Dynamic topology-based method and apparatus for analyzing collective anomalous behavior

Abstract

Disclosed in the present invention are a dynamic topology-based method and apparatus for analyzing a collective anomalous behavior. The method comprises: acquiring examination monitoring video data of an examination room; selecting key frames from examination frame data, performing portrait identification on the key frames sequentially, and labelling examinee information in the key frames; constructing a skeleton topology of each examinee, and continuously monitoring the skeleton topology and the examination monitoring video data; sequentially recording a suspicious posture of each skeleton topology, and, on the basis of coordinate information in the examination monitoring video data of the skeleton topology with the suspicious posture, obtaining an actual interval distance between the skeleton topologies with the suspicious postures; using two skeleton topologies with suspicious postures corresponding to the actual interval distance as an anomalous behavior skeleton, and outputting the examinee information labelled in the anomalous behavior skeleton, so as to complete the analysis of the collective anomalous behavior. The present invention solves the technical problem of failing to accurately and efficiently detect and analyze collective anomalous behaviors of examinees in the prior art.

Description

A method and device for analyzing collective abnormal behavior based on dynamic topology Technical Field

[0001] The present invention relates to the field of behavior recognition technology, and in particular to a method and device for analyzing collective abnormal behavior based on dynamic topology. Background Art

[0002] With the development of current visual recognition technology, machine vision recognition technology used in examination halls can accurately obtain the movements and behaviors of individual examinees, and process the images of the examinees obtained, so as to determine whether the examinees' behaviors in the examination halls are suspected of abnormal behavior.

[0003] At present, the detection of abnormal behavior is mainly achieved through machine vision recognition to monitor the abnormal behavior of candidates. However, since image recognition can only identify the behavior of candidates based on images, there will be a large visual recognition blind spot. For example, when a candidate puts his hand under the table, image recognition cannot predict and identify the candidate's hand movement, resulting in the inability to detect the specific movement in this situation. At the same time, the existing abnormal behavior detection is only for individual candidates, and it is impossible to accurately identify candidates with collective abnormal behavior. As a result, candidates with collective abnormal behavior can circulate answers through collective actions, resulting in the inability of existing abnormal behavior recognition to accurately analyze and thus cannot be effectively used in the identification of collective abnormal behavior.

[0004] Therefore, there is an urgent need for a method that can be used to detect and analyze collective abnormal behavior of examinees and improve the accuracy and efficiency of collective abnormal behavior detection and analysis. Summary of the Invention

[0005] The present invention provides a method and device for analyzing collective abnormal behavior based on dynamic topology to solve the technical problem in the prior art that it is impossible to accurately and efficiently detect and analyze the collective abnormal behavior of examinees.

[0006] To solve the above technical problems, an embodiment of the present invention provides a method for analyzing collective abnormal behavior based on dynamic topology, including:

[0007] Acquire examination monitoring video data of the examination room, and perform frame division and preprocessing on the video data; wherein the video data after frame division includes a plurality of examination frame data;

[0008] Select key frames from the test frame data, perform portrait recognition on the key frames in turn, compare and analyze the recognized portraits with the preset candidate information database, and mark the candidate information obtained from the analysis in the key frames

[0009] Capturing skeleton topology information of the annotated portraits to construct the skeleton topology of each examinee, and continuously monitoring the skeleton topology and examination surveillance video data based on a preset suspicious behavior recognition model;

[0010] The suspicious posture of each skeleton topology is recorded in turn, and the actual distance between the skeleton topologies with suspicious postures is obtained based on the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data;

[0011] If the actual interval distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual interval distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, thereby completing the analysis of collective abnormal behavior.

[0012] As a preferred solution, the acquisition of the examination monitoring video data of the examination room and the frame division and preprocessing of the video data are specifically as follows:

[0013] Obtain examination monitoring video data of the examination room through the camera;

[0014] Dividing the test monitoring video data into frames to obtain image frames;

[0015] Gaussian filtering, image denoising and image enhancement are performed on the image frame to obtain test frame data.

[0016] As a preferred solution, the key frames are selected from the test frame data, and portrait recognition is performed on the key frames in sequence, and the recognized portraits are compared and analyzed with the preset candidate information database, and the candidate information obtained from the analysis is marked in the key frames, specifically:

[0017] Performing portrait recognition on all test frame data according to the portrait recognition model, thereby selecting key frames in which portraits are not obscured from the test frame data, and obtaining portrait information after portrait recognition of the key frames; wherein all portraits can be recognized in each key frame;

[0018] The identified portrait information is compared and analyzed with the preset candidate information database, and the portrait information identified in the key frames is marked with the candidate information in turn.

[0019] As a preferred solution, the skeleton topology information of the annotated portrait is captured to construct the skeleton topology of each examinee respectively, and the skeleton topology and examination monitoring video data are continuously monitored according to a preset suspicious behavior recognition model, specifically:

[0020] Capture the skeleton topology information of the portraits with the candidate information marked in all key frames, and then construct the skeleton topology of each candidate in turn;

[0021] Based on the preset behavior recognition model, the behavior posture of each candidate is recognized on the skeleton topology to obtain the behavior information of each candidate in each key frame;

[0022] Based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the exam monitoring video data;

[0023] Based on a preset suspicious behavior identification model, all behavioral information of each candidate in the exam surveillance video data is monitored and identified to obtain a skeleton topology where suspicious behaviors exist; wherein the suspicious behaviors include all actions unrelated to writing, reading, and raising the head.

