Detection method for identifying abnormal change of number of people in examination room based on positions of examinees

Abstract

The invention provides a detection method for identifying abnormal change of the number of people in an examination room based on examinee positions, and relates to the technical field of computer vision processing. The detection method comprises the following steps: performing model training based on an examination room historical monitoring video to obtain an examination room examinee position detection model, generating an examination room state diagram based on a to-be-detected examination room monitoring video, obtaining the number and positioning of examinees in the examination room state diagram based on the model, and establishing an examinee sequence; constructing a reference examinee sequence, and generating a to-be-detected feature vector containing the number of examinees leaving the examination room in each time period based on the reference examinee sequence; and acquiring a feature vector data set of the change of the number of people normally leaving the examination room based on the historical monitoring video of the examination room, calculating the mahalanobis distance between the to-be-detected feature vector and the centroid of the feature vector data set, and judging that the to-be-detected examination room is abnormal when the mahalanobis distance is greater than the threshold value. By implementing the technical scheme provided by the invention, multi-person cheating behaviors except for examination room monitoring can be distinguished in an assisting manner.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision processing, and particularly relates to a detection method for abnormal changes in the number of examinees in an examination room based on the recognition of the positions of examinees. Background Art

[0002] By means of real-time collection through on-site inspection of the examination room, on-site invigilation, and monitoring videos, or manual detection in the later stage, it is a typical method for preventing and detecting cheating. With the development of machine learning, especially the wide application of deep learning, the combination of examination room video monitoring and machine learning enables intelligent video monitoring to have a high accuracy rate in judging abnormal behaviors in the examination room. Currently, intelligent video monitoring mainly focuses on the actions of individual examinees at a certain time node to issue warnings or judge the existence of fraudulent behaviors. However, in order to avoid the monitoring in the examination room, some examinees will go outside the examination room under the pretext of going to the toilet to obtain illegal information, leak questions and communicate, and then return to the examination room to cheat. Such abnormal behaviors generally occur in the form of multiple people cooperating, outside the examination room monitoring, and have strong concealment.

[0003] Therefore, how to address the cheating method of multiple people cooperating outside the examination room monitoring is an urgent problem to be solved currently. Summary of the Invention

[0004] The purpose of the present invention is to provide a detection method for abnormal changes in the number of examinees in an examination room based on the recognition of the positions of examinees, which can effectively assist in detecting the cheating method of multiple people cooperating outside the examination room monitoring.

[0005] The detection method for abnormal changes in the number of examinees in an examination room based on the recognition of the positions of examinees provided by the present invention adopts the following technical solutions:

[0006] Train a deep learning model based on historical monitoring videos of the same specification of the examination room to be detected to obtain an examination room examinee position detection model. Extract key frames of the monitoring video of the examination room to be detected at intervals of 30 seconds to generate an examination room state map. Detect the number of examinees and the corresponding position information in the examination room state map based on the examination room examinee position detection model, and establish an examinee sequence;

[0007] Construct a reference examinee sequence by taking the union of the examinee sequences in the previous 3 minutes. Obtain the detection sequence of examinees leaving the examination room in all examination room state maps with the reference examinee sequence as a reference, and generate a detection feature vector containing the number of examinees leaving the examination room in each time period based on the detection sequence;

[0008] Obtain a feature vector data set of the normal change in the number of examinees leaving the examination room based on the historical monitoring video of the examination room. Obtain the centroid of the feature vector data set and the threshold for judging outliers. Calculate the Mahalanobis distance between the detection feature vector and the centroid. When the Mahalanobis distance is greater than the threshold, it is judged that there is an abnormality in the examination room to be detected.

[0009] Optionally, in the process of training a deep learning model based on historical monitoring videos of the same specifications of the examination rooms to obtain an examination room candidate position detection model, it includes:

[0010] Obtain the key frames of the historical monitoring videos of the examination rooms, label the positions of the candidates in the key frames to obtain a training data set and a test data set, and preset anchor boxes suitable for candidates and the intersection over union threshold for determining candidates for the initial single-stage examination room candidate position detection model;

[0011] Pre-train the initial single-stage examination room candidate position detection model based on a public data set, adjust the hyperparameters of the pre-trained model, and fine-tune the model using the training data set to obtain a to-be-tested examination room candidate position detection model, and the fine-tuning uses cosine learning rate decay;

[0012] Evaluate the to-be-tested examination room candidate position detection model based on the test data set. When the harmonic mean of the model is less than the set expected value, reset the hyperparameters and retrain, otherwise save the weights of the to-be-tested examination room candidate position detection model.

