Classroom concentration evaluation method and system based on group sight line estimation

Abstract

The invention relates to a classroom concentration degree evaluation method and system based on group sight line estimation. The method comprises the following steps: S1, collecting a student classroom behavior video; S2, detecting the head of a classroom student group; S3, estimating the sight line based on head region features; according to the method, the YoLov8 model is improved, so that the head detection efficiency of a classroom intensive student group is improved; the sight line estimation model based on the head region features has good robustness on the definition of the actually detected head region image, and then the sight line angle is determined and the sight line estimation noise is removed, so that the result is more stable. According to the student group attention area estimation algorithm, sight line attention areas of students are projected to a space interface through sight line estimation angles, so that space calculation is converted into plane calculation. According to the classroom concentration degree evaluation method, a piecewise function is adopted, and student group concentration degree evaluation in the classroom dynamic teaching process is achieved.

Description

Technical Field

[0001] The present invention relates to the field of educational information processing, and in particular to a classroom concentration assessment method and system based on group sight line estimation. Background Art

[0002] In recent years, with the rapid development of artificial intelligence (AI) technology, its application in teaching has received widespread attention. Visual monitoring systems have become standard equipment in today's university classrooms, but they are primarily used for subjective monitoring of the teaching process, resulting in low utilization of the information captured through data mining. Deeply mining the information gained during classroom teaching has become a pressing issue at the intersection of AI and education.

[0003] Classroom focus is closely related to student behavior and can, to a certain extent, reflect the student learning state. It is a complex process that involves the interaction between individual student factors and those of the classroom learning environment. Currently, researchers primarily use computer vision to detect and identify student behavior during classroom learning using key points of the human skeleton. This allows for analysis of students' learning status and focus based on their specific behaviors. These methods often assume a single teaching mode and statically define student behavior to detect and count students' learning behavior. This makes them difficult to apply to the dynamic teaching environment of a real-world classroom. Furthermore, since most students are located on the same horizontal plane, significant occlusions between them make it difficult to obtain body language information. This limits the scope of these methods for identifying student behavior in the classroom, making their results difficult to interpret in real-world contexts. Other researchers have proposed estimating student focus by tracking eye movements. This method is more suitable for online learning scenarios, where the camera is close to the person's face. Summary of the Invention

[0004] Based on this, the present invention proposes a classroom concentration assessment method and system based on group gaze estimation. By analyzing the classroom concentration of student groups through gaze estimation, the existing vision-based classroom behavior analysis methods solve the problems of setting a single mode for the classroom teaching process, statically processing student classroom behavior, and extremely limited coverage of behavior recognition, and provide objective support for the evaluation and improvement of the classroom teaching process. The classroom learning concentration assessment method based on group gaze estimation proposed in the present invention provides a feasible solution for the deep integration of artificial intelligence and education to solve the objective evaluation of the classroom teaching process.

[0005] In order to solve the above problems, the present invention aims to provide a method and system for evaluating classroom concentration based on group gaze estimation, comprising the following steps:

[0006] S1: Collect students’ classroom behavior videos: extract images from the video frame by frame;

[0007] S2: Head detection of students in the classroom: Perform head detection on the extracted image to obtain the student head area image;

[0008] S3: Gaze estimation based on head region features: The student's head region images are uniformly scaled to the same pixel size. The gaze estimation model based on head region features is used to obtain predicted facial orientation information, including vertical and horizontal facial angles. The vertical and horizontal facial angles are then smoothed.

[0009] S4: Student focus area estimation: A calibration plate is used to map classroom surveillance images to the real space. During the modeling process, the spatial transformation matrix is ​​recorded according to the classroom number to estimate the classroom spatial position coordinates of the student head. The focus area estimation algorithm can be used to obtain the student group's focus target area.

[0010] S5: Classroom concentration assessment: The obtained group attention target area map is binarized and marked, and the number of attentions in each marked area is counted; a piecewise function is used to realize the assessment of student group concentration in the dynamic classroom teaching process.

[0011] Preferably, the head detection of the classroom student group in step S2 adopts a two-stage feature fusion attention module to replace the Concat module of YOLOv8. The main process of the two-stage feature fusion attention module includes the following steps:

[0012] S21: Input: The image is input into the feature extraction network, and two different feature representations are obtained through different mapping layers. They are used as the two inputs of the Twfa module respectively. The two feature representations are then spliced ​​together to obtain a feature map, which is input into the subsequent SEAttention channel attention mechanism.

