Method and system for sharing big data of middle channels in smart campus

Abstract

The invention relates to the technical field of data transmission, in particular to a smart campus middle station big data sharing method, which comprises the following steps of: acquiring facial images of students and parents in real time, screening out related data according to historical facial data in a storage unit, marking suspicious frame images in the related data, training and generating simulation data, and displaying the simulation data. And the restoration model restores the facial expression according to the simulation data and displays the facial expression, so that the limitation and dependency of big data in propagation caused by bandwidth and network problems can be avoided, more key facial information is reserved, the convergence speed during training of the big data model is ensured, and the communication efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission, and particularly relates to a method and system for sharing big data in the intelligent campus middle platform. Background Art

[0002] In existing intelligent campus platforms, most aspects of home-school communication adopt methods such as APPs, communication communities, or telephone communication. However, the above methods have the disadvantages of information fragmentation and information existing only for a certain period of time. Moreover, most parents rarely notice the information or emails. Traditional telephone communication can only transmit language and cannot express expressions. People's expressions can convey a large amount of information during communication, resulting in low communication efficiency. In response to the above problems, some platforms collect parents' facial information through virtual technology and display it. Moreover, in order to achieve a better communication effect, all data is often transmitted in real time. However, when all image data is transmitted in real time, it will consume a large amount of network resources and cause congestion. Moreover, if there is a network fluctuation problem, it will cause video stuttering and thus affect communication efficiency. A Chinese invention named an intelligent campus management system for big data, published on November 21, 2023, with the publication number CN116433432B, obtains various characteristic data through different sensors, compares and screens the important characteristics with the characteristic data stored in the historical database, and finally completes the display of important expressions through the restoration imaging module, which can reduce the resource occupancy rate of real-time transmission and greatly reduce the degree of influence by network fluctuations. Although the above method can enhance the anti-interference ability during transmission, in the face of multiple frames of images, the above screening and reproduction method will lead to discontinuous reproduction results and affect the communication experience between parents and teachers. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for sharing big data in the intelligent campus middle platform to solve one or more technical problems existing in the prior art and at least provide a beneficial choice or create conditions.

[0004] To achieve the above purpose, according to one aspect of the present invention, a method for sharing big data in the intelligent campus middle platform is provided. The method for sharing big data in the intelligent campus middle platform includes the following steps:

[0005] S100, collecting the facial images of parents as facial feature data;

[0006] S200, generating training data according to the collected facial feature data;

[0007] S300, generating facial simulation data according to the training data;

[0008] S400, restoring the current facial expression of the parent according to the facial simulation data.

[0009] Further, in S300, the face simulation data is generated by the restoration model according to the training data.

[0010] Further, the restoration model is any one of FaceNet, VGG-Face, ArcFace, and MTCNN.

[0011] Further, in S400, the current parent's facial expression is restored by the imaging model according to the face simulation data.

[0012] Further, the imaging model is the StyleGAN model.

[0013] Further, in S100, the specific method for collecting the parent's facial image as the facial feature data is as follows: The facial feature images of the parent are collected according to a preset cycle and used as the facial feature data of the parent to which the collected parent belongs. Let i be the serial number of the collection cycle, and B i represents the facial feature data collected in the i-th collection cycle, and all B i constitute the set B.

[0014] In real life, facial expressions usually change dynamically over time, that is, the current expression is affected by the dynamic sequence context in which it is located. Moreover, the expression shown when a parent communicates with a teacher on the phone is also affected by the expression at the previous moment. The previous moment provides a scenario basis for the current expression to generate a filter, and the current expression is constructed according to the emotional characteristics of the formed filter, so that the expression at the current moment shifts in the same direction as the expression at the previous moment, that is, there is a certain correlation between expressions. In a Chinese invention named A Smart Campus Management System of Big Data with the publication number CN116433432B and published on November 21, 2023, the current various feature data of the parent are continuously collected by the device, and the feature data more similar to those in the historical storage unit are obtained after screening and the current expression is restored through the model. Although it can accelerate the training speed of the model and can preferentially transmit important expressions in a network fluctuation environment to ensure smooth communication, the above method only uses the current image as the basis for evaluating the current parent's emotion, does not consider the correlation between expressions, and cannot capture the weak feature data in different emotion conversion periods in time, so a lot of feature data that can represent emotion conversion is lost, resulting in incomplete model training and thus the model cannot accurately reconstruct the parent's facial image. Therefore, the following method of the present application is needed to obtain complete training data.