[0013] Optionally, in the process of executing the extraction of key frames of the monitoring video of the to-be-detected examination room at 30-second intervals to generate an examination room status map, it includes:

[0014] After the exam, obtain the monitoring video of the to-be-detected examination room, divide the monitoring video of the examination room at 30-second intervals, obtain a sequence of image frames of 30 seconds, use a key frame extraction tool to extract key frames from the video within 30 seconds, convert the extracted one or several key frame images into standard images through brightness processing and size transformation, and use a no-reference image sharpness detection algorithm for the standard images to obtain the clearest image as the examination room status map for this time period.

[0015] Optionally, in the process of detecting the number of candidates and the corresponding position information in the examination room status map based on the examination room candidate position detection model and establishing a candidate sequence, it includes:

[0016] Input the examination room status map into the examination room candidate position detection model for detection to obtain the number of prediction boxes and the positions of the prediction boxes, remove redundant anchor boxes based on the non-maximum suppression method, and determine whether to leave the examination room based on whether there are candidates in the prediction boxes to determine the number of candidates;

[0017] When the intervals between candidate seats are small, calculate the center point coordinates of each prediction box, sort the center point coordinates, and make a one-to-one correspondence with the examination room seat numbers based on the sorting result to obtain candidate positioning;

[0018] When the intervals between examination room seats are large, perform grid division according to the positions of the prediction boxes, number each grid according to the examination room seat arrangement rules to obtain candidate positioning;

[0019] Construct a binary sequence of the corresponding examination room status diagram based on the standard number of people in the examination room and the number and positions of candidates obtained from the detection.

[0020] Optionally, obtain a sequence to be detected for candidates leaving the examination room in all examination room status diagrams by using the reference candidate sequence as a comparison, including:

[0021] Obtain the number and positioning of candidates taking this examination in the examination room to be detected based on the reference candidate sequence, perform an exclusive OR operation on the reference sequence and the candidate sequences of each examination room status diagram, and obtain a sequence to be detected containing the number of candidates leaving the examination room during the time interval in which each examination room status diagram is located.

[0022] Optionally, generate a sequence to be detected feature vector containing the number of candidates leaving the examination room in each time period, including:

[0023] Divide the time periods according to the total examination duration, perform an OR operation on the sequences to be detected within the time periods, obtain a sequence of changes in the number of candidates leaving the examination room in each time period, and construct a sequence to be detected feature vector containing the number of candidates leaving the examination room in each time period based on the sequence of changes in the number of people.

[0024] Optionally, in the process of dividing the time periods according to the total examination duration, including:

[0025] When the examination duration is 120 minutes, the divided time periods are respectively set as 1 minute, 2 minutes, 5 minutes, 10 minutes, 30 minutes and 60 minutes.

[0026] Optionally, obtain a feature vector data set of the normal change in the number of people leaving the examination room based on the historical monitoring video of the examination room, including:

[0027] Manually detect and eliminate suspicious monitoring videos with candidates frequently entering and leaving in the historical monitoring video, detect the remaining monitoring videos through the examination room candidate position detection model, establish a candidate sequence based on the detection results, and obtain a feature vector data set containing the number of candidates leaving the examination room in each time period based on the candidate sequence.

[0028] Optionally, in the process of obtaining the centroid of the feature vector data set and the threshold for judging outliers, including:

[0029] Calculate the average value of the features in each time period of the feature vector data set to obtain the centroid of the data set, calculate the Mahalanobis distance between all feature vectors in the data set and the centroid, and use an adaptive threshold algorithm to calculate the Mahalanobis distance threshold for judging outliers.

[0030] Optionally, calculate the Mahalanobis distance between the sequence to be detected feature vector and the centroid, using the following formula:

[0031]

[0032] Among them, D M (x) is the Mahalanobis distance between the centroid and the feature vector to be detected, and (x - μ) T is the covariance matrix of each time period between the centroid and the feature vector to be detected, and ∑ -1 (x - μ) is the inverse matrix of the covariance matrix, x is the feature vector to be detected, and μ is the centroid.

