Abnormity detection method, device and equipment for double-recording video and storage medium

Abstract

The invention relates to the field of image processing, and provides a double-recording video anomaly detection method, which comprises the steps of determining at least one quality inspection point of a to-be-detected double-recording video based on a target detection frequency; in the to-be-detected double-recording video, acquiring a video frame corresponding to the quality inspection point as a to-be-detected video frame; determining a corresponding target detection model based on the scene content corresponding to the to-be-detected video frame; and performing abnormal behavior detection on the to-be-detected video frame based on the target detection model. And based on the scene content of each to-be-detected video frame, the corresponding deep learning model is adopted as a target detection model to perform abnormal behavior detection of each to-be-detected video frame, so that the deep learning model is adopted to identify and judge the user behavior of the insurance double-recording scene in the to-be-detected double-recording video, manual auditing is not needed, and the user experience is improved. The labor cost is reduced, and the auditing efficiency and accuracy of the double-recording video are improved.

Description

Technical Field

[0001] The present invention relates to the fields of image processing and financial technology, and in particular to a method, device, computer equipment and computer-readable storage medium for detecting anomalies in dual-recording videos. Background Art

[0002] When users handle insurance business online, they need to double-record the video of signing the contract. The recorded video contains multiple voice commands to inform users of precautions, let them fill in relevant information, sign and confirm the contract, etc. At the same time, the user's affirmative or negative answer needs to be recorded to retain the basis for determining whether the contract is effective or not. At present, in order to prevent abnormal behaviors such as answering questions on behalf of others, bystanders entering the camera, and agents violating regulations in the recorded video, the double-recorded video can only be reviewed manually. However, manual review of double-recorded videos is not only intensive and easy to reduce the accuracy of the review results, but also has low review efficiency. Summary of the invention

[0003] The main purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for detecting anomalies in dual-recorded videos, aiming to improve the technical problem of improving the review efficiency and accuracy of review results of dual-recorded videos.

[0004] To achieve the above object, the present invention provides a method for detecting anomalies in dual-recording videos, wherein the frame rate improvement method comprises the following steps:

[0005] Based on the target detection frequency, determining at least one quality inspection point of the dual-recorded video to be detected;

[0006] In the dual-recorded video to be detected, obtaining the video frame corresponding to the quality inspection point as the video frame to be detected;

[0007] Determining a corresponding target detection model based on the scene content corresponding to the video frame to be detected;

[0008] Based on the target detection model, abnormal behavior detection is performed on the video frame to be detected.

[0009] In addition, to achieve the above-mentioned purpose, the present invention further provides a dual-recording video anomaly detection device, the dual-recording video anomaly detection device comprising:

[0010] A quality inspection point determination module, used to determine at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency;

[0011] A video frame acquisition module, used to acquire the video frame corresponding to the quality inspection point in the dual-recorded video to be detected as the video frame to be detected;

[0012] A detection model determination module, used to determine a corresponding target detection model based on the scene content corresponding to the video frame to be detected;

[0013] The abnormal behavior detection module is used to perform abnormal behavior detection on the video frame to be detected based on the target detection model.

[0014] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a processor, a memory, and a dual-recording video anomaly detection program stored in the memory and executable by the processor, wherein when the dual-recording video anomaly detection program is executed by the processor, the steps of the dual-recording video anomaly detection method as described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a dual-recording video anomaly detection program is stored, wherein when the dual-recording video anomaly detection program is executed by a processor, the steps of the dual-recording video anomaly detection method as described above are implemented.

