Intelligent security and protection monitoring system and method based on computer vision

Abstract

The invention discloses an intelligent security and protection monitoring system and method based on computer vision, and belongs to the technical field of data processing.The method comprises the steps that firstly, a system performance state model, an efficiency state model and a value state model are constructed, and a system performance state coefficient, an efficiency state coefficient and a value state coefficient are output respectively; and the comprehensive operation state of the system is quantitatively evaluated. Furthermore, based on the effective alarm rate and the alarm response time under the performance and efficiency state coefficient, a quality-speed adaptation model is constructed, and the adaptation degree reflecting the balance degree of the system between the alarm quality and the response speed is output. And finally, combining the adaptation degree, the value state coefficient and the basic bandwidth occupancy peak value to construct a bandwidth optimization model, and dynamically outputting a target bandwidth occupancy peak value. According to the method, comprehensive evaluation of the multi-dimensional state of the security and protection system and self-adaptive optimal configuration of key resources are realized, and the intelligent level and the overall operation efficiency of the system are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to an intelligent security monitoring system and method based on computer vision. Background Technology

[0002] With the increasing demands for social security, traditional security monitoring systems are no longer sufficient to meet the requirements of modern security management. Traditional systems primarily rely on manual monitoring of video footage, which suffers from low efficiency, slow response times, and susceptibility to missed alarms due to fatigue. Therefore, the industry has begun researching intelligent security monitoring systems based on computer vision, aiming to achieve automated and intelligent security through artificial intelligence technology.

[0003] Currently, existing intelligent security technologies mostly focus on implementing single functions, such as simple target detection, facial recognition, or abnormal behavior analysis. These systems typically evaluate their performance metrics independently, such as detection accuracy or processing speed, lacking comprehensive modeling and coordinated optimization of the overall system operation. Furthermore, the allocation of system resources (such as network bandwidth) is often based on static policies or simple rules, failing to dynamically adjust according to real-time system performance, operational efficiency, and business value. This results in low resource utilization efficiency and makes it difficult to maintain optimal performance in complex and ever-changing real-world application scenarios.

[0004] Therefore, the existing technology has the following main defects: First, the state assessment of multiple dimensions such as system performance, operational efficiency and business value is fragmented and lacks unified quantification and comprehensive correlation analysis; Second, the configuration of key system parameters (such as bandwidth) is usually static or semi-static and cannot be adaptively and dynamically adjusted according to the internal state of the system (such as the balance between quality and speed, and the level of business value), thus limiting the maximization of the overall system efficiency. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent security monitoring system and method based on computer vision, which solves the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent security monitoring method based on computer vision, comprising the following steps: The system performance status model is constructed based on the target detection rate (the ratio of the number of detected targets to the total number of actual targets in a specific time period), the recognition accuracy (the ratio of the number of times the system correctly identifies targets to the total number of times targets are identified), and the average processing delay, and the system performance status coefficient is output. An efficiency status model is constructed based on the time saved by manpower (the number of hours saved by the security team each week by replacing manual monitoring with automatic analysis) and the efficiency improvement of intelligent retrieval (the ratio of the time required for manual retrieval to the time required for intelligent retrieval), and the efficiency status coefficient is output. A value state model is constructed based on the key area control coverage rate (the ratio of the number of cameras connected to the intelligent analysis system to the total number of cameras in all key areas that should be controlled), the number of security incidents (the number of specific types of security incidents (such as illegal intrusion, theft, and damage) after the system is deployed) and the security incident reduction rate (the percentage decrease compared to the same period before deployment), and the value state coefficient is output. A quality-speed adaptation model is constructed based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. Based on quality-speed fit, value state coefficient, and basic bandwidth usage peak (the highest network bandwidth occupied by the system when transmitting video streams and alarm data), a bandwidth usage peak optimization model is constructed to output the target bandwidth usage peak.

[0007] Based on the above technical solutions, the present invention also provides the following optional technical solutions: A further technical solution: The bandwidth occupancy peak optimization model is expressed as follows:

[0008] in, Indicates the peak value of the target bandwidth usage. Indicates the peak value of the base bandwidth usage. Indicates the adjustment range coefficient. Indicates the fit response coefficient. Indicates quality-speed fit. Indicates the value compensation coefficient. Represents the value state coefficient. This represents the baseline threshold for the value state coefficient.

