CT Self-Powering Video Monitoring and Warning Method and Device Based on Environmental Perception

Abstract

The present invention discloses a method and device for CT self-powered video monitoring and early warning based on environmental perception. The method obtains historical working data of CT self-powered video monitoring in a target area, constructs a working task recognition model based on the hidden Markov model; uses the model to identify working tasks within a preset time period, analyzes power changes and evaluates working quality, and constructs a working task-power-quality data matrix; performs cluster division based on the matrix to determine a monitoring task allocation scheme; obtains video monitoring data according to the scheme, identifies abnormal events, and conducts monitoring and early warning. The present invention optimizes video monitoring task allocation, improves monitoring efficiency and early warning accuracy, and is applicable to the fields of intelligent monitoring and power management.

Description

Technical Field

[0001] The present invention relates to the technical field of video monitoring and early warning, and particularly relates to a CT self-powered video monitoring and early warning method and device based on environmental perception. Background Art

[0002] With the rapid development of intelligent monitoring technology, intelligent early warning systems based on video monitoring have been widely used in fields such as urban management, power inspection, and security monitoring. Traditional video monitoring systems mainly rely on fixed power supplies for power, and manage device status through regular inspections or remote control. However, in some special application scenarios, such as remote areas, power facility monitoring, forest fire prevention, etc., the traditional power supply mode is difficult to meet the demand for long-term autonomous operation. Therefore, self-powered technology has gradually become an important means to solve the power supply problem of monitoring devices.

[0003] CT (Current Transformer) self-powered technology uses the electromagnetic induction principle in the transmission line to power the monitoring device, enabling it to operate stably for a long time without external power supply. However, limited by grid load fluctuations, environmental factors, and the power consumption characteristics of the device itself, the power supply capacity of CT self-powered video monitoring devices has certain uncertainties, which may lead to fluctuations in the device operation state, and further affect the execution quality of the monitoring task. How to reasonably allocate video monitoring tasks under the condition of limited power supply to ensure monitoring quality and early warning reliability has become an important research direction for the optimization of current intelligent monitoring systems.

[0004] In the prior art, some intelligent monitoring systems already have basic task management and anomaly detection functions, but usually lack comprehensive perception of environmental factors and power status, and cannot dynamically adjust task allocation according to the remaining power of the monitoring device, resulting in some devices being unable to execute high-energy-consuming tasks due to insufficient power, or having monitoring blind spots during task execution, reducing the monitoring coverage rate and early warning accuracy. In addition, traditional anomaly detection methods are mostly based on fixed threshold judgment or simple pattern matching, without fully combining monitoring task characteristics and environmental change factors, and are prone to false alarms or missed alarms, affecting the accurate identification and rapid response of abnormal events.

[0005] To address the above problems, there is an urgent need for a CT self-powered video monitoring and early warning method and device based on environmental perception, which can improve the energy utilization efficiency of the video monitoring system, enhance the ability to identify abnormal events, and achieve efficient and accurate monitoring and early warning through intelligent task recognition and allocation strategies. Summary of the Invention

[0006] In order to solve at least one of the above technical problems, the present invention proposes a CT self-powered video monitoring and early warning method and device based on environmental perception.

[0007] The first aspect of the present invention provides a method for monitoring and warning of CT self-powered video based on environmental perception, including:

[0008] Obtain the historical working data of each CT self-powered video monitor in the target area, analyze the historical working data based on the hidden Markov model, and construct a working task recognition model for video monitoring;

[0009] According to the working task recognition model, identify the working tasks of each video monitor within a preset time period, obtain the power change data of the video monitor within the preset time period, and evaluate the working quality of each working task of the video monitor to construct a working task - power - quality data matrix;

[0010] According to the working task - power - quality data matrix, perform cluster division on each video monitor, determine the monitoring tasks of each monitoring cluster, and obtain a monitoring task allocation plan;

[0011] Obtain the video monitoring data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video monitoring data, and perform monitoring and warning operations according to the abnormal event identification results.

[0012] In this solution, the obtaining of the historical working data of each CT self-powered video monitor in the target area, analyzing the historical working data based on the hidden Markov model, and constructing a working task recognition model for video monitoring is specifically as follows:

[0013] Obtain the historical working data of each CT self-powered video monitor in the target area. The historical working data includes the working current and voltage data, processor load status, camera movement trajectory parameters during the operation of the video monitor, and the working task labels marked at the corresponding moments. The working task labels include target tracking and abnormal detection;

[0014] Perform preprocessing of timestamp alignment and missing value interpolation on the historical working data. After being segmented into continuous time slot segments by a sliding window, normalize the working current and voltage data, processor load status, and camera movement trajectory parameters to generate an observation sequence of the input data type of the hidden Markov model;

[0015] Define the set of hidden states of the hidden Markov model as various types of working task types marked in the working task labels, and import the observation sequence into the hidden Markov model as the observation state and the set of hidden states of the hidden Markov model;

[0016] Calculate the initial state transition probability matrix and observation probability matrix of the hidden Markov model through the forward-backward algorithm, use the labeled work task tags to perform supervised constraints on the true hidden state of each time slot segment, and adopt the Baum-Welch algorithm with label correction to iteratively optimize the distribution parameters of the state transition probability and observation probability;

[0017] When the improvement amplitude of the path matching rate between the hidden state sequence generated by the model and the actual labeled work task sequence is less than the preset threshold for three consecutive iterations, terminate the training, and output the work task recognition model that fuses the video surveillance work data features and work task association rules.