[0016] The present application provides a method for detecting anomalies in dual-recording videos, wherein the method determines at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency; in the dual-recording video to be detected, obtain the video frame corresponding to the quality inspection point as the video frame to be detected; determine the corresponding target detection model based on the scene content corresponding to the video frame to be detected; and perform abnormal behavior detection on the video frame to be detected based on the target detection model. In the above manner, the present application first determines the quality inspection point based on the target detection frequency. The dual-recording with high detection frequency has more quality inspection points. Thus, according to different detection standards of dual-recording videos, the number of video frames to be detected is determined, so as to adapt to the detection requirements of different dual-recording videos; then obtain the video frames corresponding to each quality inspection point as the video frame to be detected, and based on the scene content of each video frame to be detected, use the corresponding deep learning model as the target detection model to perform abnormal behavior detection on each video frame to be detected, so as to use the deep learning model to identify and judge the user behavior of the insurance dual-recording scene in the dual-recording video to be detected, without manual review, which not only reduces the labor cost, but also improves the review efficiency and accuracy of the dual-recording video. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a first embodiment of the method for detecting anomalies in dual-recording videos of the present invention;

[0018] Figure 2 It is a flow chart of a second embodiment of the method for detecting anomalies in dual-recording videos of the present invention;

[0019] Figure 3It is a flowchart of a third embodiment of the method for detecting anomalies in dual-recording videos of the present invention;

[0020] Figure 4 It is a schematic diagram of an abnormality detection device for dual-recording videos of the present invention;

[0021] Figure 5 The figure is a schematic diagram of the hardware structure of a computer device involved in an embodiment of the present invention.

[0022] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0025] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0026] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0027] The dual-recording video anomaly detection method involved in the embodiment of the present invention is mainly applied to computer equipment. The dual-recording video anomaly detection generation device can be a PC, a portable computer, a mobile terminal, or other device with display and processing functions.

[0028] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0029] Reference Figure 1 , Figure 1Schematic diagram of the process of the first embodiment of the dual-recording video anomaly detection method of the present invention.

[0030] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting anomalies in dual-recording videos, and the method for detecting anomalies in dual-recording videos includes steps S10 to S40.

[0031] In this embodiment, the dual-recording video anomaly detection method includes the following steps:

[0032] Step S10, determining at least one quality inspection point of the dual-recorded video to be detected based on the target detection frequency;

[0033] In the insurance field, when users need to handle business online, they are usually required to double-record a video of signing a contract. The video will contain many voice commands broadcast by a voice robot to inform users of precautions, guide users to fill in information, sign and confirm the contract, etc. It also needs to include the user's affirmative or negative answers to the voice robot's questions, which are retained as the basis for determining whether the contract is effective or not. The double-recorded video defined in this method includes audio and video recording, and usually more than one person needs to appear in the video (for example: insurance business processing usually requires the insured and the agent to appear in the same frame at all times, and the insured is required to appear in other stages). In such a video with multiple people appearing at the same time, there may be abnormal behaviors such as the agent answering on behalf of the insured (answering on behalf of the insured), the insured denying the terms (denying the intention), the other person appearing in the camera, and the insured or the agent appearing in the camera in violation of the rules (same frame anomaly). On the business side, these abnormal behaviors can usually only be determined by manually reviewing the video, but the average length of a video is about 20 minutes. This high-intensity and high-repetitive work is a severe test for the accuracy and efficiency of the review.

[0034] In this embodiment, quality inspection points are first determined based on the target detection frequency. The higher the detection frequency of the dual recording, the more quality inspection points there are. Therefore, the number of video frames to be detected is determined according to different detection standards of the dual recording video, so as to adapt to the detection requirements of different dual recording videos; then, the video frames corresponding to each quality inspection point are obtained as the video frames to be detected, and based on the scene content of each video frame to be detected, the corresponding deep learning model is used as the target detection model to perform abnormal behavior detection on each video frame to be detected, so as to use the deep learning model to identify and judge the user behavior of the insurance dual recording scenario in the dual recording video to be detected, without the need for manual review, which not only reduces labor costs but also improves the review efficiency and accuracy of the dual recording video.

[0035] Specifically, the detection frequency is the number of frames extracted in a preset time period, such as 1 frame in 3 seconds or 1 frame in 1 second, or 10 frames in 1 minute, etc. The quality inspection point is the number of video frames corresponding to the target detection frequency in the dual-recorded video to be detected. For example, if 1 frame is extracted in 1 second, the number of video frames corresponding to each second, starting from the start time 0 second in the dual-recorded video to be detected, 1S, 2S, 3S, etc., are all quality inspection points.