[0009] Further technical solution: Based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, construct a quality-speed adaptation model, and output the quality-speed adaptation degree. The steps are as follows: The effective alarm rate and alarm response time are subjected to maximum-min normalization to obtain the effective alarm rate index and alarm response time index. A quality-speed adaptation model is constructed based on the effective alarm rate index and alarm response time index under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. The quality-speed adaptation model is expressed as follows:

[0010] in, Indicates quality-speed fit. Represents the system performance state coefficient. This represents the efficiency state coefficient. This represents the effective alarm rate index. This indicates the alarm response time index. Indicates the prevention of decimals except zero (usually 10 -6 ), the ,when At that time, the system strongly favors quality, sacrificing speed to ensure alarm quality. At that time, mass and velocity reach an ideal balance. At that time, the system strongly favors the speed side, sacrificing quality to ensure response speed.

[0011] Further technical solutions: Based on the key area control coverage rate (the ratio of the number of cameras connected to the intelligent analysis system to the total number of cameras in all key areas that should be controlled), the number of security incidents (the number of specific types of security incidents (such as illegal intrusion, theft, and vandalism) after the system is deployed), and the security incident reduction rate (the percentage decrease compared to the same period before deployment), the steps to construct a value state model and output the value state coefficients are as follows: The control coverage rate, the number of security incidents, and the rate of decrease of security incidents in key areas are subjected to maximum-min normalization to obtain the control coverage rate index, the number of security incidents index, and the rate of decrease of security incidents index. A value state model is constructed based on the control coverage index, the security incident occurrence index, and the security incident decline rate index, and the value state coefficients are obtained. The value state model is expressed as follows:

[0012] in, Represents the value state coefficient. This represents the control coverage index. This represents an index indicating the number of security incidents. Indicators representing the rate of decrease in security incidents Represents the weight coefficient and The The higher the value, the higher the system value.

[0013] Further technical solutions: Based on the time saved by manpower (the number of hours saved for the security team each week by automatically analyzing and replacing manual monitoring) and the efficiency improvement of intelligent retrieval (the ratio of the time required for manual retrieval to the time required for intelligent retrieval), an efficiency status model is constructed, and the steps to output the efficiency status coefficient are as follows: The ratio of labor saving time and intelligent retrieval efficiency improvement to the corresponding maximum values ​​is used to obtain the labor saving index and efficiency improvement index. An efficiency state model is constructed based on the time-saving index and the efficiency improvement index, and the efficiency state coefficient is obtained. The efficiency state model is expressed as follows:

[0014] in, This represents the efficiency state coefficient. Indicates the time-saving index. Indicator of efficiency improvement This represents the overall sensitivity coefficient. Represents the weight coefficient and The Furthermore, the larger the value, the higher the system efficiency.

[0015] Further technical solutions: The steps for constructing a system performance state model and outputting system performance state coefficients based on the target detection rate (the ratio of the number of detected targets to the total number of actual targets within a specific time period), the recognition accuracy (the ratio of the number of times the system correctly identifies targets to the total number of times targets are identified), and the average processing latency are as follows: The target detection rate, recognition accuracy, and average processing delay are subjected to maximum-minimum normalization to obtain the target detection rate index, recognition accuracy index, and processing delay index. A system performance state model is constructed based on the target detection rate index, the recognition accuracy index, and the processing latency index, and the system performance state coefficients are obtained. The system performance state model is expressed as follows:

[0016] in, Represents the system performance state coefficient. This represents the target detection rate index. This represents the recognition accuracy index. Indicates the processing latency index. Represents the weight coefficient and The The larger the value, the better the system performance.

[0017] A computer vision-based intelligent security monitoring system employs the aforementioned computer vision-based intelligent security monitoring method.