[0013] S22: SEAttention channel attention mechanism is used to fuse features at different levels, mainly including the following operations in sequence: Avgpool, Linner, Relu, Linner, Sigmoid, to obtain feature weight values;

[0014] S23: Feature operation: multiply the feature weight value and the original input, cross-add them, and finally concatenate them together to obtain the output feature.

[0015] Preferably, the specific steps of obtaining the feature weight value in step S22 are:

[0016] The feature map is processed by Avgpool (average pooling) with a step size of 2 to obtain 1*1*N features; after Linear processing, the feature map is reduced to N / R times to obtain 1*1*N / R features, and after Relu and Linear processing, the feature map size is restored to 1*1*N. Finally, the Sigmoid activation function is used to limit the output range to between 0 and 1, where N is the number of input feature channels and R is the compression ratio.

[0017] Preferably, the structure of the sight line estimation model of the head region feature in step S3 is:

[0018] Initial feature extraction is performed using two convolution operations with a window size of 3×3 and a stride of 1, followed by a maximum pooling operation with a window size of 2×2 and a stride of 2. Next, the initial features are passed through two convolution operations with upper and lower windows of 3×3 and a stride of 1, separating the two input features into two input features that are fed into Twfa. Three feature optimization operations are performed according to this structure. Finally, the optimized feature map is passed through ResNet34 for facial orientation estimation. The output Pitch is the vertical angle, and Roll is the horizontal angle.

[0019] Preferably, the smoothing process in step S3 is to smooth out the facial orientation deviation noise that occasionally occurs during students' classroom activities. The smoothing process algorithm is:

[0020]

[0021] in and Represents the horizontal Roll direction α h and vertical Pitch direction α v The estimated angle at the i-th moment, and In fact, the horizontal Roll direction α in the time period from t-Δt to t h and vertical Pitch direction α v The average value of the angles.

[0022] Preferably, the region of interest estimation algorithm in step S4 is:

[0023]

[0024] Among them, α h is the angle of the face toward the horizontal, α v is the face angle toward the vertical, x r is the distance coordinate from the center of the head to the classroom window wall, y r is the distance from the center of the head to the blackboard wall, z rrepresents the height from the center of the head to the ground; β represents the visual divergence angle of the human eye when it is focused; x d Indicates the horizontal coordinate range of the sight area, z d Indicates the vertical coordinate range of the sight focus area.

[0025] Preferably, the calculation method of the classroom concentration evaluation in step S5 is as follows:

[0026]

[0027] Among them, the number of attentions of the i-th attention area is h i , the total number of students in the group is n, m represents the total number of marked areas, μ h Represents all areas of interest h i The average value, p t Represents the estimated value of classroom concentration at time t.

[0028] The technical solution of the embodiment of the present invention has the following beneficial effects:

[0029] (1) A two-stage feature fusion attention (Twfa) module is proposed to improve the YoLov8 model, so that the shallow and deep feature information can be fused together through the attention mechanism, guiding the feature content to improve the direction of small targets, thereby improving the efficiency of head detection in densely populated classrooms.

[0030] (2) A gaze estimation model based on head region features is proposed to achieve gaze angle estimation based on the head region image. The model is robust to the clarity of the actual detected head region image. By using the mode of the vertical and horizontal angles within a time period to determine the gaze angle estimate, the gaze estimation noise is removed, making the result more stable.

[0031] (3) A student group attention area estimation algorithm is proposed. The five spatial interfaces of the classroom are projected onto a plane, and the student's line of sight attention area is projected onto the spatial interface through the line of sight estimation angle, so that spatial calculation is converted into plane calculation.

[0032] (4) A classroom concentration assessment method is proposed, which uses piecewise functions to achieve the assessment of student group concentration in the dynamic classroom teaching process. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of the classroom attention assessment method based on group gaze estimation;

[0034] Figure 2 It is based on the two-stage feature fusion attention (Twfa) module;

[0035] Figure 3 Comparison of head features obtained using different modules, (a) is the Concat module and (b) is the Twfa module;

[0036] Figure 4 It is a gaze estimation model based on head region features;

[0037] Figure 5 Focusing on the target area for facial orientation estimation, (a) top view, (b) side view;

[0038] Figure 6 Focus on the spatial distribution of target areas for the classroom student population;

[0039] Figure 7 Visualize student focus numbers in a classroom. DETAILED DESCRIPTION

[0040] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] The present invention proposes a method and system for evaluating classroom concentration based on group gaze estimation. The overall process is as follows: Figure 1 As shown, the method includes:

[0042] 1. Collect videos of students’ classroom behavior

[0043] The surveillance cameras in the classroom are used to collect real-time videos of students’ classroom behaviors, extract images every other frame, and input them into the system in chronological order to evaluate students’ classroom concentration.

