Taking identification method and device based on privacy protection

Abstract

The invention relates to the technical field of public safety management, and discloses a fight recognition method and device based on privacy protection, and the method comprises the steps: carrying out the image capture and collection of an event based on a preset triggering strategy, and carrying out the preprocessing of an image; extracting a human body contour in the preprocessed image, and carrying out pixelation processing on a preset sensitive area of the human body contour; detecting and extracting human body key point coordinates after pixelated processing, and judging whether a preset fighting behavior threshold value is exceeded or not according to the relative distance change, the moving direction and the moving speed of the key points; and evaluating the event, and executing a corresponding preset alarm strategy according to an evaluation result. According to the invention, on the premise of ensuring privacy protection, rapid and accurate identification of fight behaviors is realized, and powerful technical support is provided for public security maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of public safety management, and in particular to a fight identification method and device based on privacy protection. Background Art

[0002] With the development of society and the improvement of people's demand for security, fight recognition technology plays an increasingly important role in the security monitoring system of public places. Traditional fight recognition methods mainly rely on fixed-position cameras to record videos, and then use manual review or simple motion detection algorithms to determine whether there is a fight. However, this method has obvious limitations: on the one hand, manual review is time-consuming and labor-intensive, and inefficient; on the other hand, simple motion detection cannot accurately distinguish between normal activities and abnormal behaviors, resulting in a high false alarm rate. In addition, traditional methods often ignore privacy protection issues when processing images, which may expose sensitive information of the monitored objects.

[0003] In recent years, advances in deep learning and computer vision technology have brought new possibilities to fight identification. Technologies such as posture estimation networks and key point detection can more accurately capture human motion features, thereby improving the accuracy of fighting behavior identification. At the same time, as the public's awareness of privacy protection increases, how to effectively identify fighting behavior while protecting personal privacy has become an urgent problem to be solved. Summary of the invention

[0004] In view of this, the present invention proposes a method and device for identifying fights based on privacy protection, which can effectively identify fighting behaviors while taking into account privacy protection, and is suitable for occasions that require a high degree of privacy protection, such as security management of public space areas. This technical solution not only improves the accuracy of fighting behavior identification, but also meets the requirements for privacy protection in modern public safety management.

[0005] In order to achieve the above object, in a first aspect, the present invention provides a fight identification method based on privacy protection, comprising:

[0006] Based on a preset trigger strategy, executing image capture and collection of events, and preprocessing the images;

[0007] Extracting a human body contour in the preprocessed image, and performing pixelation processing on a preset sensitive area of ​​the human body contour;

[0008] Detecting and extracting the coordinates of the key points of the human body after the pixelation processing, and judging whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points;

[0009] The event is evaluated, and a corresponding preset alarm strategy is executed according to the evaluation result.

[0010] Preferably, the image capture and collection of the event is performed based on a preset trigger strategy, specifically:

[0011] Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period;

[0012] If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data in the spatial area is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed;

[0013] If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

[0014] Preferably, extracting the human body contour in the image after preprocessing and performing pixelation processing on a preset sensitive area of ​​the human body contour is specifically as follows:

[0015] Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy;

[0016] The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

[0017] Preferably, it also includes:

[0018] The dynamic coding threshold adjustment strategy detects the current lighting conditions through the ambient light sensor and dynamically adjusts the intensity and granularity of the mosaic according to the detected lighting conditions.

[0019] Preferably, the root detection and extraction of the coordinates of the key points of the human body after the pixelation processing are specifically:

[0020] Extracting the human body key points according to the posture estimation network, and recording the Euclidean distance between each pair of key points based on a preset distance matrix;

[0021] For continuous frame images, calculate the change in the distance between the same pair of key points in different frames to detect whether there is a sudden movement of moving closer or farther away;

[0022] The movement trajectory of the key point in the continuous frame images is tracked to obtain the movement direction and movement distance of each key point in a unit time, and the movement speed is obtained according to the movement distance, so as to identify fast or sudden movement.

[0023] Preferably, judging whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points is as follows:

[0024] According to the movement direction, movement distance and movement speed of the key point in unit time, a spatial feature that changes with time is obtained;

[0025] Analyzing and processing the spatial features based on a preset classification model to obtain a fighting behavior probability value;

[0026] The probability value is compared with the preset fighting behavior threshold value. If it exceeds the threshold value, it is judged as a fighting behavior. Otherwise, it is judged as a non-fighting behavior.