[0018] This invention provides an intelligent security monitoring system and method based on computer vision, which has the following advantages compared with the prior art: 1. This invention achieves a comprehensive quantitative assessment of the multi-dimensional operating status of security systems by constructing system performance status models, efficiency status models, and value status models. It breaks through the limitations of traditional single-indicator assessment and provides comprehensive and accurate data support for system optimization. 2. This invention accurately depicts the balance between alarm quality and response speed of the system by constructing a quality-speed adaptation model. Its output can be directly used to guide the optimization direction of system parameters, effectively solving the industry problem of difficulty in balancing quality and speed. 3. This invention achieves dynamic and adaptive adjustment of network bandwidth resources by constructing a peak bandwidth utilization optimization model. This model can intelligently provide optimal bandwidth configuration suggestions based on the system's real-time adaptation status and service value, thereby significantly improving network resource utilization efficiency while ensuring core service performance. 4. This invention deeply integrates computer vision technology, system state modeling, and resource optimization control to form a complete technical closed loop from perception and analysis to decision-making and optimization, which significantly improves the automation and intelligence level of intelligent security monitoring systems. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0022] Please see Figure 1 The present invention provides an intelligent security monitoring method based on computer vision, comprising the following steps: The system performance status model is constructed based on the target detection rate (the ratio of the number of detected targets to the total number of actual targets in a specific time period), the recognition accuracy (the ratio of the number of times the system correctly identifies targets to the total number of times targets are identified), and the average processing delay, and the system performance status coefficient is output. An efficiency status model is constructed based on the time saved by manpower (the number of hours saved by the security team each week by replacing manual monitoring with automatic analysis) and the efficiency improvement of intelligent retrieval (the ratio of the time required for manual retrieval to the time required for intelligent retrieval), and the efficiency status coefficient is output. A value state model is constructed based on the key area control coverage rate (the ratio of the number of cameras connected to the intelligent analysis system to the total number of cameras in all key areas that should be controlled), the number of security incidents (the number of specific types of security incidents (such as illegal intrusion, theft, and damage) after the system is deployed) and the security incident reduction rate (the percentage decrease compared to the same period before deployment), and the value state coefficient is output. A quality-speed adaptation model is constructed based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. Based on quality-speed fit, value state coefficient, and basic bandwidth usage peak (the highest network bandwidth occupied by the system when transmitting video streams and alarm data), a bandwidth usage peak optimization model is constructed to output the target bandwidth usage peak.

[0023] Through the above technical solutions, this application solves the problem of unreasonable resource allocation caused by fragmented multi-dimensional state assessment, and realizes dynamic optimization of network bandwidth configuration. The system can automatically balance alarm quality and response speed based on real-time performance status, improving resource utilization while ensuring core security functions. By quantifying manpower savings and improving retrieval efficiency, it provides data support for operational decisions and avoids biases from human experience-based judgment. The linkage analysis between key area control coverage and security incidents can accurately assess the system deployment effect and guide the direction of subsequent equipment upgrades and algorithm optimization.

[0024] Preferably, the steps for constructing a system performance state model and outputting system performance state coefficients based on the target detection rate (the ratio of the number of detected targets to the total number of actual targets in a specific time period), the recognition accuracy (the ratio of the number of times the system correctly identifies targets to the total number of times targets are identified), and the average processing latency are as follows: The target detection rate, recognition accuracy, and average processing delay are subjected to maximum-minimum normalization to obtain the target detection rate index, recognition accuracy index, and processing delay index. A system performance state model is constructed based on the target detection rate index, the recognition accuracy index, and the processing latency index, and the system performance state coefficients are obtained. The system performance state model is expressed as follows:

[0025] in, Represents the system performance state coefficient. This represents the target detection rate index. This represents the recognition accuracy index. Indicates the processing latency index. Represents the weight coefficient and The The larger the value, the better the system performance.

[0026] Among them, the target detection rate refers to the ratio of the number of targets detected by the system to the total number of actual targets within a specific time period, reflecting the system's detection coverage capability. Specifically, it can be achieved by combining frame-by-frame analysis of the video stream with statistical analysis of target tracking algorithms, used to quantify the system's ability to capture targets in the monitored scene. The recognition accuracy rate refers to the ratio of the number of times the system correctly identifies a target or event to the total number of recognitions, reflecting the algorithm's recognition precision. Specifically, it can be calculated by comparing the labeled test dataset with the system output results, used to evaluate the system's accuracy in judging target attributes. The average processing latency refers to the average time from receiving video data to outputting analysis results, reflecting real-time response capability. Specifically, it can be achieved by using... The timestamp records the time consumed in each processing step and calculates the average to measure the system's processing efficiency. The maximum-minimum normalization process refers to linearly mapping the original indicators to the [0,1] interval to eliminate the influence of different dimensions on the fusion calculation. Specifically, it can be achieved by collecting historical data to determine the maximum and minimum values ​​of each indicator as the normalization benchmark, which is used to make the detection rate, accuracy and latency indicators comparable. The weighted geometric mean model refers to the exponential fusion of multi-dimensional indicators to strengthen the contribution of key indicators. Specifically, it can be achieved by setting weight coefficients and calculating the power-weighted product of the normalized indicators, which is used to highlight core performance elements while balancing detection capability, recognition accuracy and response speed.