[0033] A detection method for identifying abnormal changes in the number of examinees in a test room based on the positions of examinees proposed by the present invention has the beneficial effects that:

[0034] 1. By calculating the Mahalanobis distance between the feature vector of the test room to be detected and the data set of the historical feature vectors of the test room, the abnormality of the number of examinees leaving the test room in this session can be distinguished, and further assist in detecting whether there are multiple cheating behaviors. Description of the Drawings

[0035] Figure 1 is the flowchart of the detection method provided by the present invention;

[0036] Figure 2 is the sequence diagram of examinees leaving at time t provided by the present invention;

[0037] Figure 3 is the schematic diagram of examinees leaving the test room at time t provided by the present invention;

[0038] Figure 4 is the schematic diagram for outlier judgment. Detailed Embodiments

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meaning as understood by those of ordinary skill in the art in the field to which the present invention belongs. The words such as "including" used herein mean that the elements or items appearing before this word cover the elements or items listed after this word and their equivalents, without excluding other elements or items.

[0040] An embodiment of the present invention provides a detection method for identifying abnormal changes in the number of examinees in a test room based on the positions of examinees. Refer to Figure 1 , including:

[0041] S1. Train a deep learning model based on historical monitoring videos of the same specification in the test examination room to obtain an examination room candidate position detection model. Extract key frames from the monitoring video of the test examination room at 30-second intervals to generate an examination room status map. Detect the number of candidates and their corresponding position information in the examination room status map based on the examination room candidate position detection model, and establish a candidate sequence;

[0042] S2. Construct a reference candidate sequence by taking the union of candidate sequences in the first 3 minutes. Obtain a test sequence of candidates leaving the examination room in all examination room status maps with the reference candidate sequence as a reference, and generate a test feature vector containing the number of candidates leaving the examination room in each time period based on the test sequence;

[0043] S3. Obtain a feature vector dataset of the normal change in the number of candidates leaving the examination room based on historical monitoring videos of the examination room. Obtain the centroid of the feature vector dataset and the threshold for judging outliers. Calculate the Mahalanobis distance between the test feature vector and the centroid. When the Mahalanobis distance is greater than the threshold, it is judged that there is an abnormality in the test examination room.

[0044] In some embodiments, during the execution of step S1, it includes:

[0045] S1.1. Train a deep learning model based on historical monitoring videos of the same specification in the test examination room to obtain an examination room candidate position detection model;

[0046] S1.2. Extract key frames from the monitoring video of the test examination room at 30-second intervals to generate an examination room status map;

[0047] S1.3. Detect the number of candidates and their corresponding position information in the examination room status map based on the examination room candidate position detection model, and establish a candidate sequence.

[0048] Specifically, during the execution of step S1.1, in the process of training a deep learning model based on historical monitoring videos of the same specification in the test examination room to obtain an examination room candidate position detection model, it includes:

[0049] Obtain the key frames of the historical monitoring video of the examination room, annotate the positions of candidates in the key frames to obtain a training dataset and a test dataset, and preset anchor boxes suitable for candidates and the intersection over union threshold for judging candidates for the initial single-stage examination room candidate position detection model;

[0050] Pre-train the initial single-stage examination room candidate position detection model based on a public dataset, adjust the hyperparameters of the pre-trained model, and fine-tune the model using the training dataset to obtain a test examination room candidate position detection model. The fine-tuning uses cosine learning rate decay;

[0051] Evaluate the candidate position detection model for the test examination room based on the test data set. When the harmonic mean of the model is less than the set expected value, reset the hyperparameters and retrain. Otherwise, save the weights of the candidate position detection model for the test examination room.

[0052] In fact, the present invention aims to detect the collective cheating behavior of candidates in the examination room. Considering that there are a large number of examination rooms in an exam and the exam time is long, there are high requirements for the detection efficiency. Therefore, in this example, the single-stage object detection model YOLOV5m is used for candidate detection. The YOLOV5 model can ensure both the detection speed and accuracy while occupying less memory, which is suitable for this scenario.