[0027] Preferably, the event is evaluated and a corresponding preset alarm strategy is executed according to the evaluation result, specifically:

[0028] When a fight is identified, an alarm signal is sent to the monitoring center via the wireless network, and the number of participants is detected based on the target detection algorithm;

[0029] If the number of participants mentioned exceeds two, the person in charge will be notified directly by SMS or phone;

[0030] When it is determined to be non-fighting behavior, but the monitored sound data exceeds the preset abnormal threshold, the sound data is recorded and sent to the monitoring center to avoid escalation of conflict.

[0031] In a second aspect, the present invention provides a fight identification device based on privacy protection, comprising:

[0032] A trigger module, used to execute image capture and collection of events based on a preset trigger strategy, and to pre-process the images;

[0033] A privacy module, used to extract the human body contour in the pre-processed image, and perform pixelation processing on a preset sensitive area of ​​the human body contour;

[0034] A judgment module, used to detect and extract the coordinates of the key points of the human body after the pixelation processing, and judge whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points;

[0035] The alarm module is used to evaluate the event and execute a corresponding preset alarm strategy according to the evaluation result.

[0036] Preferably, the trigger module is specifically:

[0037] Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period;

[0038] If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data in the spatial area is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed;

[0039] If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

[0040] Preferably, the privacy module is specifically:

[0041] Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy;

[0042] The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

[0043] The embodiment of the present application discloses a method and device for identifying fights based on privacy protection. The method continuously tracks and analyzes the key points of the human body in the image, and realizes efficient and accurate identification of fighting behaviors based on the relative distance changes, moving directions and speeds between the key points, combined with the preset distance matrix and classification model, which not only improves the response speed of the system, but also greatly reduces the possibility of false alarms. In the entire image processing process, the present invention always pays attention to protecting personal privacy. First, the non-human contour part is blurred in the image preprocessing stage to avoid the leakage of irrelevant details; secondly, the mosaic algorithm is applied to the preset sensitive area within the human contour to ensure that the sensitive parts will not be clearly displayed; in addition, a dynamic coding threshold adjustment strategy is added to maintain a good privacy protection effect even under different lighting conditions. The present invention reasonably arranges the time of image capture and collection by setting different trigger mechanisms for rest time periods and non-rest time periods, reduces unnecessary resource consumption, especially in the rest time period, image capture is only started when abnormal sounds are detected, effectively saving electricity and other system resources. After confirming that a fight has occurred, the present invention immediately sends an alarm signal to the monitoring center and takes graded alarm measures according to the number of participants. If more than two people participate, the person in charge will be notified directly so that intervention measures can be taken in time to prevent the situation from further deteriorating. At the same time, for situations where there is no fight but the sound is abnormal, relevant data will also be recorded and sent to the monitoring center as a basis for subsequent analysis, which helps to prevent potential risks. In summary, the present invention realizes the rapid and accurate identification of fighting behavior while ensuring privacy protection, and provides strong technical support for maintaining public safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 A flowchart of a fight identification method based on privacy protection provided by an embodiment of the present invention;

[0046] Figure 2 A schematic structural diagram of a fight identification device based on privacy protection provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0048] like Figure 1 As shown, in some embodiments of the present application, this embodiment provides a fight identification method based on privacy protection. Specifically, the method includes the following steps:

[0049] Step S101: based on a preset trigger strategy, executing image capture and collection of an event, and preprocessing the image.

[0050] As mentioned above, the system will start the image capture function according to the pre-set conditions and perform a series of processing on the captured images to ensure the effectiveness and accuracy of the subsequent processing flow. The preset trigger strategy can be time-based, motion-based or sound-based, which can effectively reduce unnecessary image capture and save storage resources and computing resources.

[0051] For example, suppose that a fight recognition system capable of executing the present invention is installed in the public space area of ​​a school dormitory building. The system sets two time periods: 10 pm to 6 am is a rest period, and the rest of the time is a non-rest period. During the rest period, the system starts a sound trigger mechanism. If the system detects that the sound in the space area exceeds a preset abnormal threshold (for example, a scream or a heavy object hitting the sound), it automatically starts recording the video and performing image acquisition. During the non-rest period, the system enables the picture trigger mechanism, and when someone enters the detection area of ​​the public space, the system immediately starts image capture and acquisition.