[0036] In this embodiment, based on the video type of the double-recorded video to be detected, the target detection frequency of the double-recorded video to be detected is determined. For example, when the double-recorded video to be detected is a video being recorded (i.e., being recorded), the corresponding target detection frequency (i.e., the first detection frequency) is relatively low, such as being set to extract 1 frame every 3 seconds; when the double-recorded video to be detected is a video after recording (i.e., recording is completed), the corresponding target detection frequency (i.e., the second detection frequency) is relatively high, such as being set to extract 1 frame every 1 second. Thus, it is ensured that the user can be prompted in real time and smoothly during recording, further improving the user experience; and the double-recorded video after recording is monitored for abnormal risk behaviors at a high frequency and in an all-round manner. Specifically, quality inspection points can also be determined according to the process nodes to which the double-recorded video to be detected belongs, such as the same-frame appearance detection link in the double-recorded video to be detected, or the link in the double-recorded video to be detected where the insured needs to perform real-name authentication according to the voice instructions broadcast by the voice robot, etc., and high-frequency frame extraction is performed; video quality detection links or person-document comparison links, etc., are performed with low frequency frame extraction.

[0037] It is understandable that each link in the dual-recording video and its corresponding frame extraction frequency can be set in advance.

[0038] Step S20, obtaining a video frame corresponding to the quality inspection point in the dual-recorded video to be detected as a video frame to be detected;

[0039] In this embodiment, video frames corresponding to all quality inspection points are obtained (screenshot or read) from the dual-recorded video to be detected as the video frames to be detected.

[0040] Furthermore, a deduplication operation is performed on the video frames to be detected to remove duplicate images in the video frames to be detected, thereby further improving the review efficiency of the dual-recorded videos.

[0041] Furthermore, a video frame adding function is provided to the user, and upon receiving a video frame adding instruction triggered by a user operation, the video frame determined by the user selection operation is obtained, or the video frame corresponding to the user selection link is obtained, and added to the video frame to be detected.

[0042] Step S30, determining a corresponding target detection model based on the scene content corresponding to the video frame to be detected;

[0043] In this embodiment, in order to improve the accuracy of the dual-recording video review results, the deep learning model is trained in advance using video frames corresponding to different scene contents as samples to obtain detection models for different scene contents.

[0044] The process link to which the video frame to be detected belongs is determined by the scene content corresponding to each video frame to be detected, and the detection model corresponding to the process link is determined by the preset model mapping table as the target detection model corresponding to the video frame to be detected.

[0045] In a specific embodiment, the object type included in each video frame to be detected can also be determined through the scene content corresponding to the video frame to be detected, and based on the object type included in the video frame, a detection model that can detect the object type can be determined as the target detection model corresponding to the video frame to be detected.

[0046] Furthermore, the determining of the corresponding target detection model based on the scene content corresponding to the video frame to be detected specifically includes:

[0047] If the scene content is the policyholder confirmation process, a face detection model, a voice detection model, a liveness detection model and a document detection model are determined as the target detection model.

[0048] Specifically, the detection model includes:

[0049] 1. The same-frame detection model is used to read the video frame to be detected after a fixed interval and determine whether the policyholder and agent in the video frame to be detected remain in the camera. If there is an abnormal same-frame in the scene being recorded, the user will be prompted to correct the behavior immediately; if there is an abnormal same-frame in the scene after recording, the abnormal time point will be output and the dual-recorded video will be determined as a high-risk video.

[0050] It can be understood that the same-frame detection model can be applied to one or more process nodes that require the policyholder and the agent to remain in the same frame.

[0051] Furthermore, the same-frame detection model includes:

[0052] A face detection sub-model, used to detect face images in videos or video frames;

[0053] The face recognition sub-model is used to recognize the face image in the video or video frame and determine the identity information corresponding to the face;

[0054] The face quality judgment sub-model is used for the quality of face images in videos or video frames, such as clarity or blurriness.