[0027] Specifically, the target detection rate index and the recognition accuracy index are converted into dimensionless values ​​through normalization, while the processing delay index is converted into a dimensionless value through... The form is transformed into a positive index. The system performance state model adopts a weighted geometric mean form, assigning weight coefficients to the three indices. , , This amplifies the influence of high-weighted indicators through exponential calculations, while utilizing the properties of geometric averages to suppress extreme value interference. For example, when identifying accuracy weights... At higher accuracy levels, the model becomes more sensitive to changes in accuracy, and the recognition algorithm can be optimized first to improve overall performance. By adjusting the weight allocation, the system can flexibly adapt to the different needs for detection coverage, recognition accuracy, or response speed in different scenarios.

[0028] Compared to existing technologies, traditional methods typically select only a single indicator to evaluate system performance. For example, they may focus solely on the target detection rate while ignoring the impact of processing latency on real-time performance, or optimize recognition accuracy alone, leading to excessive resource consumption. This solution, through multi-indicator normalization and a nonlinear fusion model, incorporates detection capability, recognition accuracy, and response speed into a unified evaluation system, thus addressing the limitations of single-indicator evaluation. Existing static threshold judgment methods cannot reflect the dynamic correlation between various indicators, while this solution dynamically adjusts the contribution of each dimension to performance evaluation through weighting coefficients, enabling performance coefficients to accurately reflect the overall operating status of the system.

[0029] Through the above technical solution, this application achieves multi-dimensional dynamic evaluation of the performance of security monitoring systems, providing a quantitative basis for network bandwidth allocation and computing resource scheduling. For example, in nighttime monitoring scenarios, when the system performance coefficient decreases due to a drop in target detection rate, a bandwidth optimization model can be triggered to reduce the resolution of video streams in non-critical areas, prioritizing analysis resources for key areas. This solution overcomes the resource allocation imbalance problem caused by the single performance evaluation dimension in traditional methods, enabling the system to adaptively adjust resource configuration strategies based on real-time performance status.

[0030] Preferably, the steps for constructing an efficiency state model based on manpower saving time (the number of hours saved for the security team each week by automatically analyzing and replacing manual monitoring) and intelligent retrieval efficiency improvement (the ratio of time required for manual retrieval to time required for intelligent retrieval), and outputting the efficiency state coefficient, are as follows: The ratio of labor saving time and intelligent retrieval efficiency improvement to the corresponding maximum values ​​is used to obtain the labor saving index and efficiency improvement index. An efficiency state model is constructed based on the time-saving index and the efficiency improvement index, and the efficiency state coefficient is obtained. The efficiency state model is expressed as follows:

[0031] in, This represents the efficiency state coefficient. Indicates the time-saving index. Indicator of efficiency improvement This represents the overall sensitivity coefficient. Represents the weight coefficient and The Furthermore, the larger the value, the higher the system efficiency.

[0032] Among these metrics, "Manpower Saving Time" refers to the number of hours saved for the security team each week by replacing manual monitoring with automated analysis. This can be achieved through cross-validation using weekly work hour statistics and system logs. This parameter reflects the automation technology's ability to replace manual monitoring. "Intelligent Retrieval Efficiency Improvement" is the ratio of the time required for manual retrieval to the time required for intelligent retrieval. This can be calculated by comparing historical work order records with system response logs. This parameter reflects the time efficiency advantage of intelligent retrieval technology. "Manpower Saving Time Index" is the ratio of manpower saving time to a preset maximum saving time. This can be achieved by setting an industry benchmark or historical system peak value as the denominator to eliminate differences in dimensions across different scale scenarios. "Efficiency Improvement Index" is the ratio of intelligent retrieval efficiency improvement to a preset maximum improvement. This can be achieved by using the best performance of similar systems as a reference benchmark to achieve cross-scenario comparability. Weighting coefficients are used to adjust the contribution ratio of saving time and efficiency improvement in the comprehensive index. These values ​​can be determined using the analytic hierarchy process or expert experience to adapt the model to the operational needs of different scenarios. The overall sensitivity coefficient controls the steepness of the logic function curve. Specifically, the parameter value can be adjusted through gradient testing to match the efficiency variation of systems of different sizes.