[0052] It should be noted that in a specific implementation scenario, a multi-level trigger strategy scheme can be adopted on the basis of the above scheme, that is, in addition to the basic time period division, the trigger strategy can be dynamically adjusted according to the flow of people or specific events (such as during large-scale events). For example, during peak hours of traffic, the system can lower the trigger threshold to increase the probability of capturing potential fighting incidents; an intelligent image preprocessing scheme can be adopted, that is, in the image preprocessing stage, in addition to basic format conversion and size adjustment, image enhancement technology such as brightness correction and contrast enhancement can be introduced to help improve the accuracy of subsequent human contour detection; a multimodal data fusion scheme can be adopted, that is, considering that single-modal data may have limitations, image capture and collection can be combined with other types of sensor data (such as infrared sensors, humidity sensors, etc.) to form multimodal data input to improve the robustness of fighting behavior recognition. The above optional schemes all fall within the scope of protection of this application.

[0053] Step S102: extracting the human body contour in the pre-processed image, and performing pixelation processing on a preset sensitive area of ​​the human body contour.

[0054] As mentioned above, based on the pre-processed image, computer vision technology is used to identify and extract the human body contours in the image, and the sensitive areas in these contours are pixelated to protect personal privacy. The sensitive areas here usually refer to the face or other parts that need to be protected, and pixelation refers to making these areas blurred or unrecognizable through certain algorithms.

[0055] For example, in a specific embodiment, the system first uses an edge detection algorithm (such as Canny edge detection) to identify the human body outline in the image. Once the human body outline is detected, the system will identify sensitive areas such as the face. For example, a face detection algorithm (such as a Haar feature classifier or a deep learning model) can be used to locate the face. Once the position of the face is determined, the system will pixelate the area, usually by applying a mosaic algorithm to the area so that the facial features cannot be clearly identified. In this way, while retaining the information required for fighting behavior identification, it also ensures that personal privacy is not violated.

[0056] It should be noted that in specific implementation scenarios, a multi-level privacy protection mechanism can be adopted on the basis of the above scheme, that is, in addition to pixelating sensitive areas such as the face, other areas that may contain sensitive information can also be blurred to varying degrees as needed. For example, different mosaic intensity levels can be set, and higher levels of pixelation can be used for particularly sensitive information; a dynamic pixelation intensity adjustment scheme can be adopted, that is, the system can dynamically adjust the intensity and granularity of the mosaic according to the ambient lighting conditions. For example, when the ambient light is dim, the pixelation intensity is appropriately increased to ensure that privacy can be effectively protected even under low light conditions; a privacy protection rule customization scheme can be adopted, that is, allowing users or administrators to customize privacy protection rules according to actual conditions, such as specifying which areas need to be pixelated and the degree of pixelation. This can better adapt to the privacy protection needs in different scenarios, and the above optional schemes all fall within the protection scope of this application.

[0057] Step S103, detecting and extracting the coordinates of the key points of the human body after the pixelation processing, and judging whether the preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points.

[0058] As mentioned above, the system detects and extracts the key points of the human body in the processed image, and then analyzes the relative distance changes, movement direction and speed between these key points to determine whether there is a possibility of fighting. The key points here usually include joint positions (such as shoulders, elbows, knees, ankles, etc.). By analyzing these points, the action state of the person can be inferred.

[0059] For example, suppose that in a video surveillance system in a public space area, the system has completed the preprocessing of the image and the pixelation of sensitive areas. Next, the system uses a posture estimation network (such as OpenPose or DeepCut, etc.) to detect and extract the coordinates of the key points of the human body. For example, the system identifies the key point positions of a person's shoulders, arms, waist, and legs. The system then analyzes the relative position changes of these key points. If the system detects that a person's arm is approaching another person quickly, and the distance between the two people is rapidly reduced in a short period of time, the system may consider this to be a potential fight. In addition, the system also analyzes the movement direction and speed of the key points. If certain key points (such as arms) are found to move toward another object at a faster speed, the possibility of fighting will also increase.