[0024] As a preferred solution, based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the test monitoring video data, specifically:

[0025] The entire behavioral information of each examinee is predicted and completed in sequence, so that in the process of predicting and completing the entire behavioral information of each examinee, the examination frame image that can identify the examinee's complete skeleton topology and behavioral information is extracted from the examination monitoring video data according to the preset behavior recognition model, and combined with the examinee's behavioral information in each key frame, the skeleton posture and behavior are predicted in sequence for the frame moments corresponding to the examination frame data that cannot identify the examinee's skeleton topology and behavioral information through the behavior prediction model, and then the predicted skeleton posture and behavior information are input into the examination frame data at the frame moment in sequence, until the examination frame data at all frame moments have the examinee's skeleton posture and behavior information, thereby obtaining the examinee's entire behavioral information in the examination monitoring video data;

[0026] Until all the behavioral information of all candidates in the examination monitoring video data is obtained.

[0027] As a preferred solution, if the actual separation distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual separation distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, specifically:

[0028] If the actual separation distance is less than the preset value, the skeleton topologies of the two suspicious postures corresponding to the actual separation distance are both used as abnormal behavior analysis skeletons;

[0029] Performing behavior recognition on the abnormal behavior analysis skeletons in sequence, and whenever a suspicious behavior is identified in one of the abnormal behavior analysis skeletons, obtaining behavior information of all accomplice skeletons within a preset distance of the abnormal behavior analysis skeleton, so that whenever the abnormal behavior analysis skeleton performs a suspicious behavior, if an accomplice skeleton is identified as performing a head-raising action and / or a writing action, then recording the number of times the abnormal behavior analysis skeleton and the accomplice skeleton cyclically perform the same action;

[0030] If the number is greater than the preset number, the abnormal behavior analysis skeleton and the accomplice skeleton are regarded as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0031] As a preferred solution, it also includes:

[0032] If the actual separation distance is greater than or equal to the preset value, the skeleton topology of the suspicious posture is sequentially subjected to behavior recognition;

[0033] Whenever a suspicious behavior is detected in a skeleton topology with a suspicious posture, the behavior information of all accomplice skeletons within a preset distance of the skeleton topology is obtained. Therefore, whenever the skeleton topology with a suspicious posture is performing a suspicious behavior, if an accomplice skeleton is identified as performing a head-raising action and / or a writing action, the number of times the skeleton topology with a suspicious posture and the accomplice skeleton repeatedly perform the same action is recorded.

[0034] If the number is greater than the preset number, the skeleton topology of the suspicious posture and the accomplice skeleton are used as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0035] Accordingly, the present invention also provides a collective abnormal behavior analysis device based on dynamic topology, comprising: an acquisition module, a recognition module, a skeleton module, a distance module and an analysis module;

[0036] The acquisition module is used to acquire the examination monitoring video data of the examination room, and to perform frame division and preprocessing on the video data; wherein the video data after frame division includes a plurality of examination frame data;

[0037] The recognition module is used to select key frames from the test frame data, perform portrait recognition on the key frames in turn, compare and analyze the recognized portraits with the preset candidate information database, and mark the candidate information obtained from the analysis in the key frames.

[0038] The skeleton module is used to capture the skeleton topology information of the annotated portraits, thereby constructing the skeleton topology of each examinee respectively, and continuously monitoring the skeleton topology and examination surveillance video data based on a preset suspicious behavior recognition model;

[0039] The distance module is used to record the suspicious posture of each skeleton topology in turn, and obtain the actual distance between the skeleton topologies with suspicious postures based on the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data;

[0040] The analysis module is used to use the skeleton topology of the two suspicious postures corresponding to the actual interval distance as the abnormal behavior skeleton if the actual interval distance is less than the preset value, and output the candidate information marked in the abnormal behavior skeleton, thereby completing the analysis of collective abnormal behavior.

[0041] Correspondingly, the present invention also provides a terminal device, characterized in that it includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the collective abnormal behavior analysis method based on dynamic topology as described in any one of the above.

[0042] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the collective abnormal behavior analysis method based on dynamic topology as described in any one of the above.

[0043] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0044] The technical solution of the present invention obtains the examination monitoring video data of the examination room, and then selects key frames after frame division and preprocessing, so as to mark the candidate information corresponding to the portraits identified in sequence, and capture the skeleton topology information of the marked portraits to construct the skeleton topology of each candidate, so as to continuously monitor the skeleton topology and examination monitoring video data through a preset suspicious behavior recognition model, and at the same time combine the actual interval distance between the skeleton topologies of each suspicious posture to accurately and quickly determine whether the skeleton topologies of two suspicious postures between the actual interval distance are abnormal behavior skeletons, and then quickly output information on the candidates corresponding to the collective abnormal behavior. The analysis of skeleton behavior can avoid the blind spots of machine vision recognition and the problem of inaccurate machine vision behavior detection. The accuracy of collective abnormal behavior detection is improved by combining the detection of actual interval distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] FIG1 is a flowchart of a method for analyzing collective abnormal behavior based on dynamic topology according to an embodiment of the present invention;

[0046] FIG2 is a schematic structural diagram of a collective abnormal behavior analysis device based on dynamic topology provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0048] Example 1

[0049] Referring to FIG. 1 , a method for analyzing collective abnormal behavior based on dynamic topology according to an embodiment of the present invention is provided, which includes the following steps S101 to S105:

[0050] Step S101: Acquire examination monitoring video data of an examination room, and perform frame division and preprocessing on the video data; wherein the video data after frame division includes a plurality of examination frame data.