[0018] In this solution, the work task recognition is performed on each video surveillance within a preset time period according to the work task recognition model, the power change data of the video surveillance within the preset time period is obtained, and the work quality under each work task of the video surveillance is evaluated to construct a work task-power-quality data matrix, specifically:

[0019] Obtain the real-time work data of each video surveillance within the preset time, import the real-time work data into the work task recognition model, and determine the work task change data of each video surveillance within the preset time period. The work task change data includes the work task name and the duration of each work task;

[0020] Obtain the power change data of each video surveillance within the preset time period and the target recognition situation data under different work tasks. The power change data includes charge and discharge data, and the target recognition situation data includes the target detection bounding box selection data or abnormal detection result data under the corresponding work task;

[0021] Determine the target recognition situation data of each video frame of the monitored video according to the target recognition situation data, and determine the movement trajectory of the target detection bounding box according to the target recognition situation data of each video frame;

[0022] For the target tracking work task, judge the accuracy of the target bounding box selection and the continuity of the tracking trajectory according to the movement trajectory, and determine the abnormal detection accuracy of the abnormal detection work task within the preset time period according to the abnormal detection result data;

[0023] Determine the work quality of each work task of the video surveillance according to the target bounding box selection accuracy, tracking trajectory continuity, and abnormal detection accuracy;

[0024] Perform time series alignment operations on the work task changes, power changes, and work quality within the preset time period to construct a work task-power-quality data matrix.

[0025] In this solution, clustering each video surveillance according to the work task - power - quality data matrix, determining the surveillance tasks of each surveillance cluster, and obtaining the surveillance task allocation scheme, specifically:

[0026] Determine the influence of different powers on the work quality of each work task according to the work task - power - quality data matrix to obtain influence data, where the influence data includes the mapping relationship between each power threshold interval and the corresponding work quality compliance rate;

[0027] Real - time monitor the current remaining power of each video surveillance, match the set of work tasks allowed to be executed in the corresponding power threshold interval in the influence data according to the current remaining power, and construct the candidate task pool for each video surveillance;

[0028] Obtain the environmental perception data of the target area at the current moment. When a moving target is detected, trigger the target tracking mode, calculate the azimuth angle between each video surveillance camera and the target movement trajectory, and determine the quality influence coefficient of the current remaining power on the target tracking task according to the influence data;

[0029] Generate a comprehensive scoring result of the azimuth matching degree and power guarantee degree according to the azimuth angle and the quality influence coefficient to obtain the applicability score of each video surveillance;

[0030] Remove the video surveillances with applicability scores lower than the preset threshold from the target tracking cluster, dynamically allocate the top N video surveillances to the target tracking cluster according to the score ranking, and assign the remaining surveillances to the anomaly detection cluster;

[0031] When the remaining power of any video surveillance drops to the first critical threshold, determine the work task type with the highest quality compliance rate at this remaining power according to the influence data, and re - allocate the video surveillance to the corresponding task cluster to obtain the surveillance task allocation scheme;

[0032] Continuously collect the power consumption rate and work quality compliance rate during the execution of the tasks of each cluster. When it is detected that the average quality compliance rate of a certain cluster is lower than the second critical threshold, recalculate the applicability scores of all surveillances in the area and adjust the surveillance task allocation scheme until the quality of the cluster tasks resumes within the safe threshold range.

[0033] In this solution, determining the influence of different powers on the work quality of each work task according to the work task - power - quality data matrix to obtain influence data, specifically:

[0034] Based on the work task-power-quality data matrix, the work quality data of each video surveillance executing each work task under different power states during the historical operation period are extracted, the power data of the continuous time series are divided into power change segments through a sliding time window, and the power values ​​in each segment are clustered and analyzed to divide multiple power threshold intervals representing the power level;

[0035] For each work task type, the proportion of samples with work quality that meet the standards in each power threshold interval to the total number of samples in the interval is counted to generate an initial mapping relationship between the power threshold interval and the quality compliance rate;

[0036] The probability density estimation method is used to fit the probability distribution of the quality compliance rate within each power threshold interval, and the significance difference degree of the work quality distribution between different power intervals is calculated according to the probability distribution. If the difference in the work quality distribution of adjacent power intervals does not reach the preset significance level, the power threshold intervals are merged;

[0037] Based on the merged power threshold intervals, the mean and variance of the work quality compliance rate in each interval are recalculated, the abnormal intervals whose variance exceeds the stability threshold are eliminated, and the power threshold intervals and their corresponding quality compliance rates that meet the confidence conditions are retained;

[0038] A dynamic matching relationship is established between the finally divided power threshold interval and the quality compliance rate, and a mapping table of the preset power threshold interval and the work quality compliance rate is generated as the influencing data.

[0039] In this solution, the video surveillance data of the target area is obtained according to the monitoring task allocation plan, abnormal events are identified according to the video surveillance data, and monitoring and early warning operations are performed according to the abnormal event identification results, specifically:

[0040] Based on the monitoring cluster task types divided in the monitoring task allocation scheme, the video monitoring devices in each cluster are dynamically scheduled to collect real-time video stream data of the target area, and the video stream data is input into a preset abnormal event recognition model, wherein the abnormal event recognition model is generated based on historical abnormal event sample training, and its input features include dynamic target features in the video frame, environmental change patterns, and spatiotemporal correlation of target motion trajectories;

[0041] The abnormal event recognition model is used to perform feature matching on the video stream data. When an abnormal pattern whose matching degree exceeds a threshold value in the preset abnormal feature library is detected to appear continuously in the video frame sequence, the time window, spatial distribution and feature intensity of the abnormal event are extracted to generate an abnormal event determination result;

[0042] According to the determination result, if the intensity of the abnormal event feature is lower than the first risk threshold, it is marked as a potential abnormal event, and the acquisition frame rate and target tracking parameters of the corresponding monitoring cluster are adjusted to continuously verify the abnormal feature;

[0043] If the feature intensity exceeds the first risk threshold, an alarm signal is generated according to the type of abnormal event. Meanwhile, the time window data, spatial coordinates, and enhanced video clips of the abnormal event are uploaded to the cloud platform, and the directional warning function of the acoustic and optical alarm device within the target area is activated;

[0044] During the duration of the alarm signal, the spatial diffusion trend and the change of feature intensity of the abnormal event are monitored in real time. If the diffusion trend exceeds the preset safety range or the feature intensity reaches the second risk threshold, the emergency resource scheduling path is determined according to the type of abnormal event and the location of the abnormal event.