[0060] It should be noted that in a specific implementation scenario, a multi-perspective analysis solution can be adopted on the basis of the above solution. That is, if the system is equipped with multiple cameras, images can be captured from different angles, and data from multiple perspectives can be integrated to improve the accuracy and robustness of fighting behavior judgment. For example, the key points of the human body can be detected from the front and side angles at the same time, and the judgment results of the fighting behavior can be cross-validated; a three-dimensional reconstruction and depth perception solution can be adopted, that is, a depth camera (such as an RGB-D camera) is used to obtain three-dimensional image data and perform three-dimensional reconstruction to more accurately capture the position changes of the human body in space. Three-dimensional reconstruction can help more accurately measure the distance between key points and provide richer spatial information; a behavior prediction solution can be adopted, that is, a behavior prediction model is established based on historical data to predict the movement trend of key points of the human body in the next few seconds, and to warn of potential fighting behavior in advance. This prediction can help the system issue a warning before the fighting behavior actually occurs. The above optional solutions all fall within the scope of protection of this application.

[0061] Step S104: Evaluate the event and execute a corresponding preset alarm strategy according to the evaluation result.

[0062] As mentioned above, the system evaluates possible behavioral events based on the data collected and analyzed in the previous steps, and selects corresponding warning measures based on the evaluation results. This process involves evaluating the probability value of fighting behavior and taking appropriate actions based on the evaluation results, such as sending an alarm to the monitoring center or directly notifying the relevant responsible person.

[0063] For example, suppose the system detects rapid movement and physical contact between two people in the monitoring of a public space area, and based on the analysis of key points, it is concluded that the probability value of fighting behavior exceeds the preset threshold. At this point, the system will evaluate this event as a fighting behavior and take action according to the preset alarm strategy. For example: the system can send an alarm signal to the monitoring center through a wireless network (such as Wi-Fi or cellular network) so that the staff can check the situation on the scene in time and take necessary measures; if the system evaluates that the number of participants exceeds two, the responsible personnel can be notified directly by SMS or phone to ensure that they can arrive at the scene as soon as possible to deal with it; if the sound data monitored by the system exceeds the preset abnormal threshold, but has not yet reached the standard of fighting behavior, the sound data will be recorded and sent to the monitoring center for subsequent analysis and to prevent the escalation of conflicts.

[0064] It should be noted that in a specific implementation scenario, on the basis of the above-mentioned scheme, an alarm linkage mechanism scheme can be adopted, that is, the system can be linked with other security systems in the building (such as access control systems, emergency broadcast systems). Once a fighting behavior is detected, the linkage mechanism is immediately activated, such as locking the entrance to the space area, playing emergency broadcasts, etc.; a historical data analysis and report generation scheme can be adopted, that is, the system can regularly analyze historical data and generate detailed event reports, including the frequency of events, time period distribution, common behavior patterns, etc., to help managers understand the use of public space areas and formulate improvement measures accordingly; a user feedback mechanism scheme can be adopted, that is, users or managers are allowed to provide feedback on alarm events so that the system can continuously optimize the alarm strategy. For example, if a user reports that an alarm is a false alarm, the system can learn from such situations and reduce the false alarm rate in the future. The above optional schemes all fall within the scope of protection of this application.

[0065] In some embodiments of the present application, in order to adopt different trigger strategies in different time periods, thereby ensuring timely capture of potential fighting events while reducing unnecessary image capture and saving storage resources and computing resources, the image capture and collection of events is executed based on the preset trigger strategy, specifically:

[0066] Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period;

[0067] If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data in the public space area is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed;

[0068] If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

[0069] As mentioned above, according to the rules of daily life, users customize each natural day into two time periods - rest time period and non-rest time period. During the rest time period, the system activates the sound trigger mechanism, which means that the system will continue to monitor the sound data in the public space area. Once the sound data recognized by the system exceeds the preset abnormal threshold (for example, abnormally loud noise or screaming), it is considered that there may be an abnormal situation. At this time, the system will automatically start the image capture and acquisition function to record the relevant video or picture data. During the non-rest time period, the system uses a screen trigger mechanism, which means that the system will monitor the screen in the public space area in real time. When the system detects that a human body has entered the pre-set detection area, it will immediately start the image capture and acquisition function to record the video or picture data in that time period.