[0033] Specifically, this technical solution first standardizes the labor saving time and the improvement in intelligent retrieval efficiency, transforming them into an index form ranging from 0 to 1, thus addressing the issue of inconsistent dimensions in multi-dimensional operational data. By introducing weighting coefficients to linearly weight the two types of indices, a comprehensive index is formed, allowing the model to adjust its evaluation focus according to the needs of actual operational scenarios. For example, in scenarios with high labor cost pressures, the weighting coefficient of the labor saving index can be increased, while in scenarios requiring rapid response to security incidents, the weighting coefficient of the efficiency improvement index can be increased. Furthermore, a logistic function is used to non-linearly map the weighted comprehensive index, generating an efficiency state coefficient ranging from 0 to 1. This function's characteristics ensure that when the comprehensive index reaches a critical value, the efficiency state coefficient exhibits a significant change, accurately reflecting the transition state of system efficiency. The introduction of a comprehensive sensitivity coefficient allows the model to adjust the sensitivity of efficiency assessment according to system scale; for example, in large monitoring networks, this coefficient can be increased to capture subtle efficiency fluctuations.

[0034] Compared to existing technologies, traditional methods typically assess efficiency using only a single operational metric, such as manpower savings or retrieval speed, which fails to comprehensively reflect the overall system efficiency. The linear superposition models used in existing technologies struggle to handle non-linear relationships between metrics and lack adaptive adjustment capabilities to varying scenarios. This solution constructs a dynamic quantitative model through multi-dimensional metric fusion and non-linear function mapping, enabling multi-level and accurate evaluation of operational efficiency.

[0035] Through the above technical solution, this application solves the problems of traditional intelligent security monitoring systems, such as a single dimension for operational efficiency evaluation and a lack of dynamic quantitative models, achieving a dual-dimensional integrated evaluation of manpower savings and improved retrieval efficiency. Through standardized processing and weight adjustment mechanisms, the model can adapt to the operational characteristics of different scenarios and accurately quantify the system's efficiency status. Based on the nonlinear mapping relationship of logistic functions, it effectively captures the critical change characteristics of efficiency indicators, providing reliable dynamic input parameters for downstream resource optimization, thereby improving the adaptive allocation accuracy of resources such as network bandwidth.

[0036] Preferably, the steps for constructing a value state model and outputting value state coefficients based on the key area control coverage rate (the ratio of the number of cameras connected to the intelligent analysis system to the total number of cameras in all key areas that should be controlled), the number of security incidents (the number of specific types of security incidents (such as illegal intrusion, theft, and vandalism) after the system is deployed), and the security incident reduction rate (the percentage decrease compared to the same period before deployment) are as follows: The control coverage rate, the number of security incidents, and the rate of decrease of security incidents in key areas are subjected to maximum-min normalization to obtain the control coverage rate index, the number of security incidents index, and the rate of decrease of security incidents index. A value state model is constructed based on the control coverage index, the security incident occurrence index, and the security incident decline rate index, and the value state coefficients are obtained. The value state model is expressed as follows:

[0037] in, Represents the value state coefficient. This represents the control coverage index. This represents an index indicating the number of security incidents. Indicators representing the rate of decrease in security incidents Represents the weight coefficient and The The higher the value, the higher the system value.

[0038] The key area control coverage rate refers to the ratio of the number of cameras connected to the intelligent analysis system to the total number of cameras in all key areas subject to control. This can be calculated by matching online camera status monitoring data with a pre-defined list of key areas, reflecting the system's coverage capability in critical areas. The security incident occurrence rate refers to the number of specific types of security incidents that occur after system deployment. This can be achieved by counting the number of triggers for pre-defined event types such as illegal intrusion, theft, and vandalism using the event log statistics module. This indicator is converted into a positive index through reverse mapping to meet comprehensive evaluation requirements. The security incident reduction rate is the percentage decrease compared to the same period before deployment. This can be calculated by year-on-year using historical event data from the same period in the database, quantifying the security improvement effect after system deployment. Weighting coefficients are used to adjust the contribution of different indicators. Specifically, the relative importance of each indicator can be determined using the analytic hierarchy process or expert experience, allowing the model to adapt to business needs in different scenarios.