[0045] The second aspect of the present invention also provides an environment-aware CT self-powered video monitoring and warning device, which includes: a memory and a processor. The memory includes an environment-aware CT self-powered video monitoring and warning method program. When the environment-aware CT self-powered video monitoring and warning method program is executed by the processor, the following steps are implemented:

[0046] Obtain the historical working data of each CT self-powered video monitor in the target area, analyze the historical working data based on the hidden Markov model, and construct a working task recognition model for the video monitor;

[0047] According to the working task recognition model, identify the working tasks of each video monitor within a preset time period, obtain the power change data of the video monitor within the preset time period, and evaluate the working quality of each video monitor under each working task to construct a working task-power-quality data matrix;

[0048] According to the working task-power-quality data matrix, divide each video monitor into clusters, determine the monitoring tasks of each monitoring cluster, and obtain a monitoring task allocation plan;

[0049] Obtain the video monitoring data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video monitoring data, and perform monitoring and warning operations according to the abnormal event identification results.

[0050] The present invention discloses a method and device for CT self-powered video monitoring and early warning based on environmental perception. The method obtains the historical working data of CT self-powered video monitoring in a target area, constructs a working task recognition model based on the hidden Markov model; uses the model to identify the working tasks within a preset time period, analyzes the power change and evaluates the working quality, constructs a working task-power-quality data matrix; performs cluster division based on the matrix to determine the monitoring task allocation scheme; obtains video monitoring data according to the scheme, identifies abnormal events, and conducts monitoring and early warning. The present invention optimizes the video monitoring task allocation, improves the monitoring efficiency and early warning accuracy, and is applicable to the fields of intelligent monitoring and power management. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The flowchart of a method for CT self-powered video monitoring and early warning based on environmental perception according to the present invention is shown;

[0052] Figure 2 The flowchart of obtaining influence data according to the present invention is shown;

[0053] Figure 3 The flowchart of performing monitoring and early warning operations according to the present invention is shown;

[0054] Figure 4 The block diagram of a device for CT self-powered video monitoring and early warning based on environmental perception according to the present invention is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0056] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0057] Figure 1 The flowchart of a method for CT self-powered video monitoring and early warning based on environmental perception according to the present invention is shown.

[0058] As Figure 1 shown, the first aspect of the present invention provides a method for CT self-powered video monitoring and early warning based on environmental perception, including:

[0059] S102, obtaining the historical working data of each CT self-powered video monitoring in the target area, analyzing the historical working data based on the hidden Markov model, and constructing a working task recognition model for video monitoring;

[0060] S104. Identify the work tasks of each video surveillance within a preset time period according to the work task recognition model, obtain the power change data of the video surveillance within the preset time period, evaluate the work quality under each work task of the video surveillance, and construct a work task - power - quality data matrix;

[0061] S106. Perform cluster partitioning on each video surveillance according to the work task - power - quality data matrix, determine the monitoring tasks of each monitoring cluster, and obtain a monitoring task allocation plan;

[0062] S108. Obtain the video surveillance data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video surveillance data, and perform monitoring and early warning operations according to the results of the abnormal event identification.

[0063] It should be noted that by constructing a work task recognition model based on the hidden Markov model, the problem of inaccurate task type recognition in traditional video surveillance systems caused by dynamic environmental changes is effectively solved. It can mine the association rules between device operating states and task types from complex historical work data, and improve the intelligent level of monitoring task classification; by establishing a work task - power - quality data matrix, the limitation of the separate analysis of power consumption and task execution quality in the existing technology is broken through, and the quantitative evaluation of the impact of dynamic power changes on task execution efficiency is realized, providing data support for subsequent resource optimization allocation; the matrix - based cluster partitioning method overcomes the defect of uneven device load under the fixed task allocation mode. By dynamically matching the relationship between the remaining power of the device and the task quality requirements, an adaptive monitoring task allocation plan is formed, significantly improving the collaborative work efficiency of the device cluster in low - power scenarios; finally, combined with the abnormal recognition and hierarchical early warning mechanism of environmental perception, the problems of missed detection and response lag of abnormal events in traditional monitoring systems caused by the limited perspective of a single device are solved. Through the collaborative data fusion analysis of the cluster, the accurate positioning of abnormal events and the dynamic evaluation of the risk level are realized, so as to effectively improve the real - time performance of monitoring and early warning and the accuracy of emergency disposal in complex environments while ensuring the battery life of the device.

[0064] According to an embodiment of the present invention, the obtaining of the historical work data of each CT self - powered video surveillance in the target area, and the analysis of the historical work data based on the hidden Markov model to construct a work task recognition model of the video surveillance is specifically as follows:

[0065] Obtain the historical work data of each CT self - powered video surveillance in the target area. The historical work data includes the working current and voltage data, processor load status, camera movement trajectory parameters during the operation of the video surveillance, and the work task labels marked at the corresponding moments. The work task labels include target tracking and abnormal detection;

[0066] Preprocess the historical work data by aligning timestamps and imputing missing values. After dividing it into continuous time slot segments using a sliding window, normalize the working current and voltage data, the processor load status, and the camera movement trajectory parameters to generate an observation sequence of the input data type for the hidden Markov model;

[0067] Define the set of hidden states of the hidden Markov model as various types of work tasks labeled in the work task tags, and import the observation sequence into the hidden Markov model as the observation states and the set of hidden states of the hidden Markov model;

[0068] Calculate the initial state transition probability matrix and the observation probability matrix of the hidden Markov model through the forward-backward algorithm. Use the labeled work task tags to perform supervised constraints on the true hidden states of each time slot segment, and use the Baum-Welch algorithm with label correction to iteratively optimize the distribution parameters of the state transition probability and the observation probability;

[0069] Terminate the training when the improvement amplitude of the path matching rate between the hidden state sequence generated by the model and the actual labeled work task sequence is less than the preset threshold for three consecutive iterations, and output a work task recognition model that fuses the work data features of video surveillance and the association rules of work tasks.