[0070] In some embodiments of the present application, in order to effectively protect the personal privacy involved in the image while retaining key information, the human body contour in the image after preprocessing is extracted, and the preset sensitive area of ​​the human body contour is pixelated, specifically:

[0071] Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy;

[0072] The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

[0073] As mentioned above, the pre-processed image is processed using an edge detection algorithm (such as the Canny edge detection algorithm) to identify and extract the human body contour in the image, highlighting the edge of the human body for subsequent processing. After the human body contour is extracted, the image details outside the contour are blurred to protect privacy and ensure that the background and other irrelevant information in the image are not clearly displayed. Use relevant functions or algorithms in image processing libraries (such as OpenCV) to detect preset sensitive areas in the human body contour. Once these sensitive areas are detected, the system will apply a mosaic algorithm to pixelate them, making the sensitive areas blurred, thereby protecting personal privacy.

[0074] In some embodiments of the present application, in order to automatically adjust the coding parameters under different lighting conditions to ensure that personal privacy can be effectively protected under any circumstances while maintaining the consistency and reliability of image processing, it also includes:

[0075] The dynamic coding threshold adjustment strategy detects the current lighting conditions through the ambient light sensor and dynamically adjusts the intensity and granularity of the mosaic according to the detected lighting conditions.

[0076] As mentioned above, the lighting conditions of the current public space area are detected by the ambient light sensor installed on the monitoring equipment. The ambient light sensor can measure the light intensity of the surrounding environment in real time. The system analyzes the detected lighting conditions. Since different light intensities will affect the quality and clarity of the image, it is necessary to adjust the parameters of subsequent processing according to the lighting conditions. According to the analysis results, the system will dynamically adjust the intensity and granularity of the mosaic processing, including: in the case of strong light, the intensity and granularity of the mosaic can be appropriately reduced, because the image itself is clearer under strong light, and smaller mosaic particles are sufficient to achieve the effect of protecting privacy; in the case of weak light, the intensity and granularity of the mosaic need to be increased, because the image may not be clear enough in a weak light environment, and larger mosaic particles can better protect privacy.

[0077] In some embodiments of the present application, in order to effectively detect and extract the key points of the human body in the image, and determine whether there is potential fighting behavior by analyzing the dynamic changes of these key points, the root detects and extracts the coordinates of the key points of the human body after the pixelation processing, specifically:

[0078] Extracting the human body key points according to the posture estimation network, and recording the Euclidean distance between each pair of key points based on a preset distance matrix;

[0079] For continuous frame images, calculate the change in the distance between the same pair of key points in different frames to detect whether there is a sudden movement of moving closer or farther away;

[0080] The movement trajectory of the key point in the continuous frame images is tracked to obtain the movement direction and movement distance of each key point in a unit time, and the movement speed is obtained according to the movement distance, so as to identify fast or sudden movement.

[0081] As mentioned above, the pixelated image is processed using a posture estimation network (such as OpenPose, Mask R-CNN, etc.) to extract the key points of the human body in the image. These key points usually include joint positions, such as head, shoulders, elbows, wrists, hips, knees and ankles. Based on the preset distance matrix, the Euclidean distance between each pair of key points is recorded. These distances can be used to describe the state of the human body posture and its changes. For continuous frame images, the change in the distance between the same pair of key points in different frames is calculated. If the distance between a certain set of key points changes significantly in a short period of time (for example, suddenly close or pull apart), it is considered that there may be abnormal behavior. Track the movement trajectory of each key point in the continuous frame image, record the movement direction and movement distance of each key point in unit time, and obtain the movement speed of the key point by calculating the movement distance. According to the movement speed of the key point, identify whether there is fast or sudden movement. If the movement speed of the key point exceeds the preset threshold, it is considered that there may be fighting behavior.

[0082] In some embodiments of the present application, in order to accurately identify whether there is a fighting behavior based on the dynamic change characteristics of the key points and make corresponding judgments accordingly, the relative distance change, moving direction and speed of the key points are used to determine whether the preset fighting behavior threshold is exceeded, specifically:

[0083] According to the movement direction, movement distance and movement speed of the key point in unit time, a spatial feature that changes with time is obtained;

[0084] Analyzing and processing the spatial features based on a preset classification model to obtain a fighting behavior probability value;

[0085] The probability value is compared with the preset fighting behavior threshold value. If it exceeds the threshold value, it is judged as a fighting behavior. Otherwise, it is judged as a non-fighting behavior.