[0015] Further, in S200, the method for generating training data according to the collected facial feature data includes the following steps:

[0016] S210, generating a representative subscale according to the obtained facial feature data;

[0017] S220. Calculate the correlation coefficient between the facial feature data in the representative component table;

[0018] S230. Mark the training data according to the correlation coefficient and store it in the database.

[0019] Furthermore, in S210, the specific method for generating the representative component table based on the obtained facial feature data is as follows: within the range of the value of i, each B i is calculated for the matching degree with the pre-stored facial feature data of the collected parents. Mark all facial feature data greater than the average matching degree as relevant data, and the others as irrelevant data. Permanently store all relevant data in the storage unit, and mark the newly added data in the storage unit as incremental data, and the others as original data. At the same time, use the relevant data and irrelevant data as the representative component table.

[0020] Furthermore, the matching degree calculation algorithm is any one of Meanshift and Camshift.

[0021] Furthermore, in S220, the method for calculating the correlation coefficient between the facial feature data in the representative component table is as follows: in the representative component table, calculate the ratio of the number of relevant data in the same period to the total number of all data in the set composed of the facial feature data in the same period as the performance probability P of the facial feature data in the current period j , where j represents the serial number of the acquisition period. The performance probabilities P of the facial feature data in all periods j constitute the set P;

[0022] All acquisition periods constitute the set H. Calculate the dispersion coefficient of the facial feature data of the component table. The calculation method is as follows: expand the representative component table to generate a component table. The specific method is: after affine transformation of the data in the current representative component table, add it to the original representative component table to generate a component table. Take the ratio of the number of incremental data in the storage unit to the number of original data as the reference coefficient;

[0023] Sum and average the absolute differences between the performance probabilities and the mean value of all periods in the set H as the dispersion coefficient of the current component table. If the dispersion coefficient is greater than the reference coefficient, remove a period with the largest absolute value of the difference between the performance probability and the reference probability from the set H, and continue to calculate the dispersion coefficient of the performance probabilities of the remaining periods in the set H. Compare the dispersion coefficient with the reference coefficient again, and repeat the above operations until the dispersion coefficient is less than the reference coefficient. Then, sum and average the performance probabilities of the remaining periods in the set H as the dispersion standard HT;

[0024] Calculate the correlation coefficient according to the discrete standard of facial feature data in set H. The calculation method is as follows: Screen out all cycles in set H with a discrete coefficient greater than HT and mark them as important cycles. All important cycles form an important set E. Denote the pre-stored parental facial feature data as historical data. For each cycle in set E, denote the image with the lowest matching degree with the historical data as the valley image, and the image with the highest matching degree with the historical data as the peak image;

[0025] Starting from the first cycle of set E, perform feature point detection on the region of interest of the valley image to obtain multiple feature points; All the collected feature points Sr form a starting point set Start in the order of detection time. Let r represent the serial number of the feature point, and Sr represent the r-th feature point, where r ∈ [1, N], and N represents the number of feature points;

[0026] Perform feature point detection on the region of interest of the peak image to obtain multiple feature points; All the collected feature points Pr form a peak point set Peak in the order of detection time, and Pr represents the r-th feature point;

[0027] Establish a reference coordinate system with the geometric center of the valley image as the origin. One-to-one correspondence between the points in the starting set and the points in the peak set generates direction vectors as reference vectors to form a reference set, and update the reference set according to the detection time order of the feature points in the set Start;