[0070] It should be noted that by leveraging the modeling ability of the Hidden Markov Model for time-series data, the dynamic association relationship between the multi-dimensional operating parameters of video surveillance devices (such as current, voltage, processor load, camera movement trajectory) and work task tags is probabilistically characterized, thereby achieving precise identification of device work tasks. Specifically, by defining the work task type as the hidden state and the device operating parameters as the observed state, the Hidden Markov Model can learn the typical feature distributions of device operating parameters under different work tasks (such as high processor load and frequent camera movement corresponding to the target tracking task), and capture the state transition rules during work task switching (such as the transition probability of triggering target tracking after the anomaly detection task ends). During the model training process, through a supervised correction mechanism with labeled tags, combined with the joint optimization of the forward-backward algorithm and the Baum-Welch algorithm, the problem of misjudgment caused by the ambiguous association between observed data and hidden states in the traditional unsupervised Hidden Markov Model for task identification is effectively solved, significantly improving the path matching accuracy of the model for actual work task sequences. The resulting work task identification model can not only real-time analyze the implicit task types in device operation data, but also adapt to parameter fluctuations caused by hardware differences or environmental interference in different devices, ultimately achieving dynamic perception and high-robustness classification of work task types in complex scenarios, providing reliable task status inputs for subsequent cluster collaborative scheduling based on power consumption and task quality. Supervised constraint refers to the fact that during the training process of the Hidden Markov Model, the authenticity of the hidden state is forced to be guided by using the labeled work task tags. By taking the known work task tags of time slot segments as the true values of the hidden state, the parameter update direction in the calculation of state transition probability and observation probability of the model is directly corrected, thereby breaking through the limitation of the traditional unsupervised Baum-Welch algorithm relying on pure statistical learning; while the Baum-Welch algorithm with label correction is an improvement of the classical algorithm based on supervised constraint. During the iterative process of the Expectation-Maximization (EM) algorithm, when calculating the state sequence probability through the forward-backward algorithm, the hidden state path corresponding to the labeled tag is injected into the parameter estimation link as prior knowledge, forcing the adjustment of the probability weights of the wrong transition paths in the state transition matrix. At the same time, when optimizing the observation probability matrix, the observed value distribution of device operating parameters (such as current fluctuations, camera movement trajectory) is directionally corrected according to the true work task type corresponding to the label, so that the model parameter update not only depends on the statistical characteristics of the observed data, but also deeply integrates the deterministic rules of the known task tags.This hybrid training mechanism that combines supervision and unsupervised learning can significantly enhance the model's ability to capture the implicit correlation between device work tasks and operating parameters, effectively solve the problems of large differences in parameter distributions for the same type of task and strong randomness in task switching timings caused by device heterogeneity in traditional methods, and ultimately achieve high-precision convergence and strong generalization ability of the work task recognition model; the distribution parameters of the state transition probability include the state transition probability matrix elements between various work task types, and the distribution parameters of the observation probability include the normalized quantization level probabilities of sensor data, processor load level probabilities, and discrete angular velocity interval probabilities of camera movement trajectories.

[0071] According to an embodiment of the present invention, the work task recognition model is used to recognize the work tasks of each video surveillance within a preset time period, obtain the power change data of the video surveillance within the preset time period, and evaluate the work quality under each work task of the video surveillance to construct a work task - power - quality data matrix, specifically as follows:

[0072] Obtain the real-time work data of each video surveillance within the preset time, import the real-time work data into the work task recognition model, and determine the work task change data of each video surveillance within the preset time period. The work task change data includes the work task name and the duration of each work task.

[0073] Obtain the power change data of each video surveillance within the preset time period and the target recognition situation data under different work tasks. The power change data includes charge and discharge data, and the target recognition situation data includes target detection bounding box selection data or anomaly detection result data corresponding to the work task.

[0074] Determine the target recognition situation data for each video frame of the monitored video according to the target recognition situation data, and determine the movement trajectory of the target detection bounding box according to the target recognition situation data for each video frame.

[0075] For the target tracking work task, judge the accuracy of target bounding box selection and the continuity of the tracking trajectory according to the movement trajectory, and determine the anomaly detection accuracy of the anomaly detection work task within the preset time period according to the anomaly detection result data.

[0076] Determine the work quality of each work task of the video surveillance according to the accuracy of target bounding box selection, the continuity of the tracking trajectory, and the anomaly detection accuracy.

[0077] Perform a time series alignment operation on the work task changes, power change data, and work quality within the preset time period to construct a work task - power - quality data matrix.

[0078] It should be noted that in the target tracking task, by analyzing the moving trajectory of the target detection box between consecutive video frames, it is possible to effectively identify the tracking jitter or target loss phenomenon caused by insufficient device power, thereby accurately evaluating the trajectory continuity quality of the tracking task; in the anomaly detection task, by comparing and verifying the anomaly detection results with the real scenario, quality indicators such as the false alarm rate and missed detection rate of the device in a specific power range can be quantitatively statistically analyzed; by aligning the task type, duration, power change curve, and quality assessment results in the time dimension, the constructed three-dimensional data matrix can intuitively reflect the quality decay law of various types of tasks under different power thresholds. For example, it can identify the characteristic that the probability of trajectory breakage increases steeply when the target tracking task is executed with the remaining power below a certain critical value, or discover the operating characteristic that the anomaly detection task maintains a high accuracy rate in the medium power range. Furthermore, it provides a data-driven decision-making basis for subsequent dynamic task allocation, effectively avoiding the problem of system reliability decline caused by the traditional system's lack of task-power-quality correlation analysis, which leads to the device forcibly executing high-precision tasks in a low-power state. At the same time, through the prediction of the quality decay law, preventive adjustment of the device working mode is realized, significantly improving the task execution success rate and anomaly event handling timeliness of the self-powered monitoring system under limited energy conditions.

[0079] According to an embodiment of the present invention, clustering each video monitoring according to the work task-power-quality data matrix, determining the monitoring tasks of each monitoring cluster, and obtaining a monitoring task allocation scheme, specifically:

[0080] Determine the influence of different powers on the work quality of each work task according to the work task-power-quality data matrix to obtain influence data, where the influence data includes the mapping relationship between each power threshold interval and the corresponding work quality compliance rate;

[0081] Real-time monitor the current remaining power of each video monitoring, match the set of work tasks allowed to be executed in the corresponding power threshold interval in the influence data according to the current remaining power, and construct a candidate task pool for each video monitoring;

[0082] Obtain the environmental perception data of the target area at the current moment, trigger the target tracking mode when a moving target is detected, calculate the azimuth angle between each video monitoring camera and the target movement trajectory, and determine the quality influence coefficient of the current remaining power on the target tracking task according to the influence data;

[0083] Generate a comprehensive scoring result of the azimuth matching degree and power guarantee degree according to the azimuth angle and the quality influence coefficient, and obtain the applicability score of each video monitoring;

[0084] Remove video monitoring with an applicability score lower than the preset threshold from the target tracking cluster, dynamically allocate the top N video monitoring to the target tracking cluster according to the score ranking, and assign the remaining monitoring to the anomaly detection cluster;

[0085] When the remaining power of any video monitoring drops to the first critical threshold, determine the type of work task with the highest quality compliance rate at the remaining power according to the impact data, and reassign the video monitoring to the corresponding task cluster to obtain a monitoring task allocation plan;

[0086] Continuously collect the power consumption rate and work quality compliance rate during the execution of tasks in each cluster. When it is detected that the average quality compliance rate of a certain cluster is lower than the second critical threshold, recalculate the applicability scores of all monitors in the area and adjust the monitoring task allocation plan until the quality of the cluster tasks returns to the safe threshold range.

