AI-based behavior identification monitoring system

Abstract

The invention relates to the technical field of behavior analysis, in particular to an AI-based behavior recognition monitoring system which comprises a facial expression analysis module, a behavior pattern recognition module, an abnormal behavior judgment module and a monitoring adjustment decision module. According to the method, the emotion is monitored in real time by using the AI, the facial expression change can be accurately evaluated, the emotion fluctuation can be accurately analyzed, the monitoring of the emotion state is more accurate, the behavior type and the possible abnormal condition are identified through the speed and position information of the action, the judgment speed and accuracy of the abnormal behavior are improved, and the accuracy of the abnormal behavior is improved. Therefore, safety monitoring in a complex environment is more effective, potential abnormal behaviors such as invasion or theft behaviors can be accurately identified, response speed and effectiveness of prevention measures are improved, distribution of monitoring resources can be dynamically optimized by adjusting a monitoring strategy according to a behavior risk level, and safety of a user is improved. Therefore, the response capability and the resource use efficiency of the monitoring system are improved, and the overall safety is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior analysis, and in particular to an AI-based behavior recognition monitoring system. Background Art

[0002] The field of behavioral analysis technology involves the use of technical means to identify, classify and interpret the behavior patterns of people or objects. This technology is often used in many fields such as video surveillance, security systems, human-computer interaction and health monitoring. Behavioral analysis technology processes and analyzes the collected data through algorithms and models, such as machine learning, neural networks or other statistical tools, to achieve automatic recognition and prediction of behavior. The technology can extract meaningful behavioral features from video streams, sensor data or direct human-computer interaction to support applications such as security monitoring, user experience optimization or health status monitoring.

[0003] Among them, the behavior recognition monitoring system is a system that uses behavior analysis technology to achieve real-time monitoring and security management in specific environments or scenarios. The main uses include security monitoring, public safety, and workplace supervision. Through cameras and other sensors deployed in key areas, the system can automatically detect and analyze individual behavior patterns, and promptly identify abnormal or potential threatening behaviors, such as intrusion, theft, or other abnormal activities, thereby initiating corresponding early warning or intervention measures. The system not only improves monitoring efficiency, but also greatly reduces dependence on manual intervention. It is an indispensable technical solution in modern security management.

[0004] When dealing with complex or changing environmental conditions, traditional monitoring systems are limited by a single data source and fixed analysis models, resulting in insufficient accuracy and real-time performance in behavior recognition. Without the integration of emotional data, they are unable to accurately interpret certain complex human behavior patterns, such as abnormal behaviors caused by stress, which can lead to false alarms or missed alarms. Especially in high-risk scenarios or those requiring rapid response, traditional monitoring technology is difficult to effectively analyze and judge an individual's specific behavior and emotional state, thereby affecting the efficiency and effectiveness of overall security management. In addition, the coverage and priority of traditional monitoring systems are relatively fixed, making it difficult to adjust them in a timely manner according to demand, resulting in an untimely or inaccurate response to abnormal behavior. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an AI-based behavior recognition monitoring system.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: an AI-based behavior recognition monitoring system, the system comprising:

[0007] The facial expression analysis module extracts video frames per second based on real-time monitoring video data, uses AI to identify facial areas, performs image segmentation on the detected facial areas, and extracts feature points from the facial areas. It uses feature point data to evaluate facial expressions, tracks facial expressions, analyzes emotional fluctuations, and obtains emotional state data.

[0008] The behavior pattern recognition module extracts the speed and spatial position information of individual actions based on real-time monitoring video data, performs time window analysis on the information, compares the behavior features in the behavior pattern database, identifies the behavior type through the matching degree of the action speed and spatial position, and obtains the behavior pattern analysis results;

[0009] The abnormal behavior determination module compares the behavior pattern analysis result with the preset normal behavior, analyzes the deviation between the behavior pattern and the normal pattern, quantifies the behavior deviation, and evaluates the abnormal level according to the deviation size to obtain abnormal behavior identification information;

[0010] The monitoring and adjustment decision module evaluates the behavioral risk level based on the abnormal behavior identification information and emotional state data, combined with the required security level of the current area and the risk sensitivity of the current time period, implements the corresponding risk alarm notification, and adjusts the camera coverage priority of the area according to the risk level, optimizes the coverage efficiency, and obtains the adjusted monitoring strategy.