[0028] Successively obtain the feature points of the regions of interest of all the images (excluding the valley image and the peak image) collected corresponding to each cycle in set E, and mark all the feature points as 01; Generate a feature sequence for each of the feature points detected in each frame of the image in the order of detection time. Starting from the first feature sequence, successively form direction vectors between the image feature points collected and the corresponding feature points in the peak image as change vectors, and then calculate the cosine value of the angle between the change vector and the corresponding reference vector in the reference set. If the cosine value is less than zero, modify the label of the current feature point to 00; Complete the labeling of all feature points in sequence according to the above method.

[0029] The reference vector can represent the change trend of feature points in the region of interest in the valley image and the peak image. The valley image represents the starting point of the expression in the current cycle, and the peak image represents the ending point of the expression in the current cycle. The change vector can represent the change trend of feature points in the region of interest in the current image and the peak image in the current cycle. The cosine value of the angle formed between the change vector and the reference vector can reflect the direction of the change trend. If the cosine value is less than zero, it proves that the current feature point has a large difference from the overall change trend of the expression in the current cycle, and the image where the current feature point is located can be considered as a transitional image that can represent the expression change in the current cycle. If the cosine value is greater than zero, it proves that the current feature point has a small difference from the overall change trend of the expression in the current cycle, and the image where the current feature point is located can be considered as an ordinary image in the current cycle.

[0030] Further, the shape_predictor function in the DLIB open-source library is used for feature point detection.

[0031] Further, using the face detection technology provided by DLIB, the nose region, mouth region, left eyebrow and eye region, and right eyebrow and eye region are marked as regions of interest.

[0032] Non-zero determination is made on the feature points in all the marked images in sequence (that is, it is judged in sequence whether the marks of all the feature points in the image frame are 01), and the determination results are subjected to an AND operation. If the result is true (that is, if they are all 01), the current image is marked as an ordinary image. If the result is false, the current image is marked as a suspicious image.

[0033] Further, all the cycles containing suspicious images are marked as suspicious cycles Sp k , k represents the serial number of the suspicious cycle, and the cycle Sp k The number of images in is denoted as Num. According to the formula CL = P j ×Num α Calculate the correlation coefficient of Sp k , where α is a degree factor, and the calculation formula is where the function exp() is an exponential function with the natural constant e as the base, Nnum is the total number of images in all suspicious cycles, and the suspicious cycles with correlation coefficients greater than the average correlation coefficient are marked as associated cycles;

[0034] Further, the images included in the associated cycles are marked as training data and saved in the database.

[0035] The beneficial effects are as follows: The above method can screen out all images with more concentrated features, thereby reducing the errors introduced by other redundant images. The low-valley image and the peak image represent the most neutral and most intense degrees of the current expression (a neutral expression means a shallow expression degree of the current image, and an intense expression means a deep expression degree of the current image). The reference vector is a direction vector formed by using the feature points in the low-valley image as the starting point and the corresponding feature points in the peak image as the ending point, which can better illustrate the overall trend of the muscles in the current facial key area. When facing multiple frames of images, frame-by-frame detection can discover more subtle expression changes. By calculating the correlation coefficient between images, the suspicious image frames of expression changes can be accurately captured. The suspicious image frames can show the transitional state of expression transformation, which can make the output result of the restoration model smoother. At the same time, screening data by expression probability can remove the redundant data collected, ensuring the superiority of the model in training speed. The sub-scale and discrete standard can reduce the number of training data while ensuring the quality of training data, reducing the resource occupancy rate when transmitting data, solving the problem that some servers are extremely vulnerable to network fluctuations due to insufficient bandwidth, and achieving the stability of replicating expressions under network fluctuations, enhancing the communication experience between parents and teachers.