[0044] 2. Head detection of students in the classroom

[0045] YOLOv8 is a leading object detection model. Its performance has been widely recognized by researchers and has been successfully applied in people's daily lives. However, head detection in classroom environments is affected by factors such as ambient lighting, the adhesion between various objects and the head, and background interference, which can affect YOLOv8's head detection performance.

[0046] In order to improve the detection performance of YOLOv8 for student heads in a classroom environment, this paper proposes a two-stage feature fusion attention (Twfa) module to replace the Concat module of YOLOv8. This module can input features at different positions into the attention mechanism and perform feature weight fusion respectively, so as to achieve the fusion of shallow and deep feature information through the attention mechanism, and guide the feature content to improve towards small targets. The structure diagram of the Twfa module is shown in the figure. Figure 2 As shown. The main process of the module is as follows:

[0047] (1) Input: The image is input into the feature extraction network, and two different feature representations are obtained through different mapping layers, which serve as the two inputs of the Twfa module respectively. The two feature representations are then concatenated (C) together to obtain a feature map, which is input into the subsequent SEAttention channel attention mechanism.

[0048] (2) SEAttention channel attention mechanism is used to fuse features at different levels, mainly including the following operations: Avgpool, Linner, Relu, Linner, Sigmoid. Specifically: Assume that the input feature map is H*W*N (where H represents the height of the image, W represents the width of the image, and N represents the number of input feature channels). After Avgpool (average pooling) with a step size of 2, a 1*1*N feature is obtained. After Linear (linear) processing, the feature map is reduced to N / R (where N represents the number of input feature channels and R represents the compression ratio) times to obtain a 1*1*N / R feature. After Relu and Linear, the feature map size is restored to 1*1*N. Finally, the output range is limited to between 0 and 1 through the Sigmoid activation function to obtain the feature weight value.

[0049] (3) Feature operation: Multiply the feature weight value and the original input, cross-add them, and finally concatenate (C) them together to obtain the output feature.

[0050] In order to better illustrate the improvement of the Twfa output features proposed in this invention compared with the original Concat module features of YOLOv8, Figure 3 A comparison of the features extracted by the two modules is given. From the enlarged image of the local area in the second row, it can be seen that the head features obtained using the Twfa module have better consistency.

[0051] To train the YOLOv8+Twfa model parameters, we used the SCUT-HEAD dataset for classroom head detection, released by Huazhong University of Science and Technology. Part A of this dataset contains 2,000 images from university classroom surveillance videos, and Part B contains 2,405 images of various types of classrooms, with the head regions annotated. The images were randomly divided into training, validation, and test sets in a ratio of 7:1:2.

[0052] 3. Gaze Estimation Based on Head Region Features

[0053] In classroom surveillance videos, it is often difficult to obtain eye information for face images at a certain distance. In a natural state without deliberate eye movement, the facial orientation of a face can be approximated as the person's line of sight. The human head posture has three degrees of freedom of rotation in space, namely: horizontal rotation, vertical rotation, and axial rotation. The facial orientation is mainly the rotation of the head in the horizontal and vertical directions. The present invention combines the angles of these two directions to form the facial orientation. Therefore, the present invention proposes a line of sight estimation model based on head region features. Figure 4 As shown in the figure, the head area image detected by YOLOv8+Twfa is uniformly scaled to 200×200 pixels and fed into the model for facial orientation prediction. First, preliminary feature extraction is performed through two convolution operations with a window size of 3×3 and a stride of 1, and a maximum pooling operation with a window size of 2×2 and a stride of 2. Next, the preliminary features are subjected to two convolution operations with upper and lower windows of 3×3 and a stride of 1, and two input features are separated and fed into Twfa. According to this structure, three feature optimization operations are performed. Finally, the optimized feature map is passed through ResNet34 for facial orientation estimation. The output Pitch is the vertical angle and Roll is the horizontal angle.