[0086] As described above, by analyzing the movement direction, moving distance and moving speed of the key points in a unit time, the spatial features that change over time are extracted. These features reflect the relative position relationship between the key points and their changes. The extracted spatial features are analyzed and processed using a preset classification model (such as support vector machine SVM, random forest, deep learning model, etc.). The model calculates the probability value of fighting behavior by learning the features. The calculated fighting behavior probability value is compared with the preset fighting behavior threshold. If the probability value exceeds the preset threshold, the system determines that there is fighting behavior; if the probability value does not exceed the threshold, it is judged as non-fighting behavior.

[0087] In some embodiments of the present application, in order to take corresponding warning measures according to different situations and ensure timely and effective handling of potential security issues, the event is evaluated and the corresponding preset warning strategy is executed according to the evaluation result, specifically:

[0088] When a fight is identified, an alarm signal is sent to the monitoring center via the wireless network, and the number of participants is detected based on the target detection algorithm;

[0089] If the number of participants mentioned exceeds two, the person in charge will be notified directly by SMS or phone;

[0090] When it is determined to be non-fighting behavior, but the monitored sound data exceeds the preset abnormal threshold, the sound data is recorded and sent to the monitoring center to avoid escalation of conflict.

[0091] As mentioned above, when the system determines that the current event is a fight, an alarm signal is sent to the monitoring center via a wireless network (such as Wi-Fi or cellular network); at the same time, the system uses a target detection algorithm to detect the number of people involved in the fight. If the number of participants detected exceeds two, the relevant person in charge is directly notified by text message or phone call to ensure that the person in charge can quickly learn about the situation and take appropriate measures. When the system determines that the current event is not a fight, but the monitored sound data exceeds the preset abnormal threshold, the sound data is recorded; the recorded sound data is sent to the monitoring center so that the monitoring center can further analyze the situation and prevent the conflict from escalating.

[0092] Compared with the prior art, the embodiment of the present application discloses a fight identification method based on privacy protection. The method continuously tracks and analyzes the key points of the human body in the image, and combines the preset distance matrix and classification model according to the relative distance change, moving direction and speed between the key points to achieve efficient and accurate identification of fighting behavior, which not only improves the response speed of the system, but also greatly reduces the possibility of false alarms. In the entire image processing process, the present invention always pays attention to protecting personal privacy. First, the non-human contour part is blurred in the image preprocessing stage to avoid the leakage of irrelevant details; secondly, the mosaic algorithm is applied to the preset sensitive area within the human contour to ensure that the sensitive part will not be clearly displayed; in addition, a dynamic coding threshold adjustment strategy is added to maintain a good privacy protection effect even under different lighting conditions. The present invention reasonably arranges the time of image capture and collection by setting different trigger mechanisms for rest time periods and non-rest time periods, reduces unnecessary resource consumption, and especially in the rest time period, image capture is started only when abnormal sound is detected, effectively saving electricity and other system resources. After confirming that a fight has occurred, the present invention immediately sends an alarm signal to the monitoring center and takes graded alarm measures according to the number of participants. If more than two people participate, the person in charge will be notified directly so that intervention measures can be taken in time to prevent the situation from further deteriorating. At the same time, for situations where there is no fight but the sound is abnormal, relevant data will also be recorded and sent to the monitoring center as a basis for subsequent analysis, which helps to prevent potential risks. In summary, the present invention realizes the rapid and accurate identification of fighting behavior while ensuring privacy protection, and provides strong technical support for maintaining public safety.

[0093] Based on the same inventive concept as the above method, the embodiment of the present application also proposes a fight identification device based on privacy protection, such as Figure 2 FIG. 1 is a schematic diagram of a fight identification device based on privacy protection, and the device includes:

[0094] A trigger module, used to execute image capture and collection of events based on a preset trigger strategy, and to pre-process the images;

[0095] A privacy module, used to extract the human body contour in the pre-processed image, and perform pixelation processing on a preset sensitive area of ​​the human body contour;

[0096] A judgment module, used to detect and extract the coordinates of the key points of the human body after the pixelation processing, and judge whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points;

[0097] The alarm module is used to evaluate the event and execute a corresponding preset alarm strategy according to the evaluation result.