[0055] The face calibration sub-model is used to calibrate the face image in the video or video frame when the face quality judgment sub-model determines that the quality of the face image in the video or video frame is blurred, so as to improve the quality of the face image in the video or video frame.

[0056] 2. The signature behavior recognition model is used to identify the signature behavior of the insured, where the signature behavior refers to the behavior of the insured being asked to use the mobile phone to sign and confirm the contract after the specific words broadcast by the voice robot. If the scene in the recording detects that the insured is not the signature or the signature behavior is not detected, the user will be immediately prompted to regulate the behavior and restart the recording of the signature stage; if the scene after the recording detects that the insured is not the signature or the signature behavior is not detected, the abnormal time period will be output and determined as a high-risk video.

[0057] Furthermore, the signature behavior recognition model includes:

[0058] Face detection sub-model;

[0059] Face recognition sub-model;

[0060] Hand detection sub-model;

[0061] The handwriting detection sub-model is used to perform handwriting detection on the signature detected in the signature behavior recognition model.

[0062] 3. The person-document comparison model is used to check whether the information of the insured and his / her ID card, and the agent and his / her ID card are consistent. Among them, the presentation of ID card means that the voice robot requires the insured and the agent to present their ID cards in turn after reading specific words, and judges whether the current presenter is consistent with the presented ID card information. If there is any inconsistency in the recorded scene, the user will be immediately prompted to regulate the behavior and restart the recording of the ID card presentation stage; if there is any inconsistency in the recorded scene, the abnormal time period will be output and it will be judged as a high-risk video.

[0063] The human-evidence comparison model includes:

[0064] Face detection sub-model;

[0065] Face recognition sub-model;

[0066] The ID detection sub-model is used to detect the ID presented by the person in the video or video frame;

[0067] The text recognition (OCR) sub-model is used to recognize the text in the document.

[0068] The willingness verification model is used to determine the insured's willingness to answer. The willingness to answer refers to the insured's answer to questions with standard answer words (for example, yes, no, agree, disagree, clear, unclear, etc.) after the voice robot broadcasts a specific line of speech, and determines whether the current answer is a positive or negative willingness. If there is a negative willingness in the recorded scene, the user will be immediately prompted to regulate the behavior and restart the recording of the inquiry stage; if there is a negative willingness in the recorded scene, the abnormal time period will be output and determined as a high-risk video.

[0069] Furthermore, the intention verification model includes:

[0070] Voice Activity Detection (VAD) sub-model;

[0071] ASR sub-model;

[0072] Text matching sub-model;

[0073] Noise filtering sub-model;

[0074] 4. Contract text recognition model, used to detect and identify the contract confirmed by the policyholder and the agent. The contract is a contract that the voice robot will ask the agent to display the policyholder's signature and confirmation in front of the camera for 3 seconds after reading a specific line. It also determines whether there is a specific contract text in the current frame. If the specific contract text is not detected in the recorded scene, the user is immediately prompted to regulate the behavior and the contract is displayed again; if the specific contract text is not detected in the recorded scene, the abnormal time period is output and it is determined as a high-risk video.

[0075] The contract text recognition model includes:

[0076] A paper detection sub-model, used to detect contract paper in a video or video frame;

[0077] Text recognition (OCR) sub-model;

[0078] Text matching sub-model.

[0079] Step S40: performing abnormal behavior detection on the video frame to be detected based on the target detection model.

[0080] In this embodiment, each target detection model determined by each video frame to be detected is used to perform abnormal behavior detection on each video frame to be detected.