[0054] The model training process uses the UMDFaces and Head Pose Image datasets. The UMDFaces dataset contains 367,888 facial images of 8,277 subjects with annotated face orientations. The Head Pose Image dataset contains 5,580 facial images of 372 subjects with annotated face orientations. Samples from both datasets are randomly divided into training, validation, and test sets in a ratio of 7:1:2.

[0055] Use α h and α v In order to better estimate the sight angle and smooth the occasional facial orientation noise during classroom activities, the present invention calculates the face angle at time t. and The following smoothing formula is used:

[0056]

[0057] in, and Represents the horizontal Roll direction α h and vertical Pitch direction α v The estimated angle at the i-th moment, and In fact, the horizontal Roll direction α in the time period from t-Δt to t h and vertical Pitch direction α v The average value of the angle. In practice, Δt is taken as 1s, that is, the maximum number of times within 1 second is counted as the facial orientation at that time.

[0058] 4. Estimation of student group focus areas

[0059] Since the position and viewing angle of the surveillance camera in the classroom are fixed, the present invention uses a calibration plate to map the classroom surveillance image to the real space. During the modeling process, the spatial transformation matrix is ​​recorded according to the classroom number, so as to estimate the spatial position of the head. The blackboard, roof, floor, and two side walls of the classroom are defined as the visual limit boundaries. Let (x r ,y r ,z r ) is the coordinate of the head of the image corresponding to the actual space of the classroom, x r is the distance coordinate from the center of the head to the classroom window wall, y r is the distance from the center of the head to the blackboard wall, z r Indicates the height from the center of the head to the ground. This coordinate refers to the actual location of the head in the video, obtained through visual calibration. There are relatively mature visual calibration solutions in the existing technology.

[0060] (x d ,z d ) indicates the target area of ​​interest, x d Indicates the horizontal coordinate range of the sight area, z d Indicates the vertical coordinate range of the sight focus area. Both values ​​are ranges. β represents the visual divergence angle of the human eye when it is focused. This paper defines it as 5° through experimental measurement. Figure 5 , the algorithm for estimating the gaze attention area of ​​each student is as follows:

[0061]

[0062] Among them, α h is the angle of the face toward the horizontal, α vis the vertical angle of the face. By calculation, we can get the projection of the student group's attention target area on the visual limit boundary. Figure 6 The calculation of the target area of ​​attention of a student group in class at a certain moment is given. In order to facilitate the calculation of classroom concentration, the five boundary interfaces of the three-dimensional space of the classroom are projected onto the same plane, and the gray area is the invalid area.

[0063] 5. Classroom concentration assessment

[0064] Before evaluating classroom concentration, it is necessary to perform binary processing and regional marking on the obtained group attention target area map, and count the number of attentions in each marked area. Assume that at time t, the number of attentions received by the i-th attention area is h i , the total number of students is n, if there is h i >0.5n, then the i-th attention area is directly defined as the group attention area, and the concentration estimation value is h i Ratio to n; if any h i <0.5n, the quantitative uniformity of each focus area is calculated as the estimated value of focus, and the expression is as follows:

[0065]

[0066] Where m represents the total number of marked regions, μ h Represents all areas of interest h i The average value, p t Represents the estimated value of classroom concentration at time t.

[0067] This piecewise function can effectively evaluate classroom concentration. The physical meaning of this function is that when the teacher is teaching, the student group will form a focused attention. At this time, the attention areas of most students will overlap in a range. Therefore, the number of students overlapping in this area is the current classroom concentration. When discussing, practicing, or signing in and answering questions on mobile phones in class, the attention areas of the student group will be dispersed. At this time, it is necessary to calculate the uniformity of the distribution of the number of students in the dispersed areas.

[0068] Through the above steps, the dynamic learning concentration of student groups in the digital visualization classroom can be finally achieved.

[0069] In order to better analyze the head detection performance of the proposed YOLOv8+Twfa model, three evaluation criteria are introduced: precision, recall, and mean average precision (mAP).

[0070]

[0071] Among them, TP represents the number of samples correctly predicted as positive examples by the model, FP represents the number of samples incorrectly predicted as positive examples by the model, and FN represents the number of samples incorrectly predicted as negative examples by the model. N represents the total number of heads detected in the image, S overlap Represents the overlapping area between the predicted bounding box and the true bounding box. S union Represents the union area of ​​the predicted bounding box and the true bounding box.