[0098] Preferably, the trigger module is specifically:

[0099] Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period;

[0100] If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data based on privacy protection is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed;

[0101] If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

[0102] Preferably, the privacy module is specifically:

[0103] Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy;

[0104] The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

[0105] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0106] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A fight identification method based on privacy protection, characterized in that: include: Based on a preset trigger strategy, executing image capture and collection of events, and preprocessing the images; Extracting a human body contour in the preprocessed image, and performing pixelation processing on a preset sensitive area of ​​the human body contour; Detecting and extracting the coordinates of the key points of the human body after the pixelation processing, and judging whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points; The event is evaluated, and a corresponding preset alarm strategy is executed according to the evaluation result.

2. The method according to claim 1, characterized in that The image capture and collection of the event is performed based on the preset trigger strategy, specifically: Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period; If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data in the spatial area is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed; If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

3. The method according to claim 1, characterized in that The extracting of the human body contour in the image after preprocessing and pixelation processing of the preset sensitive area of ​​the human body contour is specifically as follows: Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy; The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

4. The method according to claim 3, characterized in that Also includes: The dynamic coding threshold adjustment strategy detects the current lighting conditions through the ambient light sensor and dynamically adjusts the intensity and granularity of the mosaic according to the detected lighting conditions.

5. The method according to claim 1, characterized in that The root detection and extraction of the coordinates of the key points of the human body after the pixelation processing are specifically: Extracting the human body key points according to the posture estimation network, and recording the Euclidean distance between each pair of key points based on a preset distance matrix; For continuous frame images, calculate the change in the distance between the same pair of key points in different frames to detect whether there is a sudden movement of moving closer or farther away; The movement trajectory of the key point in the continuous frame images is tracked to obtain the movement direction and movement distance of each key point in a unit time, and the movement speed is obtained according to the movement distance, so as to identify fast or sudden movement.

6. The method according to claim 5, characterized in that The step of judging whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points is as follows: According to the movement direction, movement distance and movement speed of the key point in unit time, a spatial feature that changes with time is obtained; Analyzing and processing the spatial features based on a preset classification model to obtain a fighting behavior probability value; The probability value is compared with the preset fighting behavior threshold value. If it exceeds the threshold value, it is judged as a fighting behavior. Otherwise, it is judged as a non-fighting behavior.

7. The method according to claim 6, characterized in that The event is evaluated, and a corresponding preset alarm strategy is executed according to the evaluation result, specifically: When a fight is identified, an alarm signal is sent to the monitoring center via the wireless network, and the number of participants is detected based on the target detection algorithm; If the number of participants mentioned exceeds two, the person in charge will be notified directly by SMS or phone; When it is determined to be non-fighting behavior, but the monitored sound data exceeds the preset abnormal threshold, the sound data is recorded and sent to the monitoring center to avoid escalation of conflict.

8. A fight identification device based on privacy protection, characterized in that: include: A trigger module, used to execute image capture and collection of events based on a preset trigger strategy, and to pre-process the images; A privacy module, used to extract the human body contour in the pre-processed image, and perform pixelation processing on a preset sensitive area of ​​the human body contour; A judgment module, used to detect and extract the coordinates of the key points of the human body after the pixelation process, and judge whether a preset fighting behavior threshold is exceeded according to the relative distance change, moving direction and speed of the key points; The alarm module is used to evaluate the event and execute a corresponding preset alarm strategy according to the evaluation result.

9. The device according to claim 8, characterized in that The trigger module is specifically: Based on the preset time division strategy, each natural day is divided into rest time period and non-rest time period; If it is in the rest period, the sound trigger mechanism is executed, that is, the sound data in the spatial area is collected and identified, and when the sound data exceeds the preset abnormal threshold, image capture and collection are executed; If it is not a resting period, the screen trigger mechanism will be executed, that is, when a human body is detected entering the detection area, image capture and collection will be performed.

10. The device according to claim 8, characterized in that The privacy module is specifically: Extracting a human body contour in the image according to an edge detection algorithm, and blurring details other than the human body contour to protect privacy; The preset sensitive area of ​​the human body contour is detected according to the image processing library, and the preset sensitive area is pixelated by a mosaic algorithm.

Images