Safety helmet wearing detection system and detection method for electric power construction

Abstract

The invention discloses a safety helmet wearing detection system and detection method for power construction, and relates to the technical field of power construction, and the system comprises a data collection module, a hidden danger analysis module, a comparison module, a comprehensive analysis module and an early warning module. And the data acquisition module is used for acquiring various parameter information generated when the image data is analyzed based on the image recognition technology, and the parameter information comprises system response performance information and algorithm stability information. According to the invention, various parameter information based on image recognition technology analysis is collected through the data acquisition module, wherein the parameter information comprises system response performance information and algorithm stability information. The abnormal perception delay index evaluates abnormal detection delay, and the identification confidence fluctuation index evaluates identification stability. By comparing the data evaluation index with the preset threshold value, the system early warns a manager in real time, ensures that the safety helmet is worn accurately, reduces the construction safety risk, and avoids the loosening of the safety supervision of the manager.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power construction, and in particular to a safety helmet wearing detection system and a detection method for electric power construction. Background Art

[0002] Safety helmet wearing detection for power construction refers to the use of technical means to monitor and judge in real time whether workers are wearing safety helmets correctly at the power construction site to ensure their personal safety. Safety helmets are important personal protective equipment to protect workers' heads from falling objects, collisions and other accidents. Especially in the power construction environment, wearing safety helmets is particularly important due to high-altitude operations, complex equipment and densely distributed lines. Technical methods for safety helmet wearing detection usually include computer vision-based image recognition technology, sensor detection, smart helmets, etc. These technologies are used to automatically identify the status of workers on site and provide timely feedback to ensure that workers comply with safety regulations and reduce the risk of work-related accidents.

[0003] At present, the technical methods for helmet wearing detection are mainly implemented through three means: image recognition technology based on computer vision, sensor detection technology and smart helmets.

[0004] Image recognition technology based on computer vision: Video or image data of the construction site is collected in real time through high-definition cameras, and the images are analyzed using deep learning algorithms. The image recognition algorithm can automatically detect the position of the worker's head based on the pre-trained model, and identify whether the worker is wearing a helmet and whether the wearing method is correct. This technology can monitor multiple workers at the same time by processing images under different angles, lighting and scenes. When it is detected that the helmet is not worn or worn incorrectly, the system can automatically trigger an alarm to notify the on-site management personnel. The advantage of image recognition is that it can cover a wide monitoring area and does not rely on the active cooperation of workers, which is suitable for the management of large-scale construction scenes. Sensor detection technology: By embedding various sensors such as pressure sensors and gravity sensors inside the helmet or in peripheral equipment, it can be detected in real time whether the worker is wearing a helmet. For example, the pressure sensor can sense the contact between the worker's head and the lining of the helmet, while the gravity sensor can monitor whether the helmet is in a vertical state or has abnormal deviation. When the sensor detects that the worker is not wearing or wearing it improperly, it can transmit the information to the monitoring system through the wireless communication module for real-time feedback. The advantage of sensor detection is high accuracy, which is particularly suitable for use in areas with complex environments. Smart helmet technology: Smart helmets integrate multiple functions such as sensors, communication modules, cameras, etc., which can monitor the wearing status in real time and transmit data to the background system. Smart helmets are usually equipped with additional safety functions, such as ambient temperature and gas detection, which help to improve the overall construction safety management level. This type of helmet can detect and alarm autonomously, and is suitable for power construction scenarios in high-risk environments.

[0005] The prior art has the following deficiencies:

[0006] When the existing power construction helmet wearing detection system analyzes the collected images based on computer vision image recognition technology, if an abnormality occurs during the analysis process and the system fails to perceive the abnormality in time, it will lead to a series of serious consequences. The system fails to detect the abnormality in image analysis in time, which may result in the inability to effectively identify workers who do not wear helmets correctly, thereby increasing the risk of safety accidents at the construction site. Especially in high-risk environments, insufficient head protection may cause serious injuries or even fatal accidents. In addition, the failure to detect system abnormalities will also give managers a false sense of security, leading to lax safety supervision and expanded management blind spots, greatly increasing the probability of work-related accidents. Once an accident occurs, the company will not only face legal liability and economic losses, but may also suffer multiple blows of project delays and damaged reputation, seriously affecting the normal operation and long-term development of the company.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0008] The purpose of the present invention is to provide a safety helmet wearing detection system and detection method for electric power construction to solve the problems in the above-mentioned background technology.

[0009] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: a safety helmet wearing detection system for electric power construction, characterized in that it includes a data acquisition module, a hidden danger analysis module, a comparison module, a comprehensive analysis module and an early warning module;

[0010] The data acquisition module collects various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and transmits the system response performance information and algorithm stability information to the abnormality analysis module after processing;

[0011] The hidden danger analysis module establishes a data analysis model with the processed system response performance information and algorithm stability information, generates a data evaluation index, and transmits the data evaluation index to the comparison module;

[0012] The comparison module compares the data evaluation index generated when analyzing the image data based on the image recognition technology with the pre-set data evaluation index reference threshold, generates a high-risk data signal or a low-risk data signal, and transmits the signal to the comprehensive analysis module;

[0013] Comprehensive analysis module. After receiving high-risk data information, it conducts comprehensive analysis on a data set of several data evaluation indexes subsequently generated when analyzing image data based on image recognition technology, generates accidental signals or non-accidental signals, and transmits the signals to the early warning module, which issues or does not issue early warning signals.