[0036] However, current research shows that the initially presented facial expression can affect an individual's perception of the subsequently presented facial expression through adaptation aftereffects. After a long exposure to a specific emotional face, the habituation of neural responses causes the individual's behavioral response to that emotional face to be inhibited, and then leads to the perception of the subsequent emotional face shifting towards the opposite emotional characteristic direction. For example, when teachers and parents have a long conversation and there is a long-lasting happy expression, when the teacher stares at the happy face for a long time, the characteristics of the happy emotion will become less positive, resulting in the current expression shifting towards a negative and neutral direction, leading to a negative transformation of the presented expression (simply put, when staring at a happy expression for a long time, the feedback is an unhappy expression). In model training, these expressions with negative transformation phenomena will cause great interference to the training results of the model, thus greatly reducing the accuracy of the replication model. To solve the above problems, the present invention proposes the following method to solve the transmission interference caused by the negative transformation phenomenon by marking out the empathy images.

[0037] In all correlation cycles, mark the first suspicious image as the starting frame image, and perform feature point detection on the region of interest of the starting frame image to obtain multiple feature points; form the starting point set TP with all the feature points Tp collected on the current image in the order of detection time, and use q to represent the serial number of the feature point, where Tp q represents the q-th feature point;

[0038] In the associated period where the starting frame image is located, retrieve backward the image with the highest matching degree with historical data as the local peak image, and perform feature point detection on the region of interest of the local peak image to obtain multiple feature points; all the feature points Hp collected on the current image are formed into a starting point set HP in the order of detection time, and q represents the serial number of the feature point, Hp q represents the q-th feature point;

[0039] Taking the geometric center of the starting frame image as the origin, establish a reference coordinate system, generate direction vectors corresponding one by one to the feature points in set HP from set TP as local reference vectors to form a local reference set, and update the local reference set according to the detection time order of the feature points in set TP;

[0040] Starting from the image after the starting frame image, perform feature point detection on the region of interest of the images in the current associated period in turn to obtain multiple feature points in the images, and mark all the feature points as 01; generate a feature sequence for the feature points detected in each frame of image according to the detection time order, starting from the first feature sequence, successively form direction vectors between the collected feature points and the corresponding feature points collected in the peak image as local change vectors, and then calculate the cosine value of the angle formed between the local change vector and the corresponding local reference vector in the local reference set. If the cosine value is greater than zero, keep the feature point mark unchanged, otherwise modify the mark of the current feature point to 00; complete the marking of all feature points in turn according to the above method.

[0041] Since the negative conversion phenomenon is manifested as a sudden transformation of the current expression, calculating the cosine value of the angle formed between the local change vector and the local reference vector of the current image in the current period can indicate whether there is a negative conversion in the current expression. If the cosine value is greater than zero, it means that there is no negative conversion phenomenon in the current image. If the cosine value is less than zero, it is considered that there is a negative conversion phenomenon in the current image. Since the suspicious image is marked as the starting frame image, the interference of the suspicious frame image on the calculation result is eliminated, so that the image with negative conversion can be accurately captured.

[0042] Make a non-zero judgment on the marked feature points in the images included in all associated periods, and then perform an AND operation on the judgment results. If the operation result is true, mark all the images between the starting frame image and the local peak image (excluding the starting frame image and the local peak image) in the associated period as recession images and form a recession sequence List with all the current recession images; if the operation result is false, keep the original mark unchanged, continue to mark the first suspicious image in the next associated period as the starting image, and repeat the above operations until all the images in the associated period are marked;

[0043] During the associated period, obtain the number of images included in the first regression sequence. Then, retrieve the same number of images forward from the first frame image of the current regression sequence to form a replacement sequence. Replace all the images in the current regression sequence with the images in the replacement sequence in turn. According to the above method, complete the replacement of all regression sequences and update the associated period.