[0011] The present invention is improved in that the method for performing image segmentation on the detected facial area is:

[0012] Based on real-time monitoring video data, extract video frames per second, use AI to identify facial areas, and obtain facial area recognition results;

[0013] Based on the facial area recognition result, by formula:

[0014] S=α S (F S -F00 2 +β S (G S -G00 2 +γ S (H S -H00 2

[0015] Calculate the segmentation score S of the pixel point, where α S , β S and γ S are the weight coefficients affecting the red, green, and blue channels, respectively. S , G S and H SRespectively represent the actual color values ​​of the image pixels in the red, green, and blue channels, and F0, G0, and H0 are the preset color thresholds;

[0016] Based on the segmentation score S of the pixel point, it is compared with the preset segmentation threshold, and the pixel point with S value higher than the preset segmentation threshold is marked as face, and it is segmented from the background to obtain the face area segmentation result.

[0017] The present invention is improved in that the step of acquiring the emotional state data is:

[0018] Based on the facial region segmentation result, feature points are extracted from the facial region, and position changes of the feature points between consecutive frames are calculated to obtain facial expression change data;

[0019] Based on the facial expression change data, the formula:

[0020]

[0021] Calculate the emotional fluctuation degree E of expression changes to obtain emotional state data, where σ x is the standard deviation of the horizontal displacement of the feature points, σ y is the standard deviation of the vertical displacement of the feature point, λ E is the weight coefficient of the horizontal displacement standard deviation, μ E is the weight coefficient of the standard deviation of vertical displacement, and E is the degree of emotional fluctuation of expression changes.

[0022] The present invention is improved in that the steps of obtaining the behavior pattern analysis result are:

[0023] Based on real-time monitoring video data, the speed and spatial position information of individual actions are extracted, the position coordinates of the individual and the speed vector relative to the previous frame are calculated, and the behavior association data is obtained;

[0024] Based on the behavior association data, it is compared with the known behavior patterns in the database, through the formula:

[0025]

[0026] Calculate the behavior matching degree M, where P(t) represents the spatial position of the individual at time t, V(t) represents the movement speed of the individual at time t relative to the previous frame, and V ref and P ref are the standard speed and position values ​​of known behaviors extracted from the database, κ v and η p is the weight coefficient, δ M and M is the attenuation parameter, M is the behavior matching degree, and e is the base of the natural logarithm;

[0027] Based on the behavior matching degree M, it is compared with a preset matching degree threshold to determine whether the behavior matches, identify the behavior type, and obtain the behavior pattern analysis result.

[0028] The present invention is improved in that the method for quantifying the behavioral deviation is:

[0029] Based on the analysis results of the behavior pattern, compared with the preset normal behavior, the formula is:

[0030]

[0031] Calculate the quantitative measure Q of behavioral deviation, where B(t) is the observed behavior data at time t, N(t) is the normal behavior reference data at time t, ω t is the weight coefficient of time t, Q is the quantitative measure of behavioral deviation;

[0032] The quantitative measurement Q based on the behavior deviation is compared with a preset behavior deviation threshold. If the Q value is greater than the preset behavior deviation threshold, it is judged that there is an obvious deviation in the behavior, and a behavior deviation analysis result is obtained.

[0033] The present invention is improved in that the step of obtaining the abnormal behavior identification information is:

[0034] Based on the behavior deviation analysis results and the behavior pattern analysis results, behavior type weights are extracted to obtain abnormal level association data;

[0035] Based on the abnormal level association data, by formula:

[0036] F=ω type ln(1+Q)

[0037] Calculate the abnormal behavior level F and obtain abnormal behavior identification information, where ω type is the behavior type weight coefficient, Q is the quantitative measure of behavior deviation, and F is the level of behavior abnormality.

[0038] The present invention is improved in that the method for implementing the corresponding risk alert notification is:

[0039] Based on the abnormal behavior identification information and the emotional state data, the security level required for the current area and the risk sensitivity of the current period are extracted to obtain risk association data;

[0040] Based on the risk association data, through the formula:

[0041] G=(α G ·F+E)·ln(1+β G ·S G +γ G ·R G)

[0042] Calculate the behavior risk level G, where E is the emotional fluctuation degree of expression changes, F is the abnormal behavior level, S G is the security level of the current area, R G is the risk sensitivity of the current period, G is the behavioral risk level, α G , β G and γ G is the weight coefficient,

[0043] Based on the behavior risk level G, it is compared with the preset risk standard, and a corresponding risk alarm notification is implemented to obtain a risk alarm result.