[0087] It should be noted that due to the fixed task allocation mode, the task execution quality is often unstable due to fluctuations in device power and dynamic environmental changes. For example, when a low-power device continuously executes a high-energy-consuming tracking task, problems such as target loss and false negatives in anomaly detection are likely to occur. When an unexpected environmental anomaly occurs, the device tasks are fixed and it is impossible to quickly form an optimal monitoring cluster, resulting in waste of monitoring resources and low response efficiency. Therefore, by establishing a power-task quality correlation model and a dynamic cluster division mechanism, efficient cooperation of monitoring devices in resource-constrained scenarios is achieved: First, based on historical data mining, the constraint laws of different power ranges on various task qualities are explored, and dynamic matching rules between power thresholds and task qualities are constructed to provide a candidate task pool adapted to the power for the device; Secondly, through the dual evaluation of real-time environmental perception data (such as the orientation of moving targets) and the remaining power of the device, through the fusion scoring of spatial coverage (azimuth angle) and power support (quality impact coefficient), the device cluster that can ensure task quality and has the optimal spatial layout in the current environment is dynamically selected; When the device power drops to the critical threshold, it is automatically switched to the task type with the highest quality compliance rate at the current power to avoid systematic risks caused by low-power devices struggling to support high-precision tasks; At the same time, through the closed-loop monitoring and dynamic reallocation of the task execution quality of the cluster, the device load balance is continuously optimized, effectively solving the problem of the overall efficiency decline of the cluster caused by environmental mutations or device performance attenuation in the traditional system. Finally, the success rate of task execution and the response speed of unexpected events of the monitoring system are synchronously improved under limited energy supply, extending the device battery life while ensuring the stability of monitoring and early warning. Environmental perception data refers to the characteristic information reflecting the dynamic changes of the monitoring scene obtained in real time through video monitoring devices and associated sensors, including multi-dimensional parameters related to the behavior patterns of moving targets and environmental interference, such as the movement trajectory, speed change, direction deviation, environmental occlusion degree, and light condition fluctuation of moving targets. When determining the quality impact coefficient of the current remaining power on the target tracking task, first, based on the pre-established mapping relationship between the power threshold interval and the quality compliance rate, the interval range to which the current remaining power belongs is located, and the historical average quality compliance rate of the target tracking task in this interval is obtained as the reference value; Subsequently, combined with the dynamic interference factors in the real-time environmental perception data (such as the intensity of target movement and the frequency of occluders), through the preset environmental complexity evaluation model, the additional loss weight of environmental interference on the task execution quality is quantified. For example, high-speed movement or frequent occlusion will trigger a quality attenuation factor; Finally, the reference quality compliance rate and the attenuation factor caused by environmental interference are dynamically coupled to generate a composite quality impact coefficient that comprehensively reflects the current power support ability and environmental challenge degree.

[0088] Figure 2 The flowchart of the present invention for obtaining influence data is shown.

[0089] According to an embodiment of the present invention, determining the influence of different power levels on the working quality of each work task based on the work task - power - quality data matrix to obtain influence data, specifically:

[0090] S202, extracting the working quality data of each video surveillance for each work task in different power states during historical operation based on the work task - power - quality data matrix, dividing the power data of the continuous time series into power change segments through a sliding time window, and performing cluster analysis on the power values within each segment to divide multiple power threshold intervals representing power levels;

[0091] S204, for each work task type, counting the proportion of the number of samples with qualified working quality in each power threshold interval to the total sample size of the interval, and generating an initial mapping relationship between the power threshold interval and the quality pass rate;

[0092] S206, using the probability density estimation method to fit the probability distribution of the quality pass rate within each power threshold interval, calculating the significant difference degree of the working quality distribution between different power intervals according to the probability distribution. If the difference in the working quality distribution between adjacent power intervals does not reach the preset significance level, then merge the power threshold intervals;

[0093] S208, recalculating the mean and variance of the working quality pass rate within each interval based on the merged power threshold intervals, removing the abnormal intervals with variance exceeding the stability threshold, and retaining the power threshold intervals that meet the confidence condition and their corresponding quality pass rates;

[0094] S210, establishing a dynamic matching relationship between the finally divided power threshold intervals and the quality pass rate, and generating a mapping table of the preset power threshold intervals and the working quality pass rate as the influence data.