[0014] Preferably, the system response performance information includes an abnormal perception delay index. After collection, the data acquisition module calibrates the abnormal perception delay index as APD, and the algorithm stability information includes an identification confidence fluctuation index. After collection, the data acquisition module calibrates the identification confidence fluctuation index as RCF.

[0015] Preferably, the logic for obtaining the abnormal perception delay index APD is as follows:

[0016] Under the detection window, the system records the image data and corresponding processing time at each time point, and the image acquisition time is calibrated as t i , the time when the system issues anomaly detection results is t d,i ;

[0017] For each detection event, the detection time delay is calculated, that is, the time difference from image acquisition to detection result. The calculation expression is as follows: Δti =t d,i -t i , where Δt i Indicates time delay;

[0018] Collect all time delays Δt within the detection window i , and perform weighted processing on it. The calculation expression is as follows: Δt w,i =ω i ·Δt i , where Δt w,i represents the weighted delay time, ω i represents the weighting factor;

[0019] The weighted delay time is processed by a nonlinear function to simulate the system's sensitivity to different delays. The Logistic function is used to process the nonlinear impact. The calculation expression is as follows:

[0020]

[0021] In the formula, Φ(Δt w,i ) represents the delay impact function, which is used to simulate the response sensitivity of the system to different delays, θ represents the threshold, and σ is the sensitivity parameter of the delay impact function, which determines the smoothness of the function;

[0022] The final abnormal perception delay index is calculated by weighting the delay impact function. The calculation expression is as follows:

[0023]

[0024] Where APD represents the abnormal perception delay index, φ i is the weight factor of the delay impact function, and N represents the total number of anomaly detection events in the dataset.

[0025] Preferably, the logic for obtaining the recognition confidence fluctuation index RCF is as follows:

[0026] Under the detection window, a series of continuous image frame data is obtained, which is represented by I = {I x}={I1,I2,I3,…,I D}, where D represents the total number of image frames;

[0027] For each frame image I x Perform feature extraction and obtain the feature vector f x , denoted as f x =[f xu ]=[f x1 ,f x2 ,...,f xm ] T, where m represents the dimension of the feature vector, and T represents the feature vector f x Perform a transpose operation to transform it into a column vector;

[0028] Calculate the familiarity between adjacent frames to measure the feature vector f x and the eigenvector f x The difference between +1 is calculated as follows: Where, Sim(I x ,I x+1 ) represents the degree of familiarity, that is, the x-th frame image I x and the next adjacent frame image I x+1 The similarity between 2 is the weight of the eigenvector difference;

[0029] The recognition confidence fluctuation index is generated by taking a weighted average of the similarities of adjacent image frames. The calculation expression is as follows: RCF represents the recognition confidence fluctuation index, α x is the weight of the x-th frame image.

[0030] Preferably, after the hidden danger analysis module obtains the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, a data analysis model is established to generate a data evaluation index Data As π , based on the following formula:

[0031]

[0032] Where r1 and r2 are preset proportional coefficients of the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, respectively, and both r1 and r2 are greater than 0.

[0033] Preferably, the comparison module will compare the data evaluation index generated when analyzing the image data based on the image recognition technology with a pre-set data evaluation index reference threshold. If the data evaluation index is greater than or equal to the pre-set data evaluation index reference threshold, a high-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module. If the data evaluation index is less than the pre-set data evaluation index reference threshold, a low-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module.

[0034] Preferably, after receiving the high-risk data information, the comprehensive analysis module establishes a data set for a number of data evaluation indexes subsequently generated when analyzing the image data based on the image recognition technology, and marks the data set as H, then:

[0035] H={Data As π}={Data As1, Data As2, Data As3,...Data As f}, then π=1, 2, 3, ..., f, f is a positive integer;

[0036] Calculate the mean and standard deviation of several data evaluation indexes in the data set, and calibrate the mean and standard deviation of the data evaluation index to E1 and E2 respectively, then:

[0037]

[0038] The average values ​​E1 and standard deviations E2 of several data evaluation indices in the data set are compared with the pre-set average value reference threshold value and standard deviation reference threshold value of the data evaluation indices, respectively. The comparison results are as follows:

[0039] If the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is less than the preset standard deviation reference threshold of the data evaluation index, an accidental signal is generated through the comprehensive analysis module and the signal is transmitted to the early warning module, and an early warning signal is not issued through the early warning module;

[0040] If the average value E1 of several data evaluation indexes in the data set is less than the preset reference threshold value of the average value of the data evaluation index and the standard deviation E2 of the data evaluation index is greater than or equal to the preset reference threshold value of the standard deviation of the data evaluation index, or the average value E1 of several data evaluation indexes in the data set is greater than the preset reference threshold value of the average value of the data evaluation index, a non-accidental signal is generated by the comprehensive analysis module, and the signal is transmitted to the early warning module, and an early warning prompt is issued through the early warning module.