[0053] Furthermore, since there are many examination rooms in an exam and the exam time is long, there is too much redundant information in the examination room surveillance video. In this example, ffmpeg is used to obtain all the key frames in the historical surveillance video of the examination room. After obtaining the key frames, labelImg is used to annotate the candidate positions in the key frames and save them in the YOLO format. In this way, the candidate position annotation images from the examination room surveillance are obtained, and the Mosaic method is used to perform data augmentation on the training data, where 75% of the images are used as the training data set and 25% as the test data set.

[0054] Furthermore, in this example, the public pedestrian data sets USC Pedestrian Database and MIT Pedestrian Database are first used for pre-training the model. After the pre-training is completed, the model hyperparameters are adjusted, and then the training data set is used for fine-tuning the model. In the model fine-tuning, cosine learning rate decay is used, the initial learning rate is set to 0.01, the decay period T_max is set to 20, and eta_min, that is, the minimum value of the learning rate decay, is set to 0.001. The Adam optimizer is used, the momentum is set to 0.937, and the weight decay is 0.0005. A total of 600 rounds of training are performed. Since candidates may appear densely in the examination room environment, in this example, the standard NMS (Non-Maximum Suppression) used in the YOLOV5m model is replaced with adaptive NMS (Adaptive Non-Maximum Suppression) to adapt to the examination room environment.

[0055] Finally, after training the model with the training data set, the test data set is used for testing. In this example, the IOU (Intersection over Union) threshold between the prediction box and the ground truth box is set to 0.5, and its calculation formula is where A is the ground truth box, B is the prediction box, and A∩B is the area of the intersection of A and B. Since there are only the human class and the background class, only the P value (precision) and R value (recall) of the model need to be calculated, and the harmonic mean is calculated based on the P value and R value In this example, the model is evaluated every 20 training rounds. If the F value is less than 0.8, the hyperparameters are reset and training is restarted. If there is an F value greater than 0.8, the weights of the model with the highest F value are saved.

[0056] Specifically, in the process of executing step S1.2 and extracting key frames of the monitored examination room video to generate an examination room status map at intervals of 30 seconds, it includes:

[0057] After the exam ends, obtain the monitored examination room video, divide the monitored examination room video at intervals of 30 seconds, obtain a sequence of image frames of 30 seconds, use a key frame extraction tool to extract key frames from the video within 30 seconds, convert the extracted one or several key frame images into standard images through brightness processing and size transformation, use a reference-free image sharpness detection algorithm for the standard images, and obtain the clearest image as the examination room status map for this time period.

[0058] Further, according to the above method, a total of 240 examination room status maps of the examination room to be detected can be obtained, denoted as {f 1 , f 2 , f 3 , …, f 240}.

[0059] Specifically, in the process of executing step S1.3 and detecting the number of candidates and their corresponding position information in the examination room status map based on the examination room candidate position detection model and establishing a candidate sequence, it includes:

[0060] Input the examination room status map into the examination room candidate position detection model for detection to obtain the number of prediction boxes and the positions of the prediction boxes. Based on the non-maximum suppression method, redundant anchor boxes are removed, and whether a candidate has left the examination room is determined based on whether there is a candidate in the prediction box to determine the number of candidates;

[0061] When the intervals between candidate seats are small, calculate the center point coordinates of each prediction box, sort the center point coordinates, and establish a one-to-one correspondence with the examination room seat numbers based on the sorting result to obtain candidate positioning;

[0062] When the intervals between examination room seats are large, perform grid division according to the positions of the prediction boxes, number each grid according to the examination room seat arrangement rules to obtain candidate positioning;

[0063] Construct a binary sequence corresponding to the examination room status map based on the standard number of candidates in the examination room and the number and positions of the candidates detected.

[0064] In some embodiments, in the process of executing step S2, it includes:

[0065] S2.1. Construct a benchmark candidate sequence by taking the union of candidate sequences in the first 3 minutes;

[0066] S2.2. Obtain the sequence to be detected of candidates leaving the examination room in all examination room status diagrams with reference to the benchmark candidate sequence;

[0067] S2.3. Generate a feature vector to be detected containing the number of candidates leaving the examination room in each time period based on the sequence to be detected.