[0095] It should be noted that, based on the sliding time window and clustering analysis techniques, multiple threshold intervals that can characterize the typical power states of devices are identified from continuously changing power data, overcoming the possible empirical biases in manually setting fixed power segments; subsequently, by statistically analyzing the distribution patterns of the task quality compliance rates within each power interval, a preliminary correlation between the power level and task quality is established; furthermore, probability density estimation and significance difference testing are used to scientifically verify and merge and optimize the initially divided intervals, eliminating statistical noise caused by data sampling randomness or overly fine intervals, ensuring that there are statistically significant differences and distribution consistencies in the task quality compliance rates within each remaining power threshold interval; finally, through analysis of variance and confidence level screening, intervals with excessive fluctuations caused by device individual differences or abnormal operating conditions are excluded, and power intervals with stable quality characterization capabilities are retained, thereby generating a dynamically matched power threshold - quality compliance rate mapping table. Through multi-level data cleaning and statistical verification, this process not only avoids the over-simplification of the complex power - quality non-linear relationship in traditional methods by simple threshold segmentation, but also ensures the robustness of the influencing data under device heterogeneity and environmental fluctuations, ultimately forming influencing data that can accurately reflect the variation laws of task execution efficiency under different power states, providing a reliable power constraint model for dynamic task allocation. The calculation of the confidence level is achieved through a method combining probability density estimation and hypothesis testing: First, based on the historical quality compliance rate samples within each power threshold interval, a probability distribution model of the quality compliance rate is constructed using parametric estimation (such as the mean-variance model under normal distribution) or non-parametric kernel density estimation methods; subsequently, by calculating the confidence interval range corresponding to a preset confidence level (such as 95%) under this distribution, the probability that the true value of the quality compliance rate falls within this interval is determined, and its calculation depends on the sample mean, standard deviation, and sample size (such as the critical value of the t-distribution or Z-score), ultimately forming a confidence level indicator that characterizes the reliability of the interval estimation. The confidence level condition is that the width of the confidence interval is less than the preset value.

[0096] Figure 3 Fig. shows the flowchart of the monitoring and early warning operation of the present invention.

[0097] According to an embodiment of the present invention, obtaining video monitoring data of a target area according to the monitoring task allocation scheme, identifying abnormal events based on the video monitoring data, and performing a monitoring and early warning operation according to the abnormal event identification result specifically include:

[0098] S302, based on the monitoring cluster task types divided in the monitoring task allocation scheme, dynamically scheduling the video monitoring devices within each cluster to collect real-time video stream data of the target area, and inputting the video stream data into a preset abnormal event identification model, where the abnormal event identification model is trained based on historical abnormal event samples, and its input features include dynamic target features within video frames, environmental change patterns, and spatio-temporal correlations of target movement trajectories;

[0099] S304, perform feature matching on the video stream data through the abnormal event recognition model. When it is detected that abnormal patterns with a matching degree exceeding the threshold continuously appear in the video frame sequence in the preset abnormal feature library, extract the time window, spatial distribution, and feature intensity of the abnormal event to generate an abnormal event determination result;

[0100] S306, according to the determination result, if the feature intensity of the abnormal event is lower than the first risk threshold, mark it as a potential abnormal event and adjust the acquisition frame rate and target tracking parameters of the corresponding monitoring cluster to continuously verify the abnormal features;

[0101] S308, if the feature intensity exceeds the first risk threshold, generate an alarm signal according to the abnormal event type, synchronously upload the time window data, spatial coordinates, and enhanced video segment of the abnormal event to the cloud platform, and activate the directional warning function of the acoustic and optical alarm device in the target area;

[0102] S310, during the duration of the alarm signal, continuously monitor the spatial diffusion trend and the change of feature intensity of the abnormal event. If the diffusion trend exceeds the preset safety range or the feature intensity reaches the second risk threshold, determine the emergency resource scheduling path according to the abnormal event type and the location of the abnormal event.

[0103] It should be noted that the multi-dimensional feature analysis of video stream data (dynamic targets, environmental changes, and temporal and spatial correlations) based on the pre-trained anomaly recognition model can effectively capture hidden abnormal patterns that are difficult to identify with traditional single-frame detection methods (such as slowly spreading leakage events or intermittent suspicious behaviors), and reduce the misjudgment rate by matching abnormal features of continuous frame sequences; when potential anomalies are detected, the acquisition frame rate and tracking parameters of the monitoring cluster are adaptively adjusted to achieve continuous high-precision tracking of suspicious targets, avoiding the problem of feature loss caused by insufficient resolution in the traditional fixed parameter mode; differentiated early warning strategies are initiated for abnormal events of different risk levels, and low-risk Events trigger the optimization and reorganization of monitoring resources to verify features, and high-risk events link the cloud and sound and light alarm devices to achieve multi-modal warnings, ensuring real-time synchronization and rapid response of key information; in addition, by real-time monitoring of the spatial diffusion trend and intensity changes of abnormal events, dynamic generation of resource scheduling paths can accurately match the needs of event evolution (such as giving priority to dispatching surrounding fire-fighting resources when a fire spreads), overcoming the response lag and resource mismatch problems of traditional plan-based scheduling, thereby reducing the system's false alarm rate while comprehensively improving the closed-loop disposal efficiency from abnormality perception to resource scheduling, and achieving coordinated optimization of the monitoring and early warning system in event identification accuracy, response real-time, and resource utilization efficiency. Abnormal modes include abnormal behavioral characteristics of dynamic targets in video scenes (such as sudden acceleration, reverse movement or group gathering), abnormal changes in environmental parameters (such as abnormal temperature gradient, regional color difference changes caused by smoke diffusion or liquid leakage), and abnormal spatiotemporal correlation of target motion trajectories (such as wandering paths exceeding preset safety routes, abnormal stays in specific areas or high-frequency returns), and also cover abnormal correlations between interactive behaviors of multiple targets (such as illegal transfer of objects, abnormal shortening of following distance); target tracking parameters include camera gimbal steering speed, optical zoom ratio, video acquisition frame rate, trajectory prediction model parameters in the tracking algorithm (such as Kalman filter coefficients, motion speed weights), target feature matching thresholds (such as color histogram similarity, contour recognition accuracy) and multi-target tracking priority rules (such as priority locking of near-field targets and continuous focus on high-threat targets).

[0104] Figure 4 The block diagram of the CT self-collecting electric video monitoring and early warning device based on environment perception of the present invention is shown.

[0105] The second aspect of the present invention further provides a CT self-service electricity video monitoring and early warning device 4 based on environmental perception, the system comprising: a memory 41, a processor 42, the memory comprising a CT self-service electricity video monitoring and early warning method program based on environmental perception, the CT self-service electricity video monitoring and early warning method program based on environmental perception when executed by the processor, implements the following steps:

[0106] Obtain the historical working data of each CT self-powered video surveillance in the target area, analyze the historical working data based on the hidden Markov model, and construct a working task recognition model for video surveillance;

[0107] According to the working task recognition model, identify the working tasks of each video surveillance within a preset time period, obtain the power change data of the video surveillance within the preset time period, and evaluate the working quality under each working task of the video surveillance to construct a working task-power-quality data matrix;

[0108] According to the working task-power-quality data matrix, perform cluster division on each video surveillance, determine the monitoring tasks of each monitoring cluster, and obtain a monitoring task allocation plan;

[0109] Obtain the video surveillance data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video surveillance data, and perform monitoring and early warning operations according to the abnormal event identification results.