[0044] The present invention is improved in that the method of the adjusted monitoring strategy is:

[0045] Based on the risk alarm results, according to the abnormal behavior risk level of the area, the formula is:

[0046]

[0047] Calculate the camera coverage priority C, where ξ C and η C is the adjustment coefficient, G is the behavioral risk level, S G is the security level of the current area, C is the camera coverage priority, λ C is the weight coefficient;

[0048] Based on the camera coverage priority C, according to the size of the C value, the camera coverage priority of the area is adjusted to optimize the coverage efficiency and obtain the adjusted monitoring strategy.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are:

[0050] In the present invention, by utilizing AI to monitor emotions in real time, it is possible to accurately evaluate changes in facial expressions, analyze emotional fluctuations, and make the monitoring of emotional states more accurate. By using the speed and position information of the action, the type of behavior and its possible abnormal conditions can be identified, thereby improving the speed and accuracy of determining abnormal behaviors, making security monitoring in complex environments more effective, and being able to accurately identify potential abnormal behaviors, such as intrusion or theft, thereby improving the response speed and effectiveness of preventive measures. By adjusting the monitoring strategy according to the risk level of the behavior, the allocation of monitoring resources can be dynamically optimized, thereby improving the responsiveness of the monitoring system and the efficiency of resource utilization, reducing resource waste, and enhancing overall security.

[0051] The AI-based behavior recognition monitoring system optimizes the accuracy of data processing and emotional judgment when analyzing the facial expressions and action information of people. By comparing behavior patterns in a continuously updated database and combining emotional state data, the system can more accurately identify the type of behavior and its possible abnormalities. This joint analysis of data improves the speed and accuracy of determining abnormal behavior, making security monitoring in complex environments more effective. In addition, by adjusting the monitoring strategy according to the risk level of the behavior, the system can dynamically optimize the allocation of monitoring resources, thereby improving the responsiveness and resource utilization efficiency of the monitoring system. These optimization measures focus on improving the system's early warning capabilities and intervention timeliness for emergencies, significantly reducing dependence on manual monitoring, and enhancing the automation and intelligence level of security management. . BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a system flow chart of the present invention;

[0053] Figure 2 A flow chart of performing image segmentation on a detected facial area according to the present invention;

[0054] Figure 3 A flow chart of obtaining emotional state data of the present invention;

[0055] Figure 4 A flow chart for obtaining behavior pattern analysis results for the present invention;

[0056] Figure 5 A flow chart for quantifying behavioral deviations for the present invention;

[0057] Figure 6 A flow chart of obtaining abnormal behavior identification information for the present invention;

[0058] Figure 7 A flowchart of the corresponding risk alert notification for the implementation of the present invention;

[0059] Figure 8 A flow chart of the monitoring strategy after adjustment is obtained for the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0062] See also Figure 1 The present invention provides a technical solution: an AI-based behavior recognition monitoring system, the system comprising:

[0063] The facial expression analysis module extracts video frames per second based on real-time monitoring video data, uses AI to identify facial areas, performs image segmentation on the detected facial areas, and extracts feature points from the facial areas. It uses feature point data to evaluate facial expressions, tracks facial expressions, analyzes emotional fluctuations, and obtains emotional state data.

[0064] The behavior pattern recognition module extracts the speed and spatial position information of individual actions based on real-time monitoring video data, performs time window analysis on the information, compares the behavior features in the behavior pattern database, identifies the behavior type through the matching degree of the action speed and spatial position, and obtains the behavior pattern analysis results;

[0065] The abnormal behavior determination module compares the behavior pattern analysis results with the preset normal behavior, analyzes the deviation between the behavior pattern and the normal pattern, quantifies the behavior deviation, and evaluates the abnormal level according to the deviation size to obtain abnormal behavior identification information;

[0066] The monitoring and adjustment decision module evaluates the behavior risk level based on abnormal behavior recognition information and emotional state data, combined with the required security level of the current area and the risk sensitivity of the current period, implements the corresponding risk alarm notification, and adjusts the camera coverage priority of the area according to the risk level, optimizes the coverage efficiency, and obtains the adjusted monitoring strategy;

[0067] Emotional state data includes emotion category, emotion intensity and degree of emotion fluctuation; behavioral pattern analysis results include behavior type, behavior frequency, behavior duration; abnormal behavior identification information includes behavioral deviation and abnormality level; and the adjusted monitoring strategy includes monitoring frequency, priority monitoring areas and resource allocation priority.