[0051] As a preferred solution of this embodiment, the acquisition of the examination monitoring video data of the examination room and the frame division and preprocessing of the video data are specifically as follows:

[0052] The examination monitoring video data of the examination room is obtained through a camera; the examination monitoring video data is frame-divided to obtain image frames; the image frames are subjected to Gaussian filtering, image denoising and image enhancement processing to obtain examination frame data.

[0053] In this embodiment, the camera can be used to obtain the examination monitoring video data of the examination room, and then the obtained video data is divided into frames, thereby reducing the volume of the entire video data. It is understandable that since an examination generally lasts more than half an hour and less than four hours, if all the video data is processed, the data processing will be complex, the data will be complicated, and the data processing efficiency will be low. Therefore, by dividing the video data into frames, the number of frames in one second can be reduced, thereby achieving the data volume of the video data. Preferably, the video data can be divided into a frame rate of one frame per second, which can significantly reduce the processing volume of the video data compared to the normal 30 frame rate or 60 frame rate.

[0054] Furthermore, Gaussian filtering, image denoising and image enhancement processing are performed on each image frame in the video data after frame division, so as to obtain test frame data that can be processed quickly, efficiently and accurately.

[0055] Step S102: Select key frames from the test frame data, perform portrait recognition on the key frames in turn, compare and analyze the recognized portraits with the preset candidate information database, and mark the candidate information obtained from the analysis in the key frames.

[0056] As a preferred solution of this embodiment, the key frames are selected from the test frame data, and portrait recognition is performed on the key frames in sequence, and the recognized portraits are compared and analyzed with the preset candidate information database, and the candidate information obtained by the analysis is marked in the key frames, specifically:

[0057] According to the portrait recognition model, portrait recognition is performed on all test frame data, so as to select key frames in which the portraits are not blocked from the test frame data, and obtain the portrait information after the portraits in the key frames are recognized; wherein, all portraits can be recognized in each key frame; the recognized portrait information is compared and analyzed with the preset candidate information database, and the portrait information recognized in the key frames is marked with the candidate information in turn.

[0058] In this embodiment, the portrait recognition model can accurately recognize portraits in all test frame data. Frames in which the portraits are not obscured are selected from the test frame data as key frames. The portrait information in the key frames after portrait recognition is then identified, so that candidate information corresponding to the portrait information can be obtained in a preset candidate information database. Preferably, the portrait information can be the result of facial recognition, and the facial recognition result is compared with images in the preset candidate information database to determine whether the portrait in the key frame corresponds to the candidate information in the candidate information database.

[0059] It should be noted that the portrait recognition model may be a face recognition model, which, after training, can accurately recognize facial feature information in an image.

[0060] It is understandable that the selection of key frames where the portrait is not occluded can improve the accuracy of portrait recognition, avoid the low accuracy of portrait recognition and skeleton feature information capture in subsequent steps due to occlusion, and thus lead to low accuracy of abnormal behavior analysis.

[0061] Step S103: capturing skeleton topology information of the annotated portrait, thereby constructing the skeleton topology of each examinee respectively, and continuously monitoring the skeleton topology and examination monitoring video data according to a preset suspicious behavior recognition model.

[0062] As a preferred solution of this embodiment, the skeleton topology information of the annotated portrait is captured, thereby constructing the skeleton topology of each examinee respectively, and continuously monitoring the skeleton topology and examination monitoring video data according to a preset suspicious behavior recognition model, specifically:

[0063] The skeleton topology information of the portraits marked with the candidate information in all key frames is captured, so as to construct the skeleton topology of each candidate in turn; according to the preset behavior recognition model, the behavior posture recognition is performed on the skeleton topology of each candidate to obtain the behavior information of each candidate in each key frame; according to the behavior information of each candidate in each key frame, the skeleton posture and behavior information of each candidate at the corresponding moment in the non-key frame are predicted and completed, so as to obtain the complete behavior information of each candidate in the test monitoring video data; according to the preset suspicious behavior recognition model, the complete behavior information of each candidate in the test monitoring video data is monitored and identified, so as to obtain the skeleton topology with suspicious behavior; wherein, the suspicious behavior includes all actions unrelated to writing actions, reading actions and head-up actions.

[0064] In this embodiment, by capturing the skeleton topology information of the portraits with the examinee information marked in all key frames, the skeleton feature information corresponding to each examinee is obtained, and then based on the skeleton feature information, the skeleton topology of each examinee can be accurately constructed. Then, through the preset behavior recognition model, the current behavior posture corresponding to each examinee's skeleton topology can be identified, and the current behavior state of the examinee's skeleton topology can also be identified. Among them, the behavior recognition model is obtained through pre-training, and the various posture data of the human skeleton during the examination are input, so that the behavior recognition model can be trained.