[0081] This embodiment provides a method for detecting anomalies in dual-recording videos, which method determines at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency; in the dual-recording video to be detected, obtain the video frame corresponding to the quality inspection point as the video frame to be detected; determine the corresponding target detection model based on the scene content corresponding to the video frame to be detected; and perform abnormal behavior detection on the video frame to be detected based on the target detection model. In the above manner, the present application first determines the quality inspection point based on the target detection frequency. The dual-recording with a high detection frequency has more quality inspection points. Thus, according to different detection standards of the dual-recording video, the number of video frames to be detected is determined, so as to adapt to the detection requirements of different dual-recording videos; then obtain the video frames corresponding to each quality inspection point as the video frame to be detected, and based on the scene content of each video frame to be detected, use the corresponding deep learning model as the target detection model to perform abnormal behavior detection on each video frame to be detected, so as to use the deep learning model to identify and judge the user behavior of the insurance dual-recording scene in the dual-recording video to be detected, without manual review, which not only reduces the labor cost, but also improves the review efficiency and accuracy of the dual-recording video.

[0082] Reference Figure 2 , Figure 2 Schematic diagram of the flow of the second embodiment of the method for detecting anomalies in dual-recording videos of the present invention.

[0083] Based on the above Figure 1 In the embodiment shown, in this embodiment, before step S10, the following steps are further included:

[0084] Step S01, obtaining the video type of the dual-recording video to be detected, and determining whether the video type is a recording video or a post-recording video;

[0085] Step S02: when the video type is a recorded video, determining a first detection frequency as the target detection frequency.

[0086] Furthermore, after step S01, the method further includes:

[0087] When the video type is a recorded video, a second detection frequency is determined as the target detection frequency, wherein the second detection frequency is higher than the first detection frequency.

[0088] In this embodiment, the target detection frequency of the dual-recording video to be detected is determined based on the video type of the dual-recording video to be detected. If the dual-recording video to be detected is a video in recording, the corresponding target detection frequency (i.e., the first detection frequency) is relatively low, such as being set to draw 1 frame every 3 seconds; if the dual-recording video to be detected is a video after recording, the corresponding target detection frequency (i.e., the second detection frequency) is relatively high, such as being set to draw 1 frame every 1 second. In this way, it is ensured that the user can be prompted in real time and smoothly during recording, further improving the user experience; and high-frequency and comprehensive abnormal risk behavior monitoring is performed on the dual-recording video after recording.

[0089] Further, the determining at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency specifically includes:

[0090] Based on the video content and voice content of the dual-recording video to be detected, determining each process node of the dual-recording video to be detected;

[0091] Based on the target detection frequency, at least one time point is determined in the video segment corresponding to each process node as the quality inspection point.

[0092] Specifically, quality inspection points can also be determined based on the process nodes to which the dual-recorded video to be tested belongs, such as the same-frame appearance detection link in the dual-recorded video to be tested, or the link in the dual-recorded video to be tested that requires the insured to perform real-name authentication according to the voice instructions broadcast by the voice robot, etc., for high-frequency frame extraction; low-frequency frame extraction is performed in the video quality detection link or the person-document comparison link, etc.

[0093] It is understandable that each link in the dual-recording video and its corresponding frame extraction frequency can be set in advance.

[0094] Through the above method, the present application first determines the quality inspection points based on the target detection frequency. For dual recordings with high detection frequency, there are many quality inspection points, that is, the detection frequency is low for the video being recorded, so as to reduce interference and improve the fluency of the video being recorded. For the video after recording, the detection frequency is high, so as to conduct a comprehensive detection of abnormal behavior in the video after recording, and improve the review accuracy of the dual recording video. Therefore, according to the different detection standards of the dual recording video, the number of video frames to be detected is determined to adapt to the detection requirements of different dual recording videos; then the video frames corresponding to each quality inspection point are obtained as the video frames to be detected, and based on the scene content of each video frame to be detected, the corresponding deep learning model is used as the target detection model to perform abnormal behavior detection on each video frame to be detected, so as to use the deep learning model to identify and judge the user behavior of the insurance dual recording scene in the dual recording video to be detected, without manual review, which not only reduces the labor cost, but also improves the review efficiency and accuracy of the dual recording video.

[0095] Reference Figure 3 , Figure 34 is a flow chart of a third embodiment of the method for detecting anomalies in dual-recording videos according to the present invention.