[0039] Specifically, the key area control coverage index eliminates the dimensional differences in the original data through max-min normalization. For example, when the actual number of cameras connected in a certain area is 80, but the total number to be controlled is 100, the index can be mapped to 0.8. The security incident occurrence index is transformed into a positive indicator through reverse calculation. For example, when the number of security incidents in a certain week is 5 and the historical maximum is 20, the original index after normalization is 0.25, which is then transformed into a positive contribution value of 1-0.25=0.75. The security incident reduction rate index directly reflects the degree of improvement. For example, when the number of incidents decreases from 50 to 30 per month after deployment, the reduction rate of 40% can be normalized to an index of 0.4. The three indices are merged into a unified value state coefficient through a weighted summation model. The weight coefficients can be dynamically adjusted according to the security level. For example, in key facility scenarios, the weight of control coverage can be increased to 0.5, while in high-crime areas, the weight of security incident reduction rate can be increased to 0.6. The resulting value state coefficient can dynamically reflect the overall value level of the system under different operational stages and business scenarios.

[0040] Compared to existing technologies, traditional methods typically use only a single indicator to evaluate system value, such as focusing solely on camera coverage or incident decline rate, resulting in an inability to comprehensively reflect multi-dimensional business value. The static weight allocation methods used in existing technologies are ill-suited to the evaluation needs of different scenarios, such as the different priorities for control coverage and incident response speed between campus security and financial security. By employing multi-indicator normalization and a dynamic weight allocation mechanism, the value assessment model can eliminate the dimensional differences between indicators and adjust the evaluation focus according to actual business needs.

[0041] Through the above technical solution, this application solves the problem of the single dimension and lack of dynamic adaptability in the business value assessment of existing technologies, and realizes a multi-dimensional comprehensive quantitative assessment of the coverage capability of key areas, the effectiveness of security incident control, and the improvement trend. By using a reverse mapping mechanism to transform the occurrence of security incidents into positive contribution indicators, the assessment distortion caused by directly using raw data in high-incidence scenarios is avoided. The dynamic weight allocation mechanism allows the model to adjust the indicator weights according to the characteristics of different security scenarios. For example, in the scenario of key cultural relic protection, the weight of the control coverage rate can be increased to 0.6, while in the scenario of commercial complexes, the weight of the security incident reduction rate can be increased to 0.5, thereby providing accurate value state input for subsequent resource allocation decisions such as bandwidth optimization.

[0042] Preferably, the steps for constructing a quality-speed adaptation model based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, and outputting the quality-speed adaptation degree, are as follows: The effective alarm rate and alarm response time are subjected to maximum-min normalization to obtain the effective alarm rate index and alarm response time index. A quality-speed adaptation model is constructed based on the effective alarm rate index and alarm response time index under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. The quality-speed adaptation model is expressed as follows:

[0043] in, Indicates quality-speed fit. Represents the system performance state coefficient. This represents the efficiency state coefficient. This represents the effective alarm rate index. This indicates the alarm response time index. Indicates the prevention of decimals except zero (usually 10 -6 ), the ,when At that time, the system strongly favors quality, sacrificing speed to ensure alarm quality. At that time, mass and velocity reach an ideal balance. At that time, the system strongly favors the speed side, sacrificing quality to ensure response speed.

[0044] The effective alarm rate index refers to the alarm effectiveness indicator after eliminating dimensional differences through maximum-minimum normalization. Specifically, it can be achieved by linearly mapping the actual effective alarm rate to the ratio of historical maximum and minimum values, reflecting the actual level of alarm quality. The alarm response time index refers to the response speed indicator after maximum-minimum normalization. Specifically, it can be achieved by standardizing the ratio of the actual response time to preset optimal and worst response times, quantifying system response efficiency. The system performance status coefficient characterizes the system's comprehensive detection and identification capabilities. The efficiency status coefficient quantifies the system resource optimization effect. Zero-decimal prevention refers to an extremely small constant used to avoid zero denominators, specifically a value on the order of 10^-6, to ensure the stability of formula calculations.

[0045] Specifically, the effective alarm rate and alarm response time are first normalized and converted into comparable exponential forms. The system performance state coefficient and efficiency state coefficient serve as dynamic weighting factors, respectively affecting the effective alarm rate exponent and the alarm response time exponent. In the quality-speed adaptation model, the product of the performance coefficient and the effective alarm rate exponent reflects the system's priority control over alarm quality, while the product of the efficiency coefficient and the response time exponent reflects the optimization requirements for response speed. The game relationship between quality and speed is quantified through the difference operation of the numerators, and the output is constrained to the [-1, 1] interval using a hyperbolic tangent function. When the system's detection capability improves, the performance coefficient increases, pushing the adaptation degree closer to 1, prompting the system to prioritize alarm accuracy; when resource savings are significant, the efficiency coefficient increases, pushing the adaptation degree closer to -1, prompting the system to prioritize shortening the response time. When the two are dynamically balanced, the adaptation degree approaches 0, achieving real-time balanced adjustment of quality and speed.