[0068] Specifically, when performing step S2.2, obtaining the sequence to be detected of candidates leaving the examination room in all examination room status diagrams with reference to the benchmark candidate sequence includes:

[0069] Obtain the number of candidates taking this examination in the examination room to be detected and their positions based on the benchmark candidate sequence, and perform an exclusive OR operation on the benchmark sequence and the candidate sequence of each examination room status diagram to obtain a sequence to be detected containing the number of candidates leaving the examination room during the time interval corresponding to each examination room status diagram.

[0070] Actually, referring to Figure 2 and Figure 3 , by performing an exclusive OR operation on the candidate sequence at time t and the benchmark sequence, the candidates leaving the examination room at time t can be obtained.

[0071] Specifically, when performing step S2.3, generating a feature vector to be detected containing the number of candidates leaving the examination room in each time period includes:

[0072] Divide the time periods according to the total examination duration, perform an OR operation on the sequences to be detected within the time periods, obtain a sequence of changes in the number of candidates leaving the examination room in each time period, and construct a feature vector to be detected containing the number of candidates leaving the examination room in each time period based on the sequence of changes in the number of candidates.

[0073] Further, during the process of dividing the time periods according to the total examination duration, it includes: when the examination duration is 120 minutes, the divided time periods are set to 1 minute, 2 minutes, 5 minutes, 10 minutes, 30 minutes, and 60 minutes respectively.

[0074] Actually, in this embodiment, the examination duration is set to 120 minutes. Calculate the sequences of changes in the number of candidates within every 1 minute, every 2 minutes, every 5 minutes, every 10 minutes, and every 30 minutes respectively. Combine the sequences of changes in the number of candidates in each time period as feature vectors, and a feature vector to be detected with a length of 220 can be obtained, denoted as x = (x 1 , x 2 ,..., x 220 ) T .

[0075] In some embodiments, during the process of performing step S3, it includes:

[0076] S3.1. Obtain a characteristic vector dataset of the change in the number of candidates leaving the examination room normally based on the historical monitoring videos of the examination room; S3.2. Obtain the centroid of the characteristic vector dataset and the threshold for judging outliers;

[0077] S3.3. Calculate the Mahalanobis distance between the characteristic vector to be detected and the centroid;

[0078] S3.4. Judge abnormal candidates based on the detection results.

[0079] Specifically, when executing step S3.1, obtaining a characteristic vector dataset of the change in the number of candidates leaving the examination room normally based on the historical monitoring videos of the examination room includes:

[0080] Manually detect and remove suspicious monitoring videos with candidates frequently entering and leaving in the historical monitoring videos, detect the remaining monitoring videos through the candidate position detection model in the examination room, establish a candidate sequence based on the detection results, and obtain a characteristic vector dataset containing the number of candidates leaving the examination room in each time period based on the candidate sequence.

[0081] Specifically, when executing step S3.2, in the process of obtaining the centroid of the characteristic vector dataset and the threshold for judging outliers, it includes:

[0082] Calculate the average value of the characteristics in each time period of the characteristic vector dataset to obtain the centroid of the dataset, calculate the Mahalanobis distance between all the characteristic vectors in the dataset and the centroid, and use the adaptive threshold algorithm to calculate the Mahalanobis distance threshold for judging outliers.

[0083] Actually, calculate the average value of the Mahalanobis distances between all the characteristic vectors in the dataset and the centroid, and then multiply by k to obtain the Mahalanobis distance threshold for judging outliers.

[0084] Furthermore, in this embodiment, k is set to 3.

[0085] Specifically, when executing step S3.3, calculate the Mahalanobis distance between the characteristic vector to be detected and the centroid, and use the following formula:

[0086]

[0087] where D M (x) is the Mahalanobis distance between the centroid and the characteristic vector to be detected, (x - μ) T is the covariance matrix of each time period between the centroid and the characteristic vector to be detected, ∑ -1 (x - μ) is the inverse matrix of the covariance matrix, x is the characteristic vector to be detected, and μ is the centroid.

[0088] Actually, taking the judgment of the outlier of the characteristic vector of the i-th examination room to be tested as an example, obtain the i-th characteristic vector, denoted as x = (x 1 , x2 ,..., x 220 ) T , calculate its Mahalanobis distance from the centroid μ. If the value is greater than t, it is determined that the number of people in the examination room has changed abnormally.