[0096] Based on the above Figure 1 In the embodiment shown, in this embodiment, the dual-recording video anomaly detection method further includes:

[0097] Step S50, based on the first detection model, determining whether the dual-recording video to be detected contains a preset scene content; or

[0098] Based on the banned word detection model, it is determined whether the dual-recording video to be detected contains any illegal words.

[0099] In this embodiment, after each video frame to be detected is detected by using the target detection model corresponding to the scene content of each video frame to be detected, the entire process of the double-recording video to be detected is further subjected to preset scene detection or banned word detection by using the first detection model or the banned word detection model and other abnormal behavior detection models that require full detection of the dual-recording video.

[0100] Specifically, the first detection model is used to read the video frame to be detected after a fixed time interval, and determine whether there is a preset scene content in the video frame to be detected. If there is a preset scene content in the scene being recorded, the user is immediately prompted to regulate behavior and restart dual recording; if there is a preset scene content in the scene after recording, the abnormal time point is output and it is judged as a high-risk video.

[0101] The prohibited language detection model is used to obtain the audio segment after the voice robot's voice broadcast, and the voice content of the current time period, and then determine whether there are any illegal words in the conversation through text matching, tone matching and other text matching methods. If there are illegal words in the scene during recording, the user will be immediately prompted to regulate the behavior and restart the dual recording; if there are illegal words in the scene after recording, the abnormal time period will be output and determined as a high-risk video.

[0102] Furthermore, the prohibited language detection model includes:

[0103] The Voice Activity Detection (VAD) sub-model is used to detect voice activity in the video;

[0104] A speech-to-text (ASR) sub-model, used to convert the speech information detected by the voice activity detection sub-model into text information;

[0105] The text matching sub-model is used to match the text information obtained by the speech-to-text sub-model with the corresponding text information;

[0106] The noise filtering sub-model is used to filter the noise in the speech in the video.

[0107] In a specific embodiment, the detection model further includes:

[0108] The video quality detection model is used to read the video clips or the video frames to be detected after a fixed interval, and determine whether there are video quality problems such as distorted screen, black screen, white screen, freeze, etc. in the video clips or the video frames to be detected. If there are video quality problems in the recorded scene, the user will be immediately prompted to pay attention and correct whether there are network or occlusion problems; if there are video quality problems in the recorded scene, the abnormal time point will be output and it will be determined as a suspected high-risk video.

[0109] The liveness detection model is used to read the video clips or the video frames to be detected after a fixed time interval, and determine whether there are non-live objects in the video clips or the video frames to be detected (for example: character posters, character pictures, character masks, etc.). If there are scenes in the recording, the user will be immediately prompted to pay attention and correct them; if there are scenes after recording, the abnormal time point will be output and it will be determined as a high-risk video. The models of this inspection point include face detection, liveness recognition, face quality judgment, face calibration, etc.

[0110] The violation detection model is used to read the video clips or the video frames to be detected after a fixed time interval, and determine whether there are violations such as entry, operation, answering, etc. in the video clips or the video frames to be detected. If there are violations in the recorded scene, the user will be immediately prompted to regulate the behavior and restart the dual recording; if there are violations in the post-recording scene, the abnormal time period will be output and determined as a high-risk video.

[0111] Furthermore, the violation detection model includes:

[0112] Face detection sub-model;

[0113] Face recognition sub-model;

[0114] A human body detection sub-model, used to detect human bodies in videos or video frames;

[0115] A hand detection sub-model, used for detecting the hand position of the human body detected in the human body detection sub-model;

[0116] The skeleton point detection sub-model is used to detect the skeleton point positions of the human body detected in the human body detection sub-model;

[0117] The lip sync detection sub-model is used to detect whether the lip movements of the characters in the video or video frame are synchronized with the speech.