[0044] The beneficial effects are as follows: Since regression images are the negative conversion phenomena caused by maintaining a certain expression for a long time, replacing them with adjacent images can eliminate the transmission interference and data errors caused by the negative conversion phenomena during the training of the model, improve the accuracy of the model in replicating expressions. Since no new processing is performed on the regression images, there is no redundant occupation of resources, ensuring the transmission speed and quality under network fluctuations, being able to accurately and quickly replicate the current real expression of the parent, avoiding communication misunderstandings caused by negative conversion phenomena, and enhancing the communication experience between the parent and the school.

[0045] Further, in S300, the method for generating facial simulation data based on training data is: input the facial data stored in the database into the restoration model to obtain the pre-simulation data of the facial features of the current parent.

[0046] Further, in S400, the method for restoring the current parent's facial expression based on the facial simulation data is: after obtaining the pre-simulation data of the facial features of the current parent from the database, generate a virtual avatar of a specific gender based on the gender of the parent, and restore the current facial expression of the parent according to the pre-simulation data of the facial features of the current parent, and display it on the face of the virtual avatar corresponding to the parent.

[0047] The present invention also provides a big data sharing system for the intelligent campus middle platform. The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program and runs in the following system units:

[0048] The data acquisition unit is used to obtain the facial feature data of the parent;

[0049] The data screening unit is used to screen out the training data from the facial feature images;

[0050] The data storage unit is used to store the historical data and the screened training data

[0051] The model training unit is used to train the data in the storage unit to obtain facial simulation information;

[0052] The facial restoration unit is used to restore the current parent's facial expression according to the facial simulation information.

[0053] Advantages of the present invention: By collecting facial image datasets in real time, screening relevant data from the datasets, and detecting the moving directions and ranges of feature points in the regions of interest of all relevant data frame by frame, the present invention can sensitively capture the subtle facial expression changes caused by emotional transformation during communication, avoiding the image freezing problem caused by real-time transmission of all data during network fluctuations, and at the same time being able to retain more emotional information during transmission to improve communication efficiency. In the training of the big data model, due to retaining more details, the generalization of the model is enhanced on the basis of accelerating the model training speed. Description of the Drawings

[0054] Figure 1 Shown is a flowchart of a big data sharing method for a smart campus middle platform;

[0055] Figure 2 Shown is a flowchart of a big data sharing system for a smart campus middle platform. Detailed Embodiments

[0056] The following details the embodiments of the present invention. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0057] Embodiment 1

[0058] Figure 1 Shown is a flowchart of a big data sharing method for a smart campus middle platform.

[0059] Referring to Figure 1 , the present invention proposes a big data sharing method for a smart campus middle platform, and the method includes the following steps:

[0060] S100, collecting the facial images of parents as facial feature data;

[0061] S200, generating training data according to the collected facial feature data;

[0062] S300, generating facial simulation data according to the training data;

[0063] S400, restoring the current facial expressions of parents according to the facial simulation data.

[0064] Further, in S300, facial simulation data is generated according to the training data through a restoration model.

[0065] Further, the restoration model is any one of FaceNet, VGG-Face, ArcFace, and MTCNN.

[0066] Further, the restoration model is FaceNet.

[0067] Further, the imaging model restores the current parent's facial expression according to the facial simulation data. Further, the imaging model is the StyleGAN model.

[0068] Further, in S100, the specific method of collecting the parent's facial image as facial feature data is:

[0069] During the call, the facial feature images of the parent are collected at a collection period of 1 minute and used as the facial feature data of the parent being collected. Let i be the serial number of the collection period, and B i represents the facial feature data collected in the i-th collection period, and all B i constitute the set B

[0070] Further, in S200, generating training data according to the collected facial feature data includes the following steps:

[0071] S210, generating a representative component table according to the obtained facial feature data;

[0072] S220, calculating the correlation coefficient between the facial feature data in the representative component table;

[0073] S230, marking the training data according to the correlation coefficient and storing it in the database.