[0046] Compared to existing technologies, traditional methods typically use fixed thresholds or single indicators to control the balance between alarm quality and response speed. For example, they adjust processing priorities only based on a preset alarm delay threshold, resulting in the system's inability to adapt to performance fluctuations or efficiency changes. This solution introduces a dynamic weighting mechanism for performance and efficiency state coefficients, and establishes a correlation model between quality and speed using nonlinear functions. This allows bandwidth resource allocation to be automatically adjusted according to the real-time status of the system, overcoming the resource waste problem caused by static configuration.

[0047] Through the above technical solution, this application can dynamically adjust the priority of quality and speed in the alarm processing process based on real-time changes in system detection capabilities and resource efficiency, thereby optimizing the allocation strategy of network bandwidth resources. For example, when security demands are low at night, the adaptation can automatically shift towards speed to reduce bandwidth consumption; when anomalies occur in key areas, the adaptation can quickly shift towards quality to improve alarm accuracy. This dynamic balancing mechanism effectively avoids the problems of bandwidth resource waste or missed reporting of critical events caused by fixed strategies in traditional systems.

[0048] Preferably, the peak bandwidth occupancy optimization model is expressed as:

[0049] in, Indicates the peak value of the target bandwidth usage. Indicates the peak value of the base bandwidth usage. Indicates the adjustment range coefficient. Indicates the fit response coefficient. Indicates quality-speed fit. Indicates the value compensation coefficient. Represents the value state coefficient. This represents the baseline threshold for the value state coefficient.

[0050] The peak bandwidth usage refers to the initial bandwidth usage limit without dynamic adjustment, which can be determined, for example, through historical data statistics. The adjustment amplitude coefficient controls the influence of quality-speed fit on bandwidth adjustment, and can be generated using preset empirical values ​​or adaptive learning algorithms. The fit response coefficient adjusts the non-linear mapping relationship of quality-speed fit, and can be determined, for example, by calibrating its value range experimentally or assigning values ​​based on expert experience. Quality-speed fit reflects the balance between alarm quality and response speed, and can be calculated from the normalized alarm rate and response time. The value compensation coefficient adjusts the compensation amplitude of the value status coefficient on bandwidth, and can be determined using preset empirical values, for example, based on business priority. The value status coefficient benchmark threshold is used to determine whether the system value meets the preset standard, for example, by using historical security event data. Specifically, this technical solution maps quality-speed fit to a smoothly varying adjustment factor using a hyperbolic tangent function, avoiding network instability caused by sudden bandwidth changes. The fit response coefficient controls the sensitivity of the quality-speed balance state to bandwidth adjustments; for example, when the fit is high, the system prioritizes alarm quality, and bandwidth demand increases with the improvement in fit. The adjustment amplitude coefficient limits the maximum range of bandwidth adjustment to prevent excessive resource consumption. The value compensation mechanism dynamically adjusts the bandwidth allocation strategy by comparing the real-time value state with a baseline threshold. For example, when the value state coefficient exceeds the baseline threshold, it indicates a high security value in the current scenario, and the system automatically increases bandwidth allocation to enhance protection capabilities. Compared to existing technologies, traditional bandwidth allocation strategies typically employ fixed thresholds or simple linear rules, failing to adapt to changes in system operating states. This solution, however, introduces a nonlinear dynamic adjustment model, combining the quality-speed balance with security value assessment to achieve precise matching of bandwidth resources with real-time demands. Existing technologies lack a quantitative compensation mechanism for security value states, while this solution, through comparison of benchmark thresholds for value state coefficients, can automatically optimize bandwidth allocation priorities in high-value scenarios. Through the above technical solution, this application can dynamically adjust the peak network bandwidth usage based on the real-time balance between alarm processing quality and response speed, avoiding resource waste or insufficient supply. In high-security-value scenarios, the system automatically increases bandwidth allocation to enhance the security capabilities of critical areas, while maintaining a basic bandwidth level in normal scenarios. This solution effectively solves the problem of low resource utilization caused by static bandwidth allocation strategies, achieving adaptive matching between network resources and system operating status.

[0051] A computer vision-based intelligent security monitoring system employs the aforementioned computer vision-based intelligent security monitoring method.