[0118] Through the above method, this embodiment reminds users to regulate their behavior in real time during recording; and after the video is finished, batch quality inspection of the double-recorded videos is performed after recording, the entire video is analyzed again and a comprehensive list of possible abnormal behaviors and time points is given, so that abnormalities can be monitored simultaneously during and after recording, greatly reducing the possibility of risks.

[0119] See also Figure 4 , Figure 4 It is a functional module diagram of a dual-recording video anomaly detection device provided in an embodiment of the present application.

[0120] like Figure 4 As shown, the dual-recording video anomaly detection device 400 includes:

[0121] A quality inspection point determination module 10, for determining at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency;

[0122] The video frame acquisition module 20 is used to acquire the video frame corresponding to the quality inspection point in the dual-recorded video to be detected as the video frame to be detected;

[0123] A detection model determination module 30, configured to determine a corresponding target detection model based on the scene content corresponding to the video frame to be detected;

[0124] The abnormal behavior detection module 40 is used to perform abnormal behavior detection on the video frame to be detected based on the target detection model.

[0125] Furthermore, the dual-recording video anomaly detection device 400 further includes a detection frequency determination module, which is used to:

[0126] Obtaining the video type of the dual-recording video to be detected, and determining whether the video type is a recording video or a post-recording video;

[0127] When the video type is a recorded video, the first detection frequency is determined to be the target detection frequency.

[0128] Furthermore, the detection frequency determination module is also used to:

[0129] When the video type is a recorded video, a second detection frequency is determined as the target detection frequency, wherein the second detection frequency is higher than the first detection frequency.

[0130] Furthermore, the quality inspection point determination module 10 further includes:

[0131] The quality inspection point determination unit is used to determine each time point as the quality inspection point based on the target detection frequency and starting from the start time of the dual-recording video to be detected.

[0132] Furthermore, the detection model determination module 30 is also used for:

[0133] If the scene content is the policyholder confirmation process, a face detection model, a voice detection model, a liveness detection model and a document detection model are determined as the target detection model.

[0134] Furthermore, the quality inspection point determination module 10 specifically includes:

[0135] A video node determination unit, configured to determine each process node of the dual-recording video to be detected based on the video content and voice content of the dual-recording video to be detected;

[0136] A time point determination unit is used to determine at least one time point in the video segment corresponding to each process node based on the target detection frequency as the quality inspection point.

[0137] Furthermore, the dual-recording video anomaly detection device 400 further includes:

[0138] A first detection module is used to determine whether the dual-recording video to be detected contains preset scene content based on a first detection model; or

[0139] The banned word detection module is used to determine whether there are any illegal words in the dual-recording video to be detected based on the banned word detection model.

[0140] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0141] The above-mentioned device can be implemented in the form of a computer program. Figure 5 Runs on the computer device shown.

[0142] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0143] See also Figure 5 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0144] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any dual-recording video anomaly detection method.

[0145] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0146] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any dual-recording video anomaly detection method.

[0147] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0148] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0149] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0150] Based on the target detection frequency, determining at least one quality inspection point of the dual-recorded video to be detected;

[0151] In the dual-recorded video to be detected, obtaining the video frame corresponding to the quality inspection point as the video frame to be detected;

[0152] Determining a corresponding target detection model based on the scene content corresponding to the video frame to be detected;

[0153] Based on the target detection model, abnormal behavior detection is performed on the video frame to be detected.

[0154] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0155] Obtaining the video type of the dual-recording video to be detected, and determining whether the video type is a recording video or a post-recording video;

[0156] When the video type is a recorded video, the first detection frequency is determined to be the target detection frequency.

[0157] When the video type is a recorded video, a second detection frequency is determined as the target detection frequency, wherein the second detection frequency is higher than the first detection frequency.

[0158] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0159] Based on the target detection frequency, starting from the start time of the dual-recording video to be detected, various time points are determined as the quality inspection points.

[0160] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0161] If the scene content is the policyholder confirmation process, a face detection model, a voice detection model, a liveness detection model and a document detection model are determined as the target detection model.