[0068] See also Figure 2 , the method for image segmentation of the detected facial area is:

[0069] Based on real-time monitoring video data, extract video frames per second, use AI to identify facial areas, and obtain facial area recognition results;

[0070] Based on the facial area recognition results, the formula is:

[0071] S=α S (F S -F0) 2 +β S (G S -G0) 2 +γ S (H S -H0) 2

[0072] Calculate the segmentation score S of the pixel point, where α S , β S and γ S are the weight coefficients affecting the red, green, and blue channels, respectively. S , G S and H S Respectively represent the actual color values ​​of the image pixels in the red, green, and blue channels, and F0, G0, and H0 are the preset color thresholds;

[0073] Based on the segmentation score S of the pixel point, it is compared with the preset segmentation threshold, and the pixel points with S value higher than the preset segmentation threshold are marked as faces, and they are segmented from the background to obtain the facial area segmentation result.

[0074] formula:

[0075] S=α S (F S -F0) 2 +β S (G S -G0) 2 +γ S (H S -H0) 2

[0076] The meaning and acquisition method of parameters:

[0077] α S , β S and γ S It is used to adjust the weights of the red, green, and blue channels in image segmentation. These weights are usually optimized through experiments to adjust the influence of different color channels and ensure the accuracy of image segmentation. For example, these weights can be determined by analyzing the contribution of each color channel to the accuracy of facial recognition in multiple sample images.

[0078] F S , GS and H S Represents the actual color value of a pixel in the red, green, and blue channels, which is read directly from the image file.

[0079] F0, G0, and H0 are preset color thresholds, usually set based on the average color of a human face in a typical environment. These thresholds can be determined by analyzing a large number of facial images and calculating the average value for each color channel.

[0080] Calculation example:

[0081] The following parameter settings and pixel values ​​are set: α S =0.3,β S =0.3,γ S =0.4, F S =120, G S =130,H S =140 is the color channel value of a certain pixel. F0=100, G0=110, H0=120 thresholds are calculated based on the average color of the face.

[0082] Plugging the values ​​into the formula:

[0083] S=0.3(120-100) 2 +0.3(130-110) 2 +0.4(140-120) 2

[0084] =0.3(400)+0.3(400)+0.4(400)

[0085] =120+120+160

[0086] =400

[0087] The calculated segmentation score S is 400, which is likely to be part of the key features of the facial region. This processing allows facial image segmentation to effectively distinguish between facial and non-facial areas while maintaining details, improving the overall effectiveness of the facial recognition system.

[0088] See also Figure 3 , the steps to obtain emotional state data are:

[0089] Based on the facial region segmentation results, feature points are extracted from the facial region, and the position changes of feature points between consecutive frames are calculated to obtain facial expression change data;

[0090] Based on the facial expression change data, the formula is:

[0091]

[0092] Calculate the emotional fluctuation degree E of expression changes to obtain emotional state data, where σ x is the standard deviation of the horizontal displacement of the feature points, σ y is the standard deviation of the vertical displacement of the feature point, λ E is the weight coefficient of the horizontal displacement standard deviation, μ E is the weight coefficient of the standard deviation of vertical displacement, and E is the degree of emotional fluctuation of expression changes.

[0093] formula:

[0094]

[0095] The meaning and acquisition method of parameters:

[0096] σ x and σ y : The standard deviation of the horizontal and vertical displacement of the feature points is the standard deviation of the displacement vectors Δx and Δy, which is obtained by calculating the root mean square difference of all displacement vectors over a period of time. It reflects the discrete degree of expression changes. Δx and Δy represent the horizontal and vertical displacements of the feature points between consecutive video frames, which are obtained by comparing the position changes of the same feature points in two consecutive frames.

[0097] λ E and μ E : The weight coefficient is used to adjust the contribution of the horizontal and vertical displacement standard deviations to ensure appropriate influence in the sentiment fluctuation analysis. The weight is usually adjusted based on experimental data to optimize the accuracy of the sentiment analysis model.