[0110] The present invention discloses a CT self-powered video monitoring and early warning method and device based on environmental perception. The method obtains the historical working data of CT self-powered video surveillance in the target area, constructs a working task recognition model based on the hidden Markov model; uses the model to identify the working tasks within a preset time period, analyzes the power change and evaluates the working quality, constructs a working task-power-quality data matrix; performs cluster division based on the matrix, determines the monitoring task allocation plan; obtains the video surveillance data according to the plan, identifies abnormal events, and performs monitoring and early warning. The present invention optimizes the video surveillance task allocation, improves the monitoring efficiency and early warning accuracy, and is applicable to the fields of intelligent monitoring and power management.

[0111] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0112] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0113] In addition, each functional unit in the embodiments of the present invention may all be integrated into one processing unit, or each unit may be separately regarded as one unit, or two or more units may be integrated into one unit; the above-mentioned integrated unit may be implemented in the form of hardware, or in the form of a hardware plus software functional unit.

[0114] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks or optical discs and other various media that can store program codes.

[0115] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks or optical discs and other various media that can store program codes.

[0116] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all of them should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A CT self-collected video monitoring and early warning method based on environmental perception, characterized in that: The following steps are involved: Obtaining historical working data of each CT self-collecting video surveillance in the target area, analyzing the historical working data based on a hidden Markov model, and constructing a working task recognition model for video surveillance; According to the work task identification model, work tasks are identified for each video surveillance within a preset time period, power change data of the video surveillance within the preset time period is obtained, and the work quality under each work task of the video surveillance is evaluated to construct a work task-power-quality data matrix; Clustering each video monitoring according to the work task-power-quality data matrix, determining the monitoring task of each monitoring cluster, and obtaining a monitoring task allocation plan; Acquire video surveillance data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video surveillance data, and perform monitoring and early warning operations according to the abnormal event identification results; The work task identification model is used to identify each video surveillance within a preset time period, obtain the power change data of the video surveillance within the preset time period, and evaluate the work quality under each work task of the video surveillance to construct a work task-power-quality data matrix, specifically: Acquire the real-time working data of each video surveillance within a preset time, import the real-time working data into the work task identification model, and determine the work task change data of each video surveillance within the preset time period, wherein the work task change data includes the work task name and the duration of each work task; Obtaining power change data of each video surveillance within a preset time period and target recognition data under different work tasks, wherein the power change data includes charge and discharge data, and the target recognition data includes target detection box selection data or abnormal detection result data under the corresponding work task; Determine the target recognition data for each video frame of the monitored video within a preset time period according to the target recognition data, and determine the moving trajectory of the target detection frame according to the target recognition data for each video frame; For the target tracking task, the target selection accuracy and tracking trajectory continuity are determined according to the moving trajectory, and the anomaly detection accuracy of the anomaly detection task within a preset time period is determined according to the anomaly detection result data; Determine the work quality of each video monitoring task based on the target selection accuracy, tracking trajectory continuity, and anomaly detection accuracy; The work task changes, power change data, and work quality within the preset time period are aligned in time series to construct a work task-power-quality data matrix.

2. The method for monitoring and early warning of CT self-collected electric video based on environment perception according to claim 1 is characterized in that: The method of obtaining the historical working data of each CT self-collecting video surveillance in the target area, analyzing the historical working data based on the hidden Markov model, and constructing a working task recognition model for video surveillance is specifically as follows: Obtain the historical working data of each CT self-collecting video surveillance in the target area, wherein the historical working data includes the working current and voltage data, the processor load status, the camera motion trajectory parameters and the work task labels marked at the corresponding time during the video surveillance operation, wherein the work task labels include target tracking and anomaly detection; Performing timestamp alignment and missing value interpolation preprocessing on the historical working data, dividing it into continuous time slot segments using a sliding window, normalizing the working current and voltage data, processor load status, and camera motion trajectory parameters, and generating an observation sequence of the hidden Markov model input data type; Define the hidden state set of the hidden Markov model as the various types of work tasks marked in the work task label, and import the observation sequence into the hidden Markov model as the observation state and hidden state set of the hidden Markov model; The initial state transition probability matrix and observation probability matrix of the hidden Markov model are calculated by the forward-backward algorithm, the true hidden state of each time slot segment is supervised and constrained by the labeled task labels, and the distribution parameters of the state transition probability and observation probability are iteratively optimized by the label-corrected Baum-Welch algorithm. When the path matching rate between the hidden state sequence generated by the model and the actual labeled work task sequence increases less than the preset threshold for three consecutive iterations, the training is terminated and a work task recognition model that integrates the video surveillance work data features and work task association rules is output.

3. The method for monitoring and early warning of CT self-collected electric video based on environment perception according to claim 1 is characterized in that: The method of clustering each video monitoring according to the work task-power-quality data matrix, determining the monitoring task of each monitoring cluster, and obtaining a monitoring task allocation scheme is as follows: Determine the influence of different power on the work quality of each work task according to the work task-power-quality data matrix, and obtain influence data, wherein the influence data includes a mapping relationship between each power threshold interval and the corresponding work quality compliance rate; Monitor the current remaining power of each video surveillance in real time, match the set of work tasks that are allowed to be executed in the corresponding power threshold interval in the impact data according to the current remaining power, and build a candidate task pool for each video surveillance; Acquire the environmental perception data of the target area at the current moment, trigger the target tracking mode when a moving target is detected, calculate the azimuth angle between each video surveillance camera and the target motion trajectory, and determine the quality influence coefficient of the current remaining power on the target tracking task according to the influence data; Generate a comprehensive scoring result of the orientation matching degree and the power guarantee degree according to the orientation angle and the quality influence coefficient, and obtain the applicability score of each video monitoring; Video surveillance with a fitness score lower than the preset threshold is removed from the target tracking cluster, and the top N video surveillances are dynamically assigned to the target tracking cluster according to the score ranking, and the remaining surveillances are assigned to the anomaly detection cluster; When the remaining power of any video monitoring drops to a first critical threshold, the work task type with the highest quality compliance rate under the remaining power is determined according to the impact data, and the video monitoring is reallocated to the corresponding task cluster to obtain a monitoring task allocation plan; The power consumption rate and work quality compliance rate of each cluster task are continuously collected during execution. When it is detected that the average quality compliance rate of a cluster is lower than the second critical threshold, the suitability scores of all monitoring in the area are recalculated and the monitoring task allocation plan is adjusted until the cluster task quality is restored to within the safety threshold range.

