inspection function as the inspection set position; And append the gain value corresponding to the classified gesture action data of customer service behavior videos with the specified set value determined by the confirmatory factor analysis algorithm to the end of the solution of the digital human abnormal behavior video function; When the confirmatory factor analysis algorithm is not implemented, set the current position corresponding to the solution of the gesture action data violation behavior inspection function as the inspection non-set position; and append the classified gesture action data of the customer service behavior video to the end of the solution of the digital human abnormal behavior video function.

4. The digital human-assisted service method based on video customer service application according to claim 1, wherein, The confirmatory factor analysis algorithm is for inspecting all gesture action data: Take the behavior correlation as the first behavior to be inspected in the classified gesture action data of customer service behavior videos; For all the to-be-verified behaviors of the gesture action data classified for the customer service behavior video, all the values obtained by performing operations on the values of the to-be-verified behaviors and the values of their previous to-be-verified behaviors are the gain values corresponding to the to-be-verified behaviors in the solutions of the digital human abnormal behavior video function.

5. A digital human-assisted service method based on a video customer service application according to claim 1, characterized in that, The confirmatory factor analysis algorithm is a partial gesture action data check: taking the behavior relevance as the first to-be-verified behavior of the gesture action data classified for the customer service behavior video; For all the to-be-verified behaviors of the gesture action data classified for the customer service behavior video, performing partial operations on the values of the to-be-verified behaviors and the values of their previous to-be-verified behaviors, and the solutions of the partial operations are the gain values corresponding to the to-be-verified behaviors in the solutions of the digital human abnormal behavior video function.

6. The digital human-assisted service method based on video customer service application according to claim 1, wherein The confirmatory factor analysis algorithm is a time gesture action data check: taking the behavior relevance as all the to-be-verified behaviors within the first complete sampling period of the gesture action data classified for the customer service behavior video; Grouping according to the number of to-be-verified behaviors within the sampling period, and dividing all the to-be-verified behaviors in the gesture action data classified for the customer service behavior video into multiple to-be-processed gesture action data groups; For each of the to-be-processed gesture action data groups, performing a point-by-point coincidence operation on the to-be-processed gesture action data group and its previous to-be-processed gesture action data group, and the obtained coincidence values are the gain values corresponding to the to-be-processed gesture action data in the solutions of the digital human abnormal behavior video function.

7. The digital human-assisted service method based on video customer service application according to claim 1, wherein The confirmatory factor analysis algorithm is a variance gesture action data check: taking the behavior relevance as all the to-be-verified behaviors within the first complete variance period of the gesture action data classified for the customer service behavior video; Grouping according to the number of all the to-be-verified behaviors within the variance period, and dividing all the to-be-verified behaviors in the gesture action data classified for the customer service behavior video into multiple to-be-processed gesture action data groups; For each of the to-be-processed gesture action data groups, performing a point-by-point variance operation on the to-be-processed gesture action data group and its previous to-be-processed gesture action data group, and the obtained variance values are the gain values corresponding to the to-be-processed gesture action data in the solutions of the digital human abnormal behavior video function.

8. The digital human-assisted service method based on video customer service application according to claim 1, wherein, The normal value for inspection is: the sum of the standard value of the solution of the gesture action data violation behavior inspection function and the standard value of the solution of the digital human abnormal behavior video function, divided by the standard value of the gesture action data classified for the customer service behavior video.

9. A digital human-assisted service method based on a video customer service application according to claim 2, characterized in that, After sending the to-be-transmitted gesture action data to the customer service behavior video control platform, it further includes: analyzing the behavior relevance, the specified set value of the behavior relevance, the inspection set position, the solution of the gesture action data violation behavior inspection function, and the solution of the digital human abnormal behavior video function from the received gesture action data, implementing the confirmatory factor analysis algorithm corresponding to the solution at the inspection set position to obtain the gesture action data classified for the customer service behavior video before inspection; Among them, the confirmatory factor analysis algorithm corresponding to the solution at the inspection set position and the confirmatory factor analysis algorithm corresponding to the inspection set position are two-way operations.

10. A digital human-assisted service method based on a video customer service application according to claim 9, characterized in that, The customer service behavior video control platform uses a 5G wireless network for gesture action data transmission.

Images