[0098] Calculation example:

[0099] The following parameters and data are set: the sequence of Δx is [4, -2, 3, -1], and the sequence of Δy is [5, -3, 4, -2]. The average values ​​of Δx and Δy are calculated as μ x =1 and μ y =1.

[0100] Calculate σ x and σ y :

[0101]

[0102] Set the weight coefficient λ E =0.5 and μ E =0.5, calculate E:

[0103]

[0104] The calculated degree of emotion fluctuation E is 3.08, indicating that facial expressions change actively and emotions fluctuate significantly during the observation period. The analysis can help identify periods of high emotion fluctuations.

[0105] See also Figure 4 , the steps to obtain the behavior pattern analysis results are:

[0106] Based on real-time monitoring video data, the speed and spatial position information of individual actions are extracted, the position coordinates of the individual and the speed vector relative to the previous frame are calculated, and the behavior association data is obtained;

[0107] Based on the behavior association data, it is compared with the known behavior patterns in the database through the formula:

[0108]

[0109] Calculate the behavior matching degree M, where P(t) represents the spatial position of the individual at time t, V(t) represents the movement speed of the individual at time t relative to the previous frame, and V ref and P ref are the standard speed and position values ​​of known behaviors extracted from the database, κ v and η p is the weight coefficient, δ M and M is the attenuation parameter, M is the behavior matching degree, and e is the base of the natural logarithm;

[0110] Based on the behavior matching degree M, it is compared with the preset matching degree threshold to determine whether the behavior matches, identify the behavior type, and obtain the behavior pattern analysis results.

[0111] formula:

[0112]

[0113] Parameter meaning and acquisition method

[0114] Position coordinates P(t): represent the spatial position of an individual at time t, and directly extract the coordinates of the individual from the video frame data.

[0115] The velocity vector V(t0: represents the moving speed of the individual at time t relative to the previous frame, calculated as the difference between the position coordinates of the current frame and the previous frame.

[0116] Reference speed V ref and the reference position P ref : These are standard speed and position values ​​for a specific behavior extracted from the behavior pattern database and used as matching reference points.

[0117] Weight coefficient κ v and η p : Adjust the influence of speed and position information in behavior matching scores based on previous experimental data.

[0118] Attenuation parameter δ M and M : Used to adjust the impact of speed difference and position difference, obtained based on system performance tuning.

[0119] Calculation example:

[0120] Set the following parameters and data: P(t) = (100, 150), V(t) = (5, 5) represents the position and velocity of the individual in the current frame. ref =(3,4),P ref = (95,145) The reference speed and position of a specific behavior in the behavior database. κ v =0.8,η p =0.7 weighting factor, adjusted based on past data. M =0.05,ξ M = 0.03 Attenuation parameter, used to control the impact of speed and position variations on matching.

[0121] Bring in the behavior matching calculation formula:

[0122]

[0123] The calculated result M≈0.406 indicates that the observed behavior has a low match with the reference behavior pattern in the database. The calculation method effectively utilizes the dynamic position and velocity information, combined with the preset database standards, to accurately identify the type of individual behavior.

[0124] See also Figure 5 , the method to quantify behavioral deviation is:

[0125] Based on the behavior pattern analysis results, compared with the preset normal behavior, the formula is:

[0126]

[0127] Calculate the quantitative measure Q of behavioral deviation, where B(t) is the observed behavior data at time t, N(t) is the normal behavior reference data at time t, ω t is the weight coefficient of time t, Q is the quantitative measure of behavioral deviation;

[0128] The quantitative measurement Q based on behavioral deviation is compared with the preset behavioral deviation threshold. If the Q value is greater than the preset behavioral deviation threshold, it is judged that there is an obvious deviation in the behavior, and the behavioral deviation analysis result is obtained.

[0129] formula:

[0130]

[0131] The meaning and acquisition method of parameters:

[0132] B(t): represents the observed behavior data at time t, extracted from the behavior monitoring video, and usually refers to the quantitative value of the behavior characteristics observed at a specific time point t. The data includes the individual's position, speed, action frequency, etc., which can be collected by sensors or behavior monitoring systems.

[0133] N(t): represents the normal behavior reference data at time t, extracted from the historical behavior pattern database.