[0162] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0163] Based on the video content and voice content of the dual-recording video to be detected, determining each process node of the dual-recording video to be detected;

[0164] Based on the target detection frequency, at least one time point is determined in the video segment corresponding to each process node as the quality inspection point.

[0165] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0166] Based on the first detection model, determining whether the dual-recording video to be detected contains a preset scene content; or

[0167] Based on the banned word detection model, it is determined whether the dual-recording video to be detected contains any illegal words.

[0168] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any dual-recording video anomaly detection method provided in an embodiment of the present application.

[0169] The computer-readable storage medium may be an internal storage unit of the computer device described in the above embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device.

[0170] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A dual-recording video anomaly detection method, characterized in that: The anomaly detection method comprises the following steps: Based on the target detection frequency, determining at least one quality inspection point of the dual-recorded video to be detected; In the dual-recorded video to be detected, obtaining the video frame corresponding to the quality inspection point as the video frame to be detected; Determining a corresponding target detection model based on the scene content corresponding to the video frame to be detected; Based on the target detection model, abnormal behavior detection is performed on the video frame to be detected.

2. The method for detecting anomalies in dual-recording videos as claimed in claim 1, characterized in that: Before determining at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency, the method further includes: Obtaining the video type of the dual-recording video to be detected, and determining whether the video type is a recording video or a post-recording video; When the video type is a recorded video, the first detection frequency is determined to be the target detection frequency.

3. The method for detecting anomalies in dual-recording videos according to claim 1, wherein: After obtaining the video type of the dual-recording video to be detected and determining whether the video type is a recording video or a post-recording video, the method further includes: When the video type is a recorded video, a second detection frequency is determined as the target detection frequency, wherein the second detection frequency is higher than the first detection frequency.

4. The method for detecting anomalies in dual-recording videos according to claim 1, wherein: The step of determining at least one quality inspection point of the dual-recorded video to be detected based on the target detection frequency specifically includes: Based on the target detection frequency, starting from the start time of the dual-recording video to be detected, various time points are determined as the quality inspection points.

5. The method for detecting anomalies in dual-recording videos as claimed in claim 1, wherein: The determining of the corresponding target detection model based on the scene content corresponding to the video frame to be detected specifically includes: If the scene content is the policyholder confirmation process, a face detection model, a voice detection model, a liveness detection model and a document detection model are determined as the target detection model.

6. The method for detecting anomalies in dual-recording videos according to claim 1, wherein: The step of determining at least one quality inspection point of the dual-recorded video to be detected based on the target detection frequency specifically includes: Based on the video content and voice content of the dual-recording video to be detected, determining each process node of the dual-recording video to be detected; Based on the target detection frequency, at least one time point is determined in the video segment corresponding to each process node as the quality inspection point.

7. The method for detecting anomalies in dual-recording videos according to any one of claims 1 to 6, characterized in that: After performing abnormal behavior detection on the video frame to be detected based on the target detection model, the method further includes: Based on the first detection model, determining whether the dual-recording video to be detected contains a preset scene content; or Based on the banned word detection model, it is determined whether the dual-recording video to be detected contains any illegal words.

8. A dual-recording video anomaly detection device, characterized in that: The dual-recording video anomaly detection device comprises: A quality inspection point determination module, used to determine at least one quality inspection point of the dual-recording video to be detected based on the target detection frequency; A video frame acquisition module, used to acquire the video frame corresponding to the quality inspection point in the dual-recorded video to be detected as the video frame to be detected; A detection model determination module, used to determine a corresponding target detection model based on the scene content corresponding to the video frame to be detected; The abnormal behavior detection module is used to perform abnormal behavior detection on the video frame to be detected based on the target detection model.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a dual-recording video anomaly detection program stored in the memory and executable by the processor, wherein when the dual-recording video anomaly detection program is executed by the processor, the steps of the dual-recording video anomaly detection method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a dual-recording video anomaly detection program, wherein when the dual-recording video anomaly detection program is executed by a processor, the steps of the dual-recording video anomaly detection method as described in any one of claims 1 to 7 are implemented.

Images