[0052] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0053] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A computer vision-based intelligent security monitoring method, characterized in that, Includes the following steps: A system performance state model is constructed based on the target detection rate, recognition accuracy, and average processing delay, and the system performance state coefficient is output. An efficiency status model is constructed based on the time saved by human resources and the improvement of intelligent retrieval efficiency, and the efficiency status coefficient is output. A value state model is constructed based on the coverage rate of key area control, the number of security incidents, and the rate of decrease in security incidents, and the value state coefficient is output. A quality-speed adaptation model is constructed based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. Based on quality-speed fit, value state coefficient, and basic bandwidth usage peak, a bandwidth usage peak optimization model is constructed to output the target bandwidth usage peak.

2. The intelligent security monitoring method based on computer vision according to claim 1, characterized in that, The peak bandwidth utilization optimization model is expressed as follows: in, Indicates the peak value of the target bandwidth usage. Indicates the peak value of the base bandwidth usage. Indicates the adjustment range coefficient. Indicates the fit response coefficient. Indicates quality-speed fit. Indicates the value compensation coefficient. Represents the value state coefficient. This represents the baseline threshold for the value state coefficient.

3. The intelligent security monitoring method based on computer vision according to claim 2, characterized in that, The steps for constructing a quality-speed fit model based on the effective alarm rate and alarm response time under the system performance state coefficient and efficiency state coefficient, and outputting the quality-speed fit degree are as follows: The effective alarm rate and alarm response time are subjected to maximum-min normalization to obtain the effective alarm rate index and alarm response time index. A quality-speed adaptation model is constructed based on the effective alarm rate index and alarm response time index under the system performance state coefficient and efficiency state coefficient, and the quality-speed adaptation degree is output. The quality-speed adaptation model is expressed as follows: in, Indicates quality-speed fit. Represents the system performance state coefficient. This represents the efficiency state coefficient. This represents the effective alarm rate index. This indicates the alarm response time index. Indicates the prevention of decimal division by zero, the stated ,when At that time, the system strongly favors quality, sacrificing speed to ensure alarm quality. At that time, mass and velocity reach an ideal balance. At that time, the system strongly favors the speed side, sacrificing quality to ensure response speed.

4. The intelligent security monitoring method based on computer vision according to claim 3, characterized in that, The steps for constructing a value state model based on the key area control coverage rate, the number of security incidents, and the security incident decline rate, and outputting the value state coefficients, are as follows: The control coverage rate, the number of security incidents, and the rate of decrease of security incidents in key areas are subjected to maximum-min normalization to obtain the control coverage rate index, the number of security incidents index, and the rate of decrease of security incidents index. A value state model is constructed based on the control coverage index, the security incident occurrence index, and the security incident decline rate index, and the value state coefficients are obtained. The value state model is expressed as follows: in, Represents the value state coefficient. This represents the control coverage index. This represents an index indicating the number of security incidents. Indicators representing the rate of decrease in security incidents Represents the weight coefficient and The The higher the value, the higher the system value.

5. The intelligent security monitoring method based on computer vision according to claim 3, characterized in that, The steps for constructing an efficiency state model based on labor saving time and improved intelligent retrieval efficiency, and outputting efficiency state coefficients, are as follows: The ratio of labor saving time and intelligent retrieval efficiency improvement to the corresponding maximum values ​​is used to obtain the labor saving index and efficiency improvement index. An efficiency state model is constructed based on the time-saving index and the efficiency improvement index, and the efficiency state coefficient is obtained. The efficiency state model is expressed as follows: in, This represents the efficiency state coefficient. Indicates the time-saving index. Indicator of efficiency improvement This represents the overall sensitivity coefficient. Represents the weight coefficient and The Furthermore, the larger the value, the higher the system efficiency.

6. The intelligent security monitoring method based on computer vision according to claim 3, characterized in that, The steps to construct a system performance state model based on target detection rate, recognition accuracy, and average processing latency, and then output the system performance state coefficients, are as follows: The target detection rate, recognition accuracy, and average processing delay are subjected to maximum-minimum normalization to obtain the target detection rate index, recognition accuracy index, and processing delay index. A system performance state model is constructed based on the target detection rate index, the recognition accuracy index, and the processing latency index, and the system performance state coefficients are obtained. The system performance state model is expressed as follows: in, Represents the system performance state coefficient. This represents the target detection rate index. This represents the recognition accuracy index. Indicates the processing latency index. Represents the weight coefficient and The The larger the value, the better the system performance.

7. A computer vision-based intelligent security monitoring system, characterized in that, The intelligent security monitoring method based on computer vision described in any one of claims 1-6 is adopted.

Images