[0134] ω t : Weight coefficients at time t, which are pre-set according to the importance of the behavior at each time point. These weights can be adjusted based on the expected impact of the behavior or the results of historical behavior analysis.

[0135] Calculation example:

[0136] The parameters are set as follows: observed behavior data B(t) = [4, 6, 5, 7], normal behavior data N(t) = [3, 6, 4, 5], time weight coefficient ω t =[1, 1.5, 1, 2], the judgment threshold of behavioral deviation is 3.

[0137] Calculate the behavioral deviation Q:

[0138] D(t)=|B(t)-N(t)|

[0139] Calculate the difference in behavior from normal mode at a given time:

[0140] D(t)=[|4-3|,|6-6|,|5-4|,|7-5|]=[1,0,1,2]

[0141] Multiply the difference values ​​by the time weights and sum them:

[0142] Q=1×1+1.5×0+1×1+2×2=1+0+1+4=6

[0143] The calculation result Q=6 indicates that the behavior deviation is 6, which exceeds the set threshold of 3, indicating that the observed behavior deviates significantly from the normal pattern and may indicate abnormal behavior.

[0144] See also Figure 6 ,The steps to obtain abnormal behavior identification information are:

[0145] Based on the behavior deviation analysis results and behavior pattern analysis results, the behavior type weight is extracted to obtain abnormal level association data;

[0146] Based on the abnormal level correlation data, through the formula:

[0147] F=ω type ln(1+Q)

[0148] Calculate the abnormal behavior level F and obtain abnormal behavior identification information, where ω type is the behavior type weight coefficient, Q is the quantitative measure of behavior deviation, and F is the level of behavior abnormality.

[0149] formula:

[0150] F=ω type ln(1+Q)

[0151] The meaning and acquisition method of parameters:

[0152] Q: A quantitative measure of behavioral deviation, calculated in the previous step.

[0153] ω type : Behavior type weights are set in advance based on the potential risks or social impact of the behavior. For example, for security-sensitive behavior types, the weights will be set higher.

[0154] Calculation Example

[0155] Set the following parameter settings: behavior deviation quantification metric Q = 15, behavior type weight ω type =2.0, assuming that this behavior type is associated with high risk.

[0156] Perform the calculation:

[0157] F=ω type ln(1+Q)

[0158] =2.0·ln(1+15)

[0159] =2.0·ln(16)

[0160] =2.0·2.77

[0161] =5.54

[0162] The calculated anomaly level F is 5.54, which is a medium value. The surface behavior has not yet reached the abnormal level that requires high attention.

[0163] See also Figure 7 , the method to implement the corresponding risk alert notification is:

[0164] Based on abnormal behavior recognition information and emotional state data, the required security level of the current area and the risk sensitivity of the current period are extracted to obtain risk association data;

[0165] Based on risk association data, through the formula:

[0166] G=(α G ·F+E)·ln(1+β G ·SG +γ G ·R G )

[0167] Calculate the behavior risk level G, where E is the emotional fluctuation degree of expression changes, F is the abnormal behavior level, S G is the security level of the current area, R G is the risk sensitivity of the current period, G is the behavioral risk level, α G , β G and γ G is the weight coefficient,

[0168] Based on the behavioral risk level G, it is compared with the preset risk standards, and the corresponding risk alarm notification is implemented to obtain the risk alarm result.

[0169] formula:

[0170] G=(α G ·F+E)·ln(1+β G ·S G +γ G ·R G )

[0171] Parameter meaning and acquisition method:

[0172] E: The degree of emotional fluctuation of expression changes, calculated through the previous steps.

[0173] F: Behavioral abnormality level, calculated by the previous step.

[0174] S G :The security level of the current area is set by the security management department based on the characteristics of the area and historical security data.

[0175] R G :The risk sensitivity of the current period is dynamically adjusted based on the security history and predicted risk changes of the time period.

[0176] α G , β G and γ G : Weight coefficient, obtained through historical data and statistical analysis, is used to adjust the contribution ratio of safety level and risk sensitivity in risk assessment.

[0177] Calculation example:

[0178] The following parameters were set: F = 5.5 behavioral abnormality level. G =2 represents a higher regional security level. G =3Higher risk sensitivity in the current period. α G =1.5,β G =0.5,γG =0.5 to adjust the weight coefficient, E=2.5.