[0065] In this embodiment, only all key frame data are captured. Since what is obtained is the behavioral information corresponding to each key frame, it is necessary to predict and complete the skeleton posture and behavioral information of each candidate at the corresponding moment in the non-key frame to ensure that the behavior of each candidate is monitored throughout the entire exam.

[0066] In this embodiment, a preset suspicious behavior recognition model is used to monitor and identify all behavioral information of each examinee in the exam surveillance video data, thereby obtaining a skeleton topology indicating the presence of suspicious behaviors. Suspicious behaviors can include looking around, looking up for a long time, hand movements or positions in a non-writing state, foot shaking, and all other actions unrelated to writing, reading, and looking up. It is understood that the identification of suspicious behaviors can be achieved through the behavior recognition model. By recording the normal examinee's exam behavior and then marking the normal examinee's exam movements, actions not marked as normal examinee's exam can be identified as suspicious behaviors. The behavior recognition model used for identifying suspicious behaviors can also be trained by pre-inputting normal examinee behaviors such as writing, reading, and looking up as training data to obtain a behavior recognition model capable of identifying normal examinee behaviors. The model outputs actions not marked as normal examinee's exam as suspicious behaviors, ultimately training to obtain a suspicious behavior recognition model.

[0067] As a preferred solution of this embodiment, based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the test monitoring video data, specifically:

[0068] The entire behavioral information of each examinee is predicted and completed in turn, so that in the process of predicting and completing the entire behavioral information of each examinee, the examinee frame image that can identify the examinee's complete skeleton topology and behavioral information is extracted from the examinee monitoring video data according to the preset behavior recognition model, and combined with the examinee's behavioral information in each key frame, the skeleton posture and behavior are predicted in turn for the frame moments corresponding to the examinee frame data that cannot identify the examinee's skeleton topology and behavioral information through the behavior prediction model, and then the predicted skeleton posture and behavior information are input into the examinee frame data at the frame moment in turn, until the examinee frame data at all frame moments have the examinee's skeleton posture and behavior information, thereby obtaining the examinee's entire behavioral information in the examinee monitoring video data; until the examinee's entire behavioral information in the examinee monitoring video data is obtained.

[0069] In this embodiment, through the preset behavior recognition model, the complete skeleton topology and behavior information of the examinee in the test monitoring video data can be identified, and then the corresponding test frame image can be extracted, and combined with the behavior information of the examinee in each key frame, so that the skeleton information and behavior information of the two frames before and after the missing frame moment can be predicted through the behavior prediction model, and then the corresponding information of the missing frame moment can be completed and obtained. The skeleton information and behavior information of the examinee in all data frames in the test monitoring video data are obtained.

[0070] Furthermore, if the two frames before and after the missing frame time are too different, there is a situation where the candidate deliberately avoids the camera action, that is, there may be suspicion of abnormal behavior, and the candidate information is directly output.

[0071] It can be understood that by completing the examinee's skeleton information and behavior information in all data frames, accurate monitoring of the examinee's movements and accurate analysis of abnormal behaviors can be ensured throughout the entire examination.

[0072] Step S104: recording the suspicious posture of each skeleton topology in sequence, and obtaining the actual distance between the skeleton topologies with suspicious postures according to the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data.

[0073] In this embodiment, by recording the suspicious posture of each skeleton topology in turn, the skeleton topology with suspicious posture can be marked, and then the coordinate information of the marked skeleton topology in the test monitoring video data can be obtained, so that the actual spacing distance between the skeleton topologies of each suspicious posture can be obtained according to the position and scaling ratio of the camera.

[0074] It is understandable that since candidates cannot move around freely after the exam begins, the position of the table can also be obtained. The candidate's seat number information can be input in advance, and then the candidate's skeleton and seat information can be matched, so that each seat can be obtained more accurately, that is, the actual spacing distance between each skeleton topology.

[0075] Step S105: If the actual interval distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual interval distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, thereby completing the analysis of collective abnormal behavior.

[0076] As a preferred solution of this embodiment, if the actual separation distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual separation distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, specifically:

[0077] If the actual interval distance is less than the preset value, the skeleton topologies of the two suspicious postures corresponding to the actual interval distance are both used as abnormal behavior analysis skeletons; behavior recognition is performed on the abnormal behavior analysis skeletons in turn, and whenever one of the abnormal behavior analysis skeletons is identified to have suspicious behavior, the behavior information of all accomplice skeletons within the preset distance of the abnormal behavior analysis skeleton is obtained, so that whenever the abnormal behavior analysis skeleton is performing a suspicious behavior, if it is identified that an accomplice skeleton is performing a head-raising action and / or a writing action, the number of times the abnormal behavior analysis skeleton and the accomplice skeleton cyclically perform the same action is recorded; if the number is greater than the preset number, the abnormal behavior analysis skeleton and the accomplice skeleton are used as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0078] In this embodiment, when two skeletal topologies with suspicious postures exist with an actual separation distance less than a preset value, this indicates that these two examinees may be engaging in abnormal behavior. For example, these two examinees are examinees A and B. Preferably, the preset value can be set based on the actual examination room and seating arrangements. Thus, the skeletal topologies with the two suspicious postures corresponding to this actual separation distance are used as abnormal behavior analysis skeletons. When suspicious behavior is identified for one of the abnormal behavior analysis skeletons (examinee A or examinee B), behavioral information is obtained for all accomplices of this abnormal behavior analysis skeleton within the preset distance. For example, the abnormal behavior analysis skeleton is examinee A, and all accomplices include examinees A, B, C, D, E, and F. Whenever candidate A performs suspicious behavior, including but not limited to: supporting the chin with the left hand, supporting the chin with the right hand, lifting glasses, crossing legs, etc. (candidates may agree on which answer each action corresponds to before the test. For example, for multiple-choice questions, the options can be matched with each different action respectively, so that the answer can be transmitted after the corresponding action is performed), if one or more of candidate a, candidate b, candidate c, candidate d, candidate e and candidate f are performing the raising head action and / or writing action, a collective abnormal behavior may occur in which candidate A transmits the answer to candidate a, candidate b, candidate c, candidate d, candidate e or candidate f who are performing the raising head action and / or writing action. However, in order to avoid accidental incidents, the collective abnormal behavior is also monitored, and the number of times the abnormal behavior analysis skeleton and the accomplice skeleton repeatedly perform the same action, that is, the collective abnormal behavior, is recorded. If it is greater than the preset number, it is sufficient to indicate the existence of collective abnormal behavior. The abnormal behavior analysis skeleton and the accomplice skeleton are regarded as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0079] As a preferred solution of this embodiment, it also includes:

[0080] If the actual interval distance is greater than or equal to the preset value, the skeleton topology of the suspicious posture will be subjected to behavioral identification in turn; whenever one of the skeleton topologies of the suspicious posture is identified to have suspicious behavior, the behavioral information of all accomplice skeletons within the preset distance of the skeleton topology will be obtained, so that whenever the skeleton topology of the suspicious posture is performing a suspicious behavior, if it is identified that there is an accomplice skeleton performing a head-raising action and / or a writing action, the number of times the skeleton topology of the suspicious posture and the accomplice skeleton perform the same action in a loop will be recorded; if the number is greater than the preset number, the skeleton topology of the suspicious posture and the accomplice skeleton will be regarded as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton will be output.

[0081] It is understandable that candidates who may attempt to participate in collective abnormal behavior may be far away from each other, but can still see each other's actions. Therefore, for the skeleton topology of suspicious postures with a large actual distance between them, it is also necessary to identify and record the suspicious behavior execution and accomplice skeletons, so as to accurately and quickly identify candidates participating in collective abnormal behavior.

[0082] It can be understood that the embodiments of the present invention are mainly aimed at the behavior recognition of collective abnormal behavior, especially for the collective abnormal behavior situation in which the corresponding actions are performed during the exam after the action answers are agreed upon before the exam, thereby realizing the answer transmission. The existing abnormal behavior recognition cannot be targeted at abnormal behavior recognition in the scenario of collective abnormal behavior, resulting in the action recognition of a single examinee cannot accurately reflect the situation of all examinees in the entire examination room. The present invention makes it possible to focus on the behavior of examinees in the entire examination room, and avoid the occurrence of collective abnormal behavior through skeleton behavior action recognition, while also improving the accuracy and efficiency of abnormal behavior detection.

[0083] The implementation of the above embodiment has the following effects:

[0084] The technical solution of the present invention obtains the examination monitoring video data of the examination room, and then selects key frames after frame division and preprocessing, so as to mark the candidate information corresponding to the portraits identified in sequence, and capture the skeleton topology information of the marked portraits to construct the skeleton topology of each candidate, so as to continuously monitor the skeleton topology and examination monitoring video data through a preset suspicious behavior recognition model, and at the same time combine the actual interval distance between the skeleton topologies of each suspicious posture to accurately and quickly determine whether the skeleton topologies of two suspicious postures between the actual interval distance are abnormal behavior skeletons, and then quickly output information on the candidates corresponding to the collective abnormal behavior. The analysis of skeleton behavior can avoid the blind spots of machine vision recognition and the problem of inaccurate machine vision behavior detection. The accuracy of collective abnormal behavior detection is improved by combining the detection of actual interval distance.

[0085] Example 2

[0086] Please refer to FIG. 2 . The present invention further provides a collective abnormal behavior analysis device based on dynamic topology, comprising: an acquisition module 201 , a recognition module 202 , a skeleton module 203 , a distance module 204 and an analysis module 205 .

[0087] The acquisition module 201 is used to acquire the examination monitoring video data of the examination room, and perform frame division and preprocessing on the video data; wherein the video data after frame division includes a plurality of examination frame data;

[0088] The recognition module 202 is used to select key frames from the test frame data, perform portrait recognition on the key frames in turn, compare and analyze the recognized portraits with the preset candidate information database, and mark the candidate information obtained from the analysis in the key frames.

[0089] The skeleton module 203 is used to capture the skeleton topology information of the annotated portrait, thereby constructing the skeleton topology of each examinee respectively, and continuously monitoring the skeleton topology and the examination monitoring video data according to the preset suspicious behavior recognition model;

[0090] The distance module 204 is used to record the suspicious posture of each skeleton topology in sequence, and obtain the actual distance between the skeleton topologies with suspicious postures according to the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data.