4. The method for monitoring and early warning of CT self-collected electric and video signals based on environment perception according to claim 3 is characterized in that: The influence of different power on the work quality of each work task is determined according to the work task-power-quality data matrix to obtain the influence data, specifically: Based on the work task-power-quality data matrix, the work quality data of each video surveillance executing each work task under different power states during the historical operation period are extracted, the power data of the continuous time series are divided into power change segments through a sliding time window, and the power values ​​in each segment are clustered and analyzed to divide multiple power threshold intervals representing the power level; For each work task type, the proportion of samples with work quality that meet the standards in each power threshold interval to the total number of samples in the interval is counted to generate an initial mapping relationship between the power threshold interval and the quality compliance rate; The probability density estimation method is used to fit the probability distribution of the quality compliance rate within each power threshold interval, and the significance difference degree of the work quality distribution between different power intervals is calculated according to the probability distribution. If the difference in the work quality distribution of adjacent power intervals does not reach the preset significance level, the power threshold intervals are merged; Based on the merged power threshold intervals, the mean and variance of the work quality compliance rate in each interval are recalculated, the abnormal intervals whose variance exceeds the stability threshold are eliminated, and the power threshold intervals and their corresponding quality compliance rates that meet the confidence conditions are retained; A dynamic matching relationship is established between the finally divided power threshold interval and the quality compliance rate, and a mapping table of the preset power threshold interval and the work quality compliance rate is generated as the influencing data.

5. The method for monitoring and early warning of CT self-collected electric video based on environment perception according to claim 1 is characterized in that: The acquiring of video surveillance data of the target area according to the monitoring task allocation scheme, identifying abnormal events according to the video surveillance data, and performing monitoring and early warning operations according to the abnormal event identification results are specifically as follows: Based on the monitoring cluster task types divided in the monitoring task allocation scheme, the video monitoring devices in each cluster are dynamically scheduled to collect real-time video stream data of the target area, and the video stream data is input into a preset abnormal event recognition model, wherein the abnormal event recognition model is generated based on historical abnormal event sample training, and its input features include dynamic target features in the video frame, environmental change patterns, and spatiotemporal correlation of target motion trajectories; The abnormal event recognition model is used to perform feature matching on the video stream data. When an abnormal pattern whose matching degree exceeds a threshold value in the preset abnormal feature library is detected to appear continuously in the video frame sequence, the time window, spatial distribution and feature intensity of the abnormal event are extracted to generate an abnormal event determination result; According to the determination result, if the abnormal event feature intensity is lower than the first risk threshold, it is marked as a potential abnormal event and the acquisition frame rate and target tracking parameters of the corresponding monitoring cluster are adjusted to continuously verify the abnormal features; If the feature intensity exceeds the first risk threshold, an alarm signal is generated according to the type of abnormal event, the time window data, spatial coordinates and enhanced video clips of the abnormal event are simultaneously uploaded to the cloud platform, and the directional warning function of the sound and light alarm device in the target area is activated; During the duration of the alarm signal, the spatial diffusion trend and characteristic intensity changes of abnormal events are monitored in real time. If the diffusion trend exceeds the preset safety range or the characteristic intensity reaches the second risk threshold, the emergency resource scheduling path is determined based on the type and location of the abnormal event.

6. A CT self-collecting video monitoring and early warning device based on environmental perception, characterized in that: The CT self-service electricity video monitoring and early warning device based on environment perception includes a storage device and a processor. The storage device includes a CT self-service electricity video monitoring and early warning method program based on environment perception. When the CT self-service electricity video monitoring and early warning method program based on environment perception is executed by the processor, the following steps are implemented: Obtaining historical working data of each CT self-collecting video surveillance in the target area, analyzing the historical working data based on a hidden Markov model, and constructing a working task recognition model for video surveillance; According to the work task identification model, work tasks are identified for each video surveillance within a preset time period, power change data of the video surveillance within the preset time period is obtained, and the work quality under each work task of the video surveillance is evaluated to construct a work task-power-quality data matrix; Clustering each video monitoring according to the work task-power-quality data matrix, determining the monitoring task of each monitoring cluster, and obtaining a monitoring task allocation plan; Acquire video surveillance data of the target area according to the monitoring task allocation plan, identify abnormal events according to the video surveillance data, and perform monitoring and early warning operations according to the abnormal event identification results; The work task identification model is used to identify each video surveillance within a preset time period, obtain the power change data of the video surveillance within the preset time period, and evaluate the work quality under each work task of the video surveillance to construct a work task-power-quality data matrix, specifically: Acquire the real-time working data of each video surveillance within a preset time, import the real-time working data into the work task identification model, and determine the work task change data of each video surveillance within the preset time period, wherein the work task change data includes the work task name and the duration of each work task; Obtaining power change data of each video surveillance within a preset time period and target recognition data under different work tasks, wherein the power change data includes charge and discharge data, and the target recognition data includes target detection box selection data or abnormal detection result data under the corresponding work task; Determine the target recognition data for each video frame of the monitored video within a preset time period according to the target recognition data, and determine the moving trajectory of the target detection frame according to the target recognition data for each video frame; For the target tracking task, the target selection accuracy and tracking trajectory continuity are determined according to the moving trajectory, and the anomaly detection accuracy of the anomaly detection task within a preset time period is determined according to the anomaly detection result data; Determine the work quality of each video monitoring task based on the target selection accuracy, tracking trajectory continuity, and anomaly detection accuracy; The work task changes, power change data, and work quality within the preset time period are aligned in time series to construct a work task-power-quality data matrix.

Images