[0179] Perform the calculation:

[0180] G=(α G ·F+E)·ln(1+β G ·S G +γ G ·R G )

[0181] =(1.5·5.5+2.5)·ln(1+0.5·2+0.5·3)

[0182] =10.75·1.2528

[0183] =13.47

[0184] The calculated behavioral risk level G is 13.47, which is a large value, indicating that the current behavior is considered high risk within the given security level and risk-sensitive period.

[0185] See also Figure 8 , the method of the adjusted monitoring strategy is:

[0186] Based on the risk alarm results, according to the abnormal behavior risk level of the area, the formula is:

[0187]

[0188] Calculate the camera coverage priority C, where ξ C and η C is the adjustment coefficient, G is the behavioral risk level, S G is the security level of the current area, C is the camera coverage priority, λ C is the weight coefficient;

[0189] Based on the camera coverage priority C, according to the size of the C value, the camera coverage priority of the area is adjusted to optimize the coverage efficiency and obtain the adjusted monitoring strategy.

[0190] formula:

[0191]

[0192] The meaning and acquisition method of parameters:

[0193] G: Behavioral risk level, calculated by the previous steps.

[0194] S G :The security level of the current area is set by the security management department based on the characteristics of the area and historical security data.

[0195] λ C : Weight coefficient, used to adjust the influence of risk level in the calculation process to ensure that the calculation results can reasonably reflect the degree of risk.

[0196] ξ C and η C : Adjustment coefficient, used to adjust the safety level S G Parameters that affect priority adjustment. ξ C is the basic adjustment value, η C It is a nonlinear response coefficient adjusted according to the change of risk level, so that the impact of safety level can be adaptively adjusted according to the change of risk level.

[0197] Calculation example:

[0198] Set the following parameters: G = 8 for high risk level, S G =3 indicates high security requirements for the area. C = 1.2 adjustment factor to ensure that the risk level is appropriately reflected, ξ C =2.0,η C =0.5 adjusts the influence of the security level so that it increases nonlinearly with the increase of risk level.

[0199] Perform the calculation:

[0200]

[0201] The calculation result C = 10.6536, which represents the value obtained by adjusting the camera coverage priority under the current risk and security requirements. This result means that the surveillance coverage needs to be enhanced in this area to cope with higher security risks.

[0202] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. AI-based behavior recognition monitoring system, characterized by: The system comprises: The facial expression analysis module extracts video frames per second based on real-time monitoring video data, uses AI to identify facial areas, performs image segmentation on the detected facial areas, and extracts feature points from the facial areas. It uses feature point data to evaluate facial expressions, tracks facial expressions, analyzes emotional fluctuations, and obtains emotional state data. The behavior pattern recognition module extracts the speed and spatial position information of individual actions based on real-time monitoring video data, performs time window analysis on the information, compares the behavior features in the behavior pattern database, identifies the behavior type through the matching degree of the action speed and spatial position, and obtains the behavior pattern analysis results; The abnormal behavior determination module compares the behavior pattern analysis result with the preset normal behavior, analyzes the deviation between the behavior pattern and the normal pattern, quantifies the behavior deviation, and evaluates the abnormal level according to the deviation size to obtain abnormal behavior identification information; The monitoring and adjustment decision module evaluates the behavioral risk level based on the abnormal behavior identification information and emotional state data, combined with the required security level of the current area and the risk sensitivity of the current time period, implements the corresponding risk alarm notification, and adjusts the camera coverage priority of the area according to the risk level, optimizes the coverage efficiency, and obtains the adjusted monitoring strategy.

2. The AI-based behavior recognition monitoring system according to claim 1, characterized in that: The method for performing image segmentation on the detected facial area is: Based on real-time monitoring video data, extract video frames per second, use AI to identify facial areas, and obtain facial area recognition results; Based on the facial area recognition result, by formula: S=a S (F S -F0) 2 +b S (G S -G0) 2 +g S (H S -H0) 2 Calculate the segmentation score S of the pixel point, where α S , β S and γ S are the weight coefficients affecting the red, green, and blue channels, respectively. S , G S and H S Respectively represent the actual color values ​​of the image pixels in the red, green, and blue channels, and F0, G0, and H0 are the preset color thresholds; Based on the segmentation score S of the pixel point, it is compared with the preset segmentation threshold, and the pixel point with S value higher than the preset segmentation threshold is marked as face, and it is segmented from the background to obtain the face area segmentation result.