[0074] Further, in S210, the specific method of generating a representative component table according to the obtained facial feature data is: within the range of the value of i, calculate the matching degree between each B i and the facial feature data of the parent being collected stored in advance. Mark all facial feature data greater than the average matching degree as relevant data, and the others as irrelevant data. Permanently store all relevant data in the storage unit, and mark the newly added data as incremental data in the storage unit, and the others as original data. At the same time, use the relevant data and irrelevant data as the representative component table.

[0075] Further, the matching degree calculation algorithm is the Meanshift algorithm.

[0076] Further, in S220, the method of calculating the correlation coefficient between the facial feature data in the representative component table is: in the representative component table, calculate the ratio of the number of relevant data in the same period to the total number of all data in the set composed of the facial feature data in the same period as the performance probability P j of the facial feature data in the current period. j represents the serial number of the collection period. The performance probabilities P jForm the set P;

[0077] Form the set H with all the acquisition cycles, calculate the coefficient of variation of the facial feature data of the sub-scale. The calculation method is as follows: expand the representative sub-scale to generate a new sub-scale. The specific method is: after affine transformation of the data in the current representative sub-scale, add it to the original representative sub-scale to generate a new sub-scale, and take the ratio between the number of incremental data in the storage unit and the number of original data as the reference coefficient;

[0078] Sum and average the absolute differences between the performance probabilities and the mean values of all cycles in the set H as the coefficient of variation of the current sub-scale. If the coefficient of variation is greater than the reference coefficient, remove a cycle with the largest absolute value of the difference between the performance probability and the reference probability from the set H, and continue to calculate the coefficient of variation of the performance probabilities of the remaining cycles in the set H. Compare the coefficient of variation with the reference coefficient again, and repeat the above operations until the coefficient of variation is less than the reference coefficient. Then sum and average the performance probabilities of the remaining cycles in the set H as the discrete standard HT;

[0079] Calculate the correlation coefficient according to the discrete standard of the facial feature data in the set H. The calculation method is as follows: screen out all cycles with a coefficient of variation greater than HT in the set H and mark them as important cycles. Form the important set E with all the important cycles. Denote the pre-stored parental facial feature data as historical data. For each cycle in the set E, denote the image with the lowest matching degree with the historical data as the valley image, and denote the image with the highest matching degree with the historical data as the peak image;

[0080] Starting from the first cycle of the set E, perform feature point detection on the region of interest of the valley image to obtain multiple feature points; form the starting point set Start with all the collected feature points Sr in the order of detection time. Let r represent the serial number of the feature point, and Sr represent the r-th feature point, where r ∈ [1, N], and N represents the number of feature points;

[0081] Perform feature point detection on the region of interest of the peak image to obtain multiple feature points; form the peak point set Peak with all the collected feature points Pr in the order of detection time. Pr represents the r-th feature point;

[0082] Establish a reference coordinate system with the geometric center of the valley image as the origin, generate direction vectors by corresponding the points in the starting set with the points in the peak set one by one as reference vectors to form a reference set, and update the reference set according to the detection time order of the feature points in the set Start;

[0083] Obtain the feature points of the regions of interest of all the images (excluding the valley images and peak images) collected corresponding to each period in set E in sequence, and mark all the feature points as 01; generate a feature sequence for each of the feature points detected in each frame of image according to the detection time sequence. Starting from the first feature sequence, form a direction vector by taking the feature points of the collected images and the corresponding feature points in the peak image in sequence as the change vector, and then calculate the cosine value of the angle between the change vector and the corresponding reference vector in the reference set. If the cosine value is less than zero, modify the label of the current feature point to 00; complete the labeling of all the feature points according to the above method.

[0084] Further, use the shape_predictor function in the DLIB open source library for feature point detection.

[0085] Further, adopt the face detection technology provided by DLIB, and mark the nose region, mouth region, left eyebrow and eye region, and right eyebrow and eye region as regions of interest.

[0086] Make a non-zero determination on the feature points in all the labeled images in sequence (that is, determine whether the labels of all the feature points in the image frame are 01 in sequence), and perform an AND operation on the determination results. If the result is true (that is, if they are all 01), mark the current image as a normal image. If the result is false, mark the current image as a suspicious image.