[0091] The analysis module 205 is used to use the skeleton topology of the two suspicious postures corresponding to the actual interval distance as the abnormal behavior skeleton if the actual interval distance is less than the preset value, and output the candidate information marked in the abnormal behavior skeleton, thereby completing the analysis of collective abnormal behavior.

[0092] As a preferred solution, the acquisition of the examination monitoring video data of the examination room and the frame division and preprocessing of the video data are specifically as follows:

[0093] Obtain examination monitoring video data of the examination room through the camera;

[0094] Dividing the test monitoring video data into frames to obtain image frames;

[0095] Gaussian filtering, image denoising and image enhancement are performed on the image frame to obtain test frame data.

[0096] As a preferred solution, the key frames are selected from the test frame data, and portrait recognition is performed on the key frames in sequence, and the recognized portraits are compared and analyzed with the preset candidate information database, and the candidate information obtained from the analysis is marked in the key frames, specifically:

[0097] Performing portrait recognition on all test frame data according to the portrait recognition model, thereby selecting key frames in which portraits are not obscured from the test frame data, and obtaining portrait information after portrait recognition of the key frames; wherein all portraits can be recognized in each key frame;

[0098] The identified portrait information is compared and analyzed with the preset candidate information database, and the portrait information identified in the key frames is marked with the candidate information in turn.

[0099] As a preferred solution, the skeleton topology information of the annotated portrait is captured to construct the skeleton topology of each examinee respectively, and the skeleton topology and examination monitoring video data are continuously monitored according to a preset suspicious behavior recognition model, specifically:

[0100] Capture the skeleton topology information of the portraits with the candidate information marked in all key frames, and then construct the skeleton topology of each candidate in turn;

[0101] Based on the preset behavior recognition model, the behavior posture of each candidate is recognized on the skeleton topology to obtain the behavior information of each candidate in each key frame;

[0102] Based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the exam monitoring video data;

[0103] Based on a preset suspicious behavior identification model, all behavioral information of each candidate in the exam surveillance video data is monitored and identified to obtain a skeleton topology where suspicious behaviors exist; wherein the suspicious behaviors include all actions unrelated to writing, reading, and raising the head.

[0104] As a preferred solution, based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the test monitoring video data, specifically:

[0105] The entire behavioral information of each examinee is predicted and completed in sequence, so that in the process of predicting and completing the entire behavioral information of each examinee, the examination frame image that can identify the examinee's complete skeleton topology and behavioral information is extracted from the examination monitoring video data according to the preset behavior recognition model, and combined with the examinee's behavioral information in each key frame, the skeleton posture and behavior are predicted in sequence for the frame moments corresponding to the examination frame data that cannot identify the examinee's skeleton topology and behavioral information through the behavior prediction model, and then the predicted skeleton posture and behavior information are input into the examination frame data at the frame moment in sequence, until the examination frame data at all frame moments have the examinee's skeleton posture and behavior information, thereby obtaining the examinee's entire behavioral information in the examination monitoring video data;

[0106] Until all the behavioral information of all candidates in the examination monitoring video data is obtained.

[0107] As a preferred solution, if the actual separation distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual separation distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, specifically:

[0108] If the actual separation distance is less than the preset value, the skeleton topologies of the two suspicious postures corresponding to the actual separation distance are both used as abnormal behavior analysis skeletons;

[0109] Performing behavior recognition on the abnormal behavior analysis skeletons in sequence, and whenever a suspicious behavior is identified in one of the abnormal behavior analysis skeletons, obtaining behavior information of all accomplice skeletons within a preset distance of the abnormal behavior analysis skeleton, so that whenever the abnormal behavior analysis skeleton performs a suspicious behavior, if an accomplice skeleton is identified as performing a head-raising action and / or a writing action, then recording the number of times the abnormal behavior analysis skeleton and the accomplice skeleton cyclically perform the same action;

[0110] If the number is greater than the preset number, the abnormal behavior analysis skeleton and the accomplice skeleton are regarded as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0111] As a preferred solution, it also includes:

[0112] If the actual separation distance is greater than or equal to the preset value, the skeleton topology of the suspicious posture is sequentially subjected to behavior recognition;

[0113] Whenever a suspicious behavior is detected in a skeleton topology with a suspicious posture, the behavior information of all accomplice skeletons within a preset distance of the skeleton topology is obtained. Therefore, whenever the skeleton topology with a suspicious posture is performing a suspicious behavior, if an accomplice skeleton is identified as performing a head-raising action and / or a writing action, the number of times the skeleton topology with a suspicious posture and the accomplice skeleton repeatedly perform the same action is recorded.

[0114] If the number is greater than the preset number, the skeleton topology of the suspicious posture and the accomplice skeleton are used as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

[0115] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0116] The implementation of the above embodiment has the following effects:

[0117] The technical solution of the present invention obtains the examination monitoring video data of the examination room, and then selects key frames after frame division and preprocessing, so as to mark the candidate information corresponding to the portraits identified in sequence, and capture the skeleton topology information of the marked portraits to construct the skeleton topology of each candidate, so as to continuously monitor the skeleton topology and examination monitoring video data through a preset suspicious behavior recognition model, and at the same time combine the actual interval distance between the skeleton topologies of each suspicious posture to accurately and quickly determine whether the skeleton topologies of two suspicious postures between the actual interval distance are abnormal behavior skeletons, and then quickly output information on the candidates corresponding to the collective abnormal behavior. The analysis of skeleton behavior can avoid the blind spots of machine vision recognition and the problem of inaccurate machine vision behavior detection. The accuracy of collective abnormal behavior detection is improved by combining the detection of actual interval distance.