3. The AI-based behavior recognition monitoring system according to claim 2, characterized in that: The steps for acquiring the emotional state data are: Based on the facial region segmentation result, feature points are extracted from the facial region, and position changes of the feature points between consecutive frames are calculated to obtain facial expression change data; Based on the facial expression change data, the formula: Calculate the emotional fluctuation degree E of expression changes to obtain emotional state data, where σ x is the standard deviation of the horizontal displacement of the feature points, σ y is the standard deviation of the vertical displacement of the feature point, λ E is the weight coefficient of the horizontal displacement standard deviation, μ E is the weight coefficient of the standard deviation of vertical displacement, and E is the degree of emotional fluctuation of expression changes.

4. The AI-based behavior recognition monitoring system according to claim 1, characterized in that: The steps for obtaining the behavior pattern analysis result are: Based on real-time monitoring video data, the speed and spatial position information of individual actions are extracted, the position coordinates of the individual and the speed vector relative to the previous frame are calculated, and the behavior association data is obtained; Based on the behavior association data, it is compared with the known behavior patterns in the database, through the formula: Calculate the behavior matching degree M, where P(t) represents the spatial position of the individual at time t, V(t) represents the movement speed of the individual at time t relative to the previous frame, and V ref and P ref are the standard speed and position values ​​of known behaviors extracted from the database, κ v and η p is the weight coefficient, δ M and M is the attenuation parameter, M is the behavior matching degree, and e is the base of the natural logarithm; Based on the behavior matching degree M, it is compared with a preset matching degree threshold to determine whether the behavior matches, identify the behavior type, and obtain the behavior pattern analysis result.

5. The AI-based behavior recognition monitoring system according to claim 1, characterized in that: The method for quantifying behavioral deviation is: Based on the analysis results of the behavior pattern, compared with the preset normal behavior, the formula is: Calculate the quantitative measure Q of behavioral deviation, where B(t) is the observed behavior data at time t, N(t) is the normal behavior reference data at time t, ω t is the weight coefficient of time t, Q is the quantitative measure of behavioral deviation; The quantitative measurement Q based on the behavior deviation is compared with a preset behavior deviation threshold. If the Q value is greater than the preset behavior deviation threshold, it is judged that there is an obvious deviation in the behavior, and a behavior deviation analysis result is obtained.

6. The AI-based behavior recognition monitoring system according to claim 5, characterized in that: The steps for obtaining the abnormal behavior identification information are as follows: Based on the behavior deviation analysis results and the behavior pattern analysis results, behavior type weights are extracted to obtain abnormal level association data; Based on the abnormal level association data, by formula: F=ω type ·ln(1+Q) Calculate the abnormal behavior level F and obtain abnormal behavior identification information, where ω type is the behavior type weight coefficient, Q is the quantitative measure of behavior deviation, and F is the level of behavior abnormality.

7. The AI-based behavior recognition monitoring system according to claim 1, characterized in that: The method for implementing the corresponding risk alert notification is: Based on the abnormal behavior identification information and the emotional state data, the security level required for the current area and the risk sensitivity of the current period are extracted to obtain risk association data; Based on the risk association data, through the formula: G=(a G ·F+E)·ln(1+β G ·S G +g G ·R G ) Calculate the behavior risk level G, where E is the emotional fluctuation degree of expression changes, F is the abnormal behavior level, S G is the security level of the current area, R G is the risk sensitivity of the current period, G is the behavioral risk level, α G , β G and γ G is the weight coefficient, Based on the behavior risk level G, it is compared with the preset risk standard, and a corresponding risk alarm notification is implemented to obtain a risk alarm result.

8. The AI-based behavior recognition monitoring system according to claim 7, characterized in that: The method of the adjusted monitoring strategy is: Based on the risk alarm results, according to the abnormal behavior risk level of the area, the formula is: Calculate the camera coverage priority C, where ξ C and η C is the adjustment coefficient, G is the behavioral risk level, S G is the security level of the current area, C is the camera coverage priority, λ C is the weight coefficient; Based on the camera coverage priority C, according to the size of the C value, the camera coverage priority of the area is adjusted to optimize the coverage efficiency and obtain the adjusted monitoring strategy.

Images