[0072] Table 1 shows the head detection test comparison results of the YOLOv8+Twfa model proposed in this paper and the classic YOLOv8 model. It can be seen that the YOLOv8+Twfa model proposed in this paper has better performance in detecting the heads of students in the classroom.

[0073] Table 1 Comparison of head detection performance of student groups in classroom

[0074]

[0075] Table 2 shows the test results of the gaze estimation model based on head region features proposed in this paper on the UMDFaces and HeadPose Image datasets. The values ​​in the table are the average errors between the estimated angles in the two directions and the labeled angles in the test set.

[0076] Table 2 Facial orientation test results (mean error ± standard deviation)

[0077]

[0078] Existing related concentration assessment technologies often determine the classroom concentration of a student group by counting the proportion of students whose faces are facing forward (it is believed that the students are paying attention to the podium or blackboard at this time). This directly defines the rationality of students' head posture in the classroom from a macro perspective, and at the same time solidifies the teaching link in which only the teacher lectures in the classroom, ignoring the dynamic nature of the teaching link in the modern teaching process. These methods emphasize that students' faces should be facing forward, but in reality, facing forward does not necessarily mean focusing on the classroom content. With the dynamic changes in the teaching link, the facial orientation of the student group will change accordingly. For example, in classroom discussions or classroom exercises, most students' faces will not be facing forward, and the obtained concentration values ​​will drop sharply. It can be seen that these methods will obviously fail when dealing with modern teaching processes. The classroom concentration assessment method based on group line of sight estimation proposed in the present invention perceives the target area of ​​attention of the student group by analyzing the facial orientation of the student group, and projects it onto a plane representing the classroom space, thereby perceiving the dynamic learning concentration of the student group. Figure 7An example of digital visualization of the concentration of a group of students in a classroom according to the present invention is given, from which it can be seen that the attention of students in the classroom will fluctuate to a certain extent with time and the switching of classroom links, but the fluctuation of attention in the classroom is limited, and the situation of drastic fluctuation of student concentration when the classroom links change will not occur as in other existing methods. It can be seen that the present invention effectively solves the static problem of macro-pattern analysis of current classroom students' learning behavior. It does not rigidly stipulate the correctness of students' behavior in the classroom, but considers the situation of students' classroom concentration from the distribution of attention, and can effectively estimate the degree of student group participation in classroom learning in dynamic classroom situations. The present invention improves the data mining utilization rate of information obtained from the classroom visual monitoring system, provides objective evaluation data and basis for teachers' teaching process and students' learning process, provides objective support for the evaluation and improvement of classroom teaching process, and provides a feasible solution for the deep integration of artificial intelligence and education to solve the objective evaluation of classroom teaching process.

[0079] The main features of this embodiment are as follows:

[0080] (1) A two-stage feature fusion attention (Twfa) module is proposed to improve the YoLov8 model, so that the shallow and deep feature information can be fused together through the attention mechanism, guiding the feature content to improve the direction of small targets, thereby improving the efficiency of head detection in densely populated classrooms.

[0081] (2) A gaze estimation model based on head region features is proposed to achieve gaze angle estimation based on the head region image. The model is robust to the clarity of the actual detected head region image. By using the mode of the vertical and horizontal angles within a time period to determine the gaze angle estimate, the gaze estimation noise is removed, making the result more stable.

[0082] (3) A method for estimating the attention area of ​​a student group is proposed. The five spatial interfaces of the classroom are projected onto a plane, and the students' line of sight attention areas are projected onto the spatial interface through the line of sight estimation angle, so that spatial calculation is converted into plane calculation.

[0083] (4) A classroom concentration assessment method is proposed, which uses piecewise functions to achieve the assessment of student group concentration in the dynamic classroom teaching process.