[0118] Example 3

[0119] Accordingly, the present invention also provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the collective abnormal behavior analysis method based on dynamic topology as described in any one of the above embodiments.

[0120] The terminal device of this embodiment includes: a processor, a memory, and a computer program or computer instructions stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the first embodiment described above, such as steps S101 to S105 shown in FIG1 . Alternatively, when the processor executes the computer program, it implements the functions of the modules / units of the apparatus embodiment described above, such as analysis module 205.

[0121] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the analysis module 205 is used to use the skeleton topology of the two suspicious postures corresponding to the actual separation distance as an abnormal behavior skeleton if the actual separation distance is less than a preset value, and output the candidate information marked in the abnormal behavior skeleton, thereby completing the analysis of collective abnormal behavior.

[0122] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or cloud server. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a terminal device and does not limit the terminal device. The terminal device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, and the like.

[0123] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the terminal device and connects various parts of the entire terminal device using various interfaces and lines.

[0124] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, etc.; the data storage area may store data created based on the use of the mobile terminal, etc. In addition, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0125] If the module / unit integrated into the terminal device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content of the computer-readable medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electric carrier signals and telecommunication signals.

[0126] Example 4

[0127] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the collective abnormal behavior analysis method based on dynamic topology as described in any one of the above embodiments.

[0128] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for analyzing collective abnormal behavior based on dynamic topology, characterized in that: include: Acquire examination monitoring video data of an examination room, divide the video data into frames to obtain image frames, and preprocess the image frames to obtain examination frame data; wherein the video data after frame division includes a plurality of unpreprocessed examination frame data; Select key frames from the test frame data, perform portrait recognition on the key frames in turn, compare and analyze the recognized portraits with the preset candidate information database, and mark the candidate information obtained from the analysis in the key frames; The skeleton topology information of the annotated portraits is captured, so as to construct the skeleton topology of each candidate respectively, and the skeleton topology and the test monitoring video data are continuously monitored according to the preset suspicious behavior recognition model; the skeleton topology information of the portraits with the candidate information annotated in all key frames is captured, so as to construct the skeleton topology of each candidate in turn; according to the preset behavior recognition model, the behavior posture recognition is performed on the skeleton topology of each candidate to obtain the behavior information of each candidate in each key frame; according to the behavior information of each candidate in each key frame, the skeleton posture and behavior information of each candidate at the corresponding moment in the non-key frame are predicted and completed, so as to obtain the full behavior information of each candidate in the test monitoring video data; according to the preset suspicious behavior recognition model, the full behavior information of each candidate in the test monitoring video data is monitored and identified, so as to obtain the skeleton topology with suspicious behavior; wherein, the suspicious behavior includes all actions unrelated to writing, reading and raising the head; The suspicious posture of each skeleton topology is recorded in turn, and the actual distance between the skeleton topologies with suspicious postures is obtained based on the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data; If the actual interval distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual interval distance is used as the abnormal behavior skeleton, and the candidate information marked in the abnormal behavior skeleton is output, thereby completing the analysis of the collective abnormal behavior; if the actual interval distance is less than the preset value, the skeleton topology of the two suspicious postures corresponding to the actual interval distance is used as the abnormal behavior analysis skeleton; the abnormal behavior analysis skeletons are sequentially subjected to behavior recognition, and whenever one of the abnormal behavior analysis skeletons is identified to have a suspicious behavior, the behavior information of all accomplice skeletons within the preset distance of the abnormal behavior analysis skeleton is obtained, so that whenever the abnormal behavior analysis skeleton is performing a suspicious behavior, if it is identified that an accomplice skeleton is performing a head-raising action and / or a writing action, the number of times the abnormal behavior analysis skeleton and the accomplice skeleton cyclically perform the same action is recorded; if the number is greater than the preset number, the abnormal behavior analysis skeleton and the accomplice skeleton are used as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

2. The method for analyzing collective abnormal behavior based on dynamic topology according to claim 1, characterized in that: The acquisition of the examination monitoring video data of the examination room and the frame division and preprocessing of the video data are specifically as follows: Obtain examination surveillance video data of the examination room through the camera; Dividing the test monitoring video data into frames to obtain image frames; Gaussian filtering, image denoising and image enhancement are performed on the image frame to obtain test frame data.

3. The method for analyzing collective abnormal behavior based on dynamic topology according to claim 2, characterized in that: The key frames are selected from the test frame data, and the key frames are sequentially subjected to portrait recognition, and the recognized portraits are compared and analyzed with the preset candidate information database, and the candidate information obtained by the analysis is marked in the key frames, specifically: Performing portrait recognition on all test frame data according to the portrait recognition model, thereby selecting key frames in which portraits are not obscured from the test frame data, and obtaining portrait information after portrait recognition of the key frames; wherein all portraits can be recognized in each key frame; The identified portrait information is compared and analyzed with the preset candidate information database, and the portrait information identified in the key frames is marked with the candidate information in turn.