[0089] Furthermore, refer to Figure 4 . The black circular data in the figure represents the eigenvectors of the normal number change obtained in the past three years (this figure only uses two-dimensional features for demonstration). The triangle is the obtained centroid μ, and the rhombus is the feature of the examination room to be judged. If its distance from μ is greater than the threshold t, it means it is an outlier outside the dotted line box. It can be understood that in this example, the Mahalanobis distance is used as the outlier judgment because the Mahalanobis distance method can detect outliers of multivariate variables, and the Mahalanobis distance is not affected by the dimension. Because the change range of the number of people in the examination room in the early, middle, and late stages of the exam is often different, the Mahalanobis distance method can effectively determine abnormal data. The parameter k for threshold calculation is generally set to 2 or 3. Since the data set in this example has been manually removed of outliers, it is set to 3 in this example.

[0090] Furthermore, if the Mahalanobis distance between x and the centroid μ is less than the threshold t, but the change range of the number of people is too large during a certain period of time. For this situation, in this example, thresholds are set for different time periods, namely 2, 2, 3, 4, 5 for 1 minute, 2 minutes, 5 minutes, 10 minutes, and 30 minutes respectively. That is, in this example, each value in the eigenvector of length 220 will be judged and compared with the corresponding threshold. If it is greater than or equal to this threshold, the eigenvector of the examination room is determined to be an outlier.

[0091] Specifically, when performing step S3.4, judging abnormal candidates based on the detection results includes:

[0092] If it is judged as an outlier, obtain the sequence stuNoout of the admission ticket numbers of the candidates entering and leaving the examination room. Carefully check the surveillance video of the examination room to confirm whether there is any cheating behavior among the candidates who leave the examination room halfway. Then obtain the information of the candidates entering and leaving the same or different examination rooms at the same time, obtain their test papers, and then judge whether there are identical test papers through professional graders, and finally confirm whether there is any fraud behavior.

[0093] Although the embodiments of the present invention have been described in detail above, it is obvious to those skilled in the art that various modifications and changes can be made to these embodiments. However, it should be understood that such modifications and changes all fall within the scope and spirit of the present invention described in the claims. Moreover, the present invention described herein may have other embodiments and can be implemented or realized in various ways.

Claims

1. A detection method for abnormal changes in the number of people in an examination room based on the position of examinees, characterized in that: The following steps are involved: Based on historical surveillance videos of the same specifications of the examination room to be tested, the deep learning model is trained to obtain the examination room candidate position detection model, and the key frames of the examination room surveillance video to be tested are extracted at intervals of 30 seconds to generate the examination room status diagram. Based on the examination room candidate position detection model, the number of candidates and the corresponding position information in the examination room status diagram are detected, and the candidate sequence is established; A benchmark candidate sequence is constructed by taking the union of the candidate sequences in the first 3 minutes, and a sequence to be detected of candidates leaving the examination room in all examination room status diagrams is obtained based on the benchmark candidate sequence, and a feature vector to be detected containing the number of candidates leaving the examination room in each time period is generated based on the sequence to be detected; Based on the historical monitoring video of the examination room, a feature vector data set of the normal change in the number of people leaving the examination room is obtained, the centroid of the feature vector data set and the threshold for judging the outlier are obtained, and the Mahalanobis distance between the feature vector to be detected and the centroid is calculated. When the Mahalanobis distance is greater than the threshold, it is judged that there is an abnormality in the examination room to be detected.

2. A method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: The process of training the deep learning model based on the historical surveillance video of the same specifications of the examination room to be detected to obtain the examination room candidate location detection model includes: Obtain key frames of historical surveillance videos of the examination room, annotate the positions of examinees in the key frames to obtain training data sets and test data sets, and preset anchor frames suitable for examinees and intersection-over-union ratio thresholds for determining examinees for the initial single-stage examinee position detection model; Perform initial single-stage pre-training of the examination room candidate location detection model based on the public data set, adjust the pre-trained model hyperparameters, and use the training data set to fine-tune the model to obtain the examination room candidate location detection model to be tested, wherein the fine-tuning uses cosine learning rate decay; The candidate position detection model for the test room is evaluated based on the test data set. When the harmonic mean of the model is less than the set expected value, the hyperparameters are reset and retraining is performed. Otherwise, the weights of the candidate position detection model for the test room are saved.

3. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: The process of extracting key frames of the surveillance video of the examination room to be detected based on 30-second intervals to generate the examination room status diagram includes: After the test, obtain the surveillance video of the examination room to be tested, divide the examination room surveillance video into 30-second intervals, obtain a 30-second image frame sequence, use the key frame extraction tool to extract key frames from the video within 30 seconds, and convert the extracted one or several key frame images into standard images through brightness processing and size transformation. Use the reference-free image clarity detection algorithm for the standard images to obtain the clearest image as the examination room status diagram for that time period.

4. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: The process of detecting the number of examinees and their corresponding position information in the examination room state diagram based on the examination room examinee position detection model and establishing the examinee sequence includes: The examination room state diagram is input into the examination room candidate position detection model for detection to obtain the number of prediction boxes and the positions of the prediction boxes. The redundant anchor boxes are removed based on the non-maximum suppression method. Based on whether there is a candidate in the prediction box, it is judged whether to leave the examination room to determine the number of candidates. When the distance between the examinees' seats is small, the coordinates of the center points of each prediction frame are calculated, the center point coordinates are sorted, and the sorting results are matched one-to-one with the examination room seat numbers to obtain the examinee's location; When the seats in the examination room are spaced far apart, grids are divided according to the position of the prediction box, and each grid is numbered according to the examination room seat arrangement rules to obtain the candidate's location; Based on the standard number of people in the examination room and the number and positions of candidates obtained by detection, a binary sequence corresponding to the examination room status diagram is constructed.

5. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: Based on the benchmark candidate sequence, the sequences to be detected in which candidates leave the examination room in all examination room status diagrams are obtained, including: Based on the benchmark candidate sequence, the number and location of candidates taking the exam in the examination room to be tested are obtained, and the benchmark sequence is XORed with the candidate sequence of each examination room status diagram to obtain the sequence to be tested that includes the number of candidates leaving the examination room within the time interval of each examination room status diagram.

6. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: Based on the sequence to be detected, a feature vector to be detected is generated, which includes the number of candidates leaving the examination room in each time period, including: The total duration of the examination is divided into time periods, and the sequences to be detected in the time periods are ORed to obtain a sequence of changes in the number of examinees leaving the examination room in each time period, and a feature vector to be detected containing the number of examinees leaving the examination room in each time period is constructed based on the sequence of changes in the number of examinees.

7. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 6, characterized in that: The process of dividing the time periods according to the total duration of the exam includes: When the exam duration is 120 minutes, the time periods are set as 1 minute, 2 minutes, 5 minutes, 10 minutes, 30 minutes and 60 minutes respectively.

8. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: Based on the historical surveillance video of the examination room, the feature vector dataset of the normal number of people leaving the examination room is obtained, including: Through manual detection, suspicious surveillance videos in which candidates frequently enter and exit the examination room are eliminated. The remaining surveillance videos are detected through the examination room candidate location detection model, and a candidate sequence is established based on the detection results. Based on the candidate sequence, a feature vector data set containing the number of candidates leaving the examination room in each time period is obtained.

9. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: The process of obtaining the centroid of the feature vector data set and the threshold value for determining an outlier includes: The features of each time period in the feature vector data set are averaged to obtain the centroid of the data set, the Mahalanobis distance between all feature vectors in the data set and the centroid is calculated, and the Mahalanobis distance threshold for judging outliers is calculated using an adaptive threshold algorithm.

10. The method for detecting abnormal changes in the number of people in an examination room based on the position of examinees according to claim 1, characterized in that: The Mahalanobis distance between the feature vector to be detected and the centroid is calculated using the following formula: Among them, D M (x) is the Mahalanobis distance between the centroid and the feature vector to be detected, (x-μ) T is the covariance matrix between the centroid and the feature vector to be detected in each time period, ∑ -1 (x-μ) is the inverse matrix of the covariance matrix, x is the eigenvector to be detected, and μ is the centroid.

Images