[0087] Further, mark all the periods containing suspicious images as suspicious periods Sp k , where k represents the serial number of the suspicious period, and denote the number of images in period Sp k as Num. Calculate the correlation coefficient of Sp j according to the formula CL = P α × Num k , where α is the degree factor, and the calculation formula is where the function exp() is the exponential function with the natural constant e as the base, Nnum is the total number of images in all the suspicious periods. Mark the suspicious periods with correlation coefficients greater than the average correlation coefficient as associated periods;

[0088] Further, mark the images included in the associated periods as training data and save them in the database.

[0089] Further, in S300, the method for generating facial simulation data according to the training data is: input the facial data stored in the database into the restoration model to obtain the facial simulation data of the facial features belonging to the current parent.

[0090] Further, in S400, the method for restoring the current parent's facial expression according to the facial simulation data is as follows: After obtaining the facial simulation data of the facial features to which the current parent belongs from the database, a virtual avatar of a specific gender is generated based on the gender of the parent, and the current facial expression of the parent is restored according to the pre-simulation data of the facial features to which the current parent belongs, and is displayed on the face of the virtual avatar corresponding to the parent.

[0091] Embodiment 2

[0092] The present invention also proposes the following method to solve the transmission interference caused by the negative conversion phenomenon by marking the empathy image:

[0093] In all association cycles, mark the first suspicious image as the starting frame image, and perform feature point detection on the region of interest of the starting frame image to obtain multiple feature points; all the feature points Tp collected on the current image are formed into a starting point set TP in the order of detection time, and q represents the serial number of the feature point, and Tp q represents the q-th feature point;

[0094] In the association cycle where the starting frame image is located, retrieve backward to find the image with the highest matching degree with the historical data as the local peak image, and perform feature point detection on the region of interest of the local peak image to obtain multiple feature points; all the feature points Hp collected on the current image are formed into a starting point set HP in the order of detection time, and q represents the serial number of the feature point, and Hp q represents the q-th feature point;

[0095] Taking the geometric center of the starting frame image as the origin, establish a reference coordinate system, and generate a direction vector by corresponding the feature points in the set TP with the feature points in the set HP one by one as the local reference vector to form a local reference set, and update the local reference set according to the detection time order of the feature points in the set TP;

[0096] Starting from the frame image after the starting frame image, perform feature point detection on the region of interest of the images in the current association cycle in turn to obtain multiple feature points in the image, and mark all the feature points as 01; generate a feature sequence for each of the feature points detected in each frame image according to the detection time order, starting from the first feature sequence, successively form a direction vector between the collected feature points and the corresponding feature points collected in the peak image as the local change vector, and then calculate the cosine value of the angle formed between the local change vector and the corresponding local reference vector in the local reference set. If the cosine value is greater than zero, keep the feature point mark unchanged, otherwise modify the current feature point mark to 00; complete the marking of all feature points in turn according to the above method.

[0097] Make a non - zero judgment on the feature points marked in the images included in all associated cycles, and then perform an AND operation on the judgment results. If the operation result is true, mark all the images (excluding the starting - frame image and the local - peak image) between the starting - frame image and the local - peak image in the associated cycle as fade - sense images, and form a fade - sense sequence List with all the current fade - sense images; if the operation result is false, keep the original marks unchanged, continue to mark the first suspicious image in the next associated cycle as the starting image, and repeat the above operations until all the images in the associated cycle are marked;

[0098] In the associated cycle, obtain the number of images included in the first fade - sense sequence, and then retrieve the same number of images forward from the first - frame image of the current fade - sense sequence to form a replacement sequence. Replace all the images in the current fade - sense sequence with the images in the replacement sequence in turn. According to the above method, complete the replacement of all fade - sense sequences and update the associated cycle.