4. The method for analyzing collective abnormal behavior based on dynamic topology according to claim 3, characterized in that: Based on the behavioral information of each examinee in each key frame, the skeleton posture and behavioral information of each examinee at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining the complete behavioral information of each examinee in the test monitoring video data, specifically: The entire behavioral information of each examinee is predicted and completed in sequence, so that in the process of predicting and completing the entire behavioral information of each examinee, the examination frame image that can identify the examinee's complete skeleton topology and behavioral information is extracted from the examination monitoring video data according to the preset behavior recognition model, and combined with the examinee's behavioral information in each key frame, the skeleton posture and behavior are predicted in sequence for the frame moments corresponding to the examination frame data that cannot identify the examinee's skeleton topology and behavioral information through the behavior prediction model, and then the predicted skeleton posture and behavior information are input into the examination frame data at the frame moment in sequence, until the examination frame data at all frame moments have the examinee's skeleton posture and behavior information, thereby obtaining the examinee's entire behavioral information in the examination monitoring video data; Until all the behavioral information of all candidates in the examination monitoring video data is obtained.

5. The method for analyzing collective abnormal behavior based on dynamic topology according to claim 1, characterized in that: Also includes: If the actual separation distance is greater than or equal to the preset value, the skeleton topology of the suspicious posture is sequentially subjected to behavior recognition; Whenever a suspicious behavior is detected in a skeleton topology with a suspicious posture, the behavior information of all accomplice skeletons within a preset distance of the skeleton topology is obtained. Therefore, whenever the skeleton topology with a suspicious posture is performing a suspicious behavior, if an accomplice skeleton is identified as performing a head-raising action and / or a writing action, the number of times the skeleton topology with a suspicious posture and the accomplice skeleton repeatedly perform the same action is recorded. If the number is greater than the preset number, the skeleton topology of the suspicious posture and the accomplice skeleton are used as abnormal behavior skeletons, and the candidate information marked in the abnormal behavior skeleton is output.

6. A collective abnormal behavior analysis device based on dynamic topology, characterized in that: include: Acquisition module, recognition module, skeleton module, distance module and analysis module; The acquisition module is used to acquire the examination monitoring video data of the examination room, and to perform frame division on the video data to obtain image frames, and to perform preprocessing on the image frames to obtain examination frame data; wherein the video data after frame division includes a plurality of unpreprocessed examination frame data; The recognition module is used to select key frames from the test frame data, perform portrait recognition on the key frames in sequence, compare and analyze the recognized portraits with a preset candidate information database, and mark the candidate information obtained from the analysis in the key frames; The skeleton module is used to capture the skeleton topology information of the annotated portraits, thereby constructing the skeleton topology of each candidate respectively, and continuously monitoring the skeleton topology and test monitoring video data according to the preset suspicious behavior recognition model; the skeleton topology information of the portraits with the candidate information marked in all key frames is captured, thereby sequentially constructing the skeleton topology of each candidate; according to the preset behavior recognition model, the behavior posture recognition is performed on the skeleton topology of each candidate to obtain the behavior information of each candidate in each key frame; according to the behavior information of each candidate in each key frame, the skeleton posture and behavior information of each candidate at the corresponding moment in the non-key frame are predicted and completed, thereby obtaining all the behavior information of each candidate in the test monitoring video data; according to the preset suspicious behavior recognition model, all the behavior information of each candidate in the test monitoring video data is monitored and identified, thereby obtaining the skeleton topology with suspicious behavior; wherein, the suspicious behavior includes all actions unrelated to writing actions, reading actions and head-up actions; The distance module is used to record the suspicious posture of each skeleton topology in turn, and obtain the actual distance between the skeleton topologies with suspicious postures based on the coordinate information of the skeleton topologies with suspicious postures in the test monitoring video data; The analysis module is used to, if the actual interval distance is less than a preset value, use the skeleton topology of the two suspicious postures corresponding to the actual interval distance as the abnormal behavior skeleton, and output the candidate information marked in the abnormal behavior skeleton, thereby completing the analysis of collective abnormal behavior; if the actual interval distance is less than the preset value, use the skeleton topology of the two suspicious postures corresponding to the actual interval distance as the abnormal behavior analysis skeleton; perform behavior recognition on the abnormal behavior analysis skeletons in turn, and whenever one of the abnormal behavior analysis skeletons is identified to have suspicious behavior, obtain the behavior information of all accomplice skeletons within the preset distance of the abnormal behavior analysis skeleton, so that whenever the abnormal behavior analysis skeleton is performing a suspicious behavior, if it is identified that an accomplice skeleton is performing a head-raising action and / or a writing action, then record the number of times the abnormal behavior analysis skeleton and the accomplice skeleton cyclically perform the same action; if the number is greater than the preset number, use the abnormal behavior analysis skeleton and the accomplice skeleton as abnormal behavior skeletons, and output the candidate information marked in the abnormal behavior skeleton.

7. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for collective abnormal behavior analysis based on dynamic topology according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the collective abnormal behavior analysis method based on dynamic topology according to any one of claims 1 to 5.

Images