[0084] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form or substance. It should be pointed out that ordinary technicians in this technical field can make several improvements and supplements without departing from the method of the present invention. These improvements and supplements should also be regarded as the scope of protection of the present invention. Any equivalent changes, modifications and evolutions made by technicians familiar with this profession without departing from the spirit and scope of the present invention by using the technical content disclosed above are all equivalent embodiments of the present invention; at the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method and system for evaluating classroom concentration based on group gaze estimation, characterized in that: The following steps are involved: S1: Collect students’ classroom behavior videos: extract images from the video frame by frame; S2: Head detection of students in the classroom: Perform head detection on the extracted image to obtain the student head area image; S3: Gaze estimation based on head region features: The student's head region images are uniformly scaled to the same pixel size. The gaze estimation model based on head region features is used to obtain predicted facial orientation information, including vertical and horizontal facial angles. The vertical and horizontal facial angles are then smoothed. S4: Student focus area estimation: A calibration plate is used to map classroom surveillance images to the real space. During the modeling process, the spatial transformation matrix is ​​recorded according to the classroom number to estimate the classroom spatial position coordinates of the student head. The focus area estimation algorithm can be used to obtain the student group's focus target area. S5: Classroom concentration assessment: The obtained group attention target area map is binarized and marked, and the number of attentions in each marked area is counted; a piecewise function is used to realize the assessment of student group concentration in the dynamic classroom teaching process.

2. A method and system for evaluating classroom concentration based on group gaze estimation according to claim 1, characterized in that: The head detection of the student group in the classroom in step S2 uses a two-stage feature fusion attention module to replace the Concat module of YOLOv8. The main process of the two-stage feature fusion attention module includes the following steps: S21: Input: The image is input into the feature extraction network, and two different feature representations are obtained through different mapping layers. They are used as the two inputs of the Twfa module respectively. The two feature representations are then spliced ​​together to obtain a feature map, which is input into the subsequent SEAttention channel attention mechanism. S22: SEAttention channel attention mechanism is used to fuse features at different levels, mainly including the following operations in sequence: Avgpool, Linner, Relu, Linner, Sigmoid, to obtain feature weight values; S23: Feature operation: multiply the feature weight value and the original input, cross-add them, and finally concatenate them together to obtain the output feature.

3. The method and system for evaluating classroom concentration based on group gaze estimation according to claim 2, characterized in that: The specific steps of obtaining the feature weight value in step S22 are: The feature map is processed by Avgpool (average pooling) with a step size of 2 to obtain 1*1*N features; after Linear processing, the feature map is reduced to N / R times to obtain 1*1*N / R features, and after Relu and Linear processing, the feature map size is restored to 1*1*N. Finally, the Sigmoid activation function is used to limit the output range to between 0 and 1, where N is the number of input feature channels and R is the compression ratio.

4. The method and system for evaluating classroom concentration based on group gaze estimation according to claim 1, characterized in that: The structure of the sight line estimation model of the head region feature in step S3 is: Initial feature extraction is performed using two convolution operations with a window size of 3×3 and a stride of 1, followed by a maximum pooling operation with a window size of 2×2 and a stride of 2. Next, the initial features are passed through two convolution operations with upper and lower windows of 3×3 and a stride of 1, separating the two input features into two input features that are fed into Twfa. Three feature optimization operations are performed according to this structure. Finally, the optimized feature map is passed through ResNet34 for facial orientation estimation. The output Pitch is the vertical angle, and Roll is the horizontal angle.

5. The method and system for evaluating classroom concentration based on group gaze estimation according to claim 4, characterized in that: The smoothing process in step S3 is to smooth out the facial orientation deviation noise that occasionally occurs during classroom activities. The smoothing process algorithm is: in and Represents the horizontal Roll direction α h and vertical Pitch direction α v The estimated angle at the i-th moment, and In fact, the horizontal Roll direction α in the time period from t-Δt to t h and vertical Pitch direction α v The average value of the angles.

6. The method and system for evaluating classroom concentration based on group gaze estimation according to claim 1, characterized in that: The algorithm for estimating the region of interest in step S4 is: Among them, α h is the angle of the face toward the horizontal, α v is the face angle toward the vertical, x r is the distance coordinate from the center of the head to the classroom window wall, y r is the distance from the center of the head to the blackboard wall, z r represents the height from the center of the head to the ground; β represents the visual divergence angle of the human eye when it is focused; x d Indicates the horizontal coordinate range of the sight area, z d Indicates the vertical coordinate range of the sight focus area.

7. The method and system for evaluating classroom concentration based on group gaze estimation according to claim 1, characterized in that: The calculation method of the classroom concentration assessment in step S5 is as follows: Among them, the number of attentions of the i-th attention area is h i , the total number of students in the group is n, m represents the total number of marked areas, μ h Represents all areas of interest h i The average value, p t Represents the estimated value of classroom concentration at time t.

Images