[0099] In addition, the present invention also provides an embodiment of a big - data sharing system for the intelligent campus middle - platform, as Figure 2 shown in the structure diagram of a big - data sharing system for the intelligent campus middle - platform of the present invention. An embodiment of a big - data sharing system for the intelligent campus middle - platform includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above - mentioned embodiment of a big - data sharing system for the intelligent campus middle - platform.

[0100] The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it runs in the following units of the system:

[0101] A data - acquisition unit, used to obtain the facial feature data of parents;

[0102] A data - screening unit, used to screen out training data from facial feature images;

[0103] A data - storage unit, used to store historical data and the screened - out training data

[0104] A model - training unit, used to train the data in the storage unit to obtain facial simulation information;

[0105] A facial - restoration unit, used to restore the current parent's facial expression according to the facial simulation information.

[0106] The described big data sharing system for the intelligent campus middle platform can run on computing devices such as desktop computers, laptops, palmtop computers, and cloud servers. The described big data sharing system for the intelligent campus middle platform, the system that can run may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above examples are only examples of a big data sharing system for the intelligent campus middle platform, and do not constitute a limitation on a big data sharing system for the intelligent campus middle platform. It may include more or fewer components than the examples, or combine some components, or different components. For example, a big data sharing system for the intelligent campus middle platform may also include input / output devices, network access devices, buses, etc.

[0107] The so-called processor can be a central processing unit (CPU), or it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the operating system of the described big data sharing system for the intelligent campus middle platform, and uses various interfaces and lines to connect all parts of the operating system of the described big data sharing system for the intelligent campus middle platform.

[0108] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the described big data sharing system for the intelligent campus middle platform by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, memory, plug-in hard disks, smart media cards (SMCs), secure digital (SD) cards, flash cards, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0109] Although the description of the present invention has been quite detailed and has particularly described several of the described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, so as to effectively cover the intended scope of the present invention. In addition, the present invention is described above in terms of embodiments foreseeable by the inventors for the purpose of providing a useful description, and non-substantive changes to the present invention that are not currently foreseeable may still represent equivalent changes to the present invention.

Claims

1. A method for sharing big data in a smart campus, characterized in that: The method comprises the following steps: S100, collecting a parent's facial image as facial feature data; S200, generating training data according to the collected facial feature data; S300, generating facial simulation data according to the training data; S400, restoring the current parent's facial expression according to the facial simulation data.

2. According to claim 1, a method for sharing big data in a smart campus platform is characterized in that: In S100, the method of collecting the facial image of the parent as the facial feature data is as follows: collecting the facial feature image of the parent according to a preset period and using it as the facial feature data of the collected parent, where i is the serial number of the collection period, and B is the serial number of the collection period. i represents the facial feature data collected in the i-th collection cycle, consisting of all B i Constitute the set B.

3. According to claim 1, a method for sharing big data in a smart campus platform is characterized in that: In S200, the method for generating training data according to the collected facial feature data includes the following steps: S210, generating a representative component table according to the acquired facial feature data; S220, calculating the correlation coefficient between the facial feature data in the representative subscale; S230, marking the training data according to the correlation coefficient and storing it in a database.

4. According to claim 1, a method for sharing big data in a smart campus platform is characterized in that: In S300, the method for generating facial simulation data according to the training data is: The facial data stored in the database is input into the restoration model to obtain facial simulation data of the facial features of the current parent.

5. According to claim 1, a method for sharing big data in a smart campus platform is characterized in that: In S400, the method for restoring the current parent's facial expression according to the facial simulation data is: After obtaining the facial simulation data of the current facial features of the parent from the database, a virtual avatar of a specific gender is generated based on the parent's gender, and the parent's current facial expression is restored based on the pre-simulated data of the current facial features of the parent, and displayed on the face of the virtual avatar corresponding to the parent.

6. A smart campus platform big data sharing system, characterized in that: The smart campus center big data sharing system comprises: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the smart campus center big data sharing method described in any one of claims 1-5 are implemented.

Images