[0041] A method for detecting the wearing of a safety helmet for electric power construction comprises the following steps:

[0042] Collect various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and process the collected system response performance information and algorithm stability information;

[0043] Establish a data analysis model with the processed system response performance information and algorithm stability information to generate a data evaluation index;

[0044] Compare the data evaluation index generated when analyzing the image data based on the image recognition technology with a preset data evaluation index reference threshold to generate a high-risk data signal or a low-risk data signal;

[0045] After receiving high-risk data information, a data set is established for comprehensive analysis of a number of data evaluation indices subsequently generated when analyzing image data based on image recognition technology, generating accidental signals or non-accidental signals, and issuing or not issuing warnings for the signals;

[0046] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0047] The present invention collects various parameter information based on image recognition technology analysis through a data acquisition module, including system response performance information and algorithm stability information. The abnormal perception delay index is used as system response performance information to evaluate the delay degree of abnormality detection in the image analysis process. The recognition confidence fluctuation index is used as algorithm stability information to evaluate the stability of the image recognition result. This information is used to establish a data analysis model through a hidden danger analysis module to generate a data evaluation index, which is transmitted to a comparison module for comparison with a preset threshold, and then a high-risk or low-risk signal is generated. After comprehensive analysis, it is transmitted to the early warning module. This process ensures that the system can detect abnormal situations in the image analysis process in real time and warn managers in time, thereby avoiding workers who do not wear safety helmets correctly from being missed, reducing the risk of safety accidents on construction sites, and avoiding the problems of lax safety supervision and expanded management blind spots due to managers' failure to notice system abnormalities.

[0048] The present invention generates accidental or non-accidental signals by comparing the mean value and standard deviation of the data evaluation index in the data set with a preset reference threshold through a comprehensive analysis module, and then decides whether to trigger a warning signal. If the evaluation index is within the preset threshold, the system generates an accidental signal and sends a warning signal to the manager to help respond to potential safety hazards in a timely manner. Through the implementation of the above scheme, not only the efficiency and accuracy of safety management are improved, but also the legal liability, economic losses, and the double blow of project progress and reputation that may be caused by safety accidents to the enterprise are effectively avoided, which helps to ensure the normal operation and long-term development of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0050] Figure 1 The present invention is a module schematic diagram of a safety helmet wearing detection system for electric power construction.

[0051] Figure 2 The present invention is a method flow chart of a method for detecting the wearing of a safety helmet for electric power construction. DETAILED DESCRIPTION

[0052] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0053] The present invention provides Figure 1 A safety helmet wearing detection system for electric power construction shown includes a data acquisition module, a hidden danger analysis module, a comparison module, a comprehensive analysis module and an early warning module;

[0054] The data acquisition module collects various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and transmits the system response performance information and algorithm stability information to the abnormality analysis module after processing;

[0055] The system response performance information includes the abnormal perception delay index. After collection, the data collection module calibrates the abnormal perception delay index as APD. The algorithm stability information includes the recognition confidence fluctuation index. After collection, the data collection module calibrates the recognition confidence fluctuation index as RCF.

[0056] The abnormal perception delay index is a key indicator in the system response performance information. It is used to measure the time delay from detection to feedback or alarm when the power construction helmet wearing detection system identifies an abnormal situation (such as workers not wearing or wearing an incorrect helmet);

[0057] When image recognition technology analyzes image data, changes in abnormal perception delay may cause a series of abnormal hidden dangers. First, if the abnormal perception delay increases, it means that the system's detection and response speed to potential abnormal conditions in the image (such as workers not wearing safety helmets) becomes slow. This delay will cause the system to be unable to capture and process the workers' wearing status in real time, so that the situation of not wearing or wearing incorrectly is not recognized in time, which will not only reduce the monitoring effect of the system, but also may prolong the time that workers are in dangerous conditions, thereby increasing the risk of safety accidents.

[0058] Secondly, the increase in abnormal perception delay may also cause the system's alarm mechanism to fail. If the system is slow to respond when it detects that a worker is not wearing a safety helmet, the relevant alarm and notification mechanism may not be triggered in time, which will affect the on-site management's rapid response to safety hazards. Since the system fails to issue an alarm in time, the management may not be able to take necessary safety measures in the shortest time, such as correcting illegal wearing behavior or conducting on-site inspections, thereby increasing the possibility of accidents.

[0059] Finally, changes in abnormality perception delay will also affect the overall reliability and stability of the system. Larger delays not only affect the efficiency of handling single abnormal events, but may also have a negative impact on the long-term performance of the system. The system may show more recognition errors and response delays in frequent or complex operating environments, leading to long-term accumulation of safety hazards and management loopholes, thereby affecting the overall safety management level and accident prevention effect. Therefore, maintaining a low abnormality perception delay is crucial to ensure the efficient operation of the system and reduce potential hazards.

[0060] The logic for obtaining the abnormal perception delay index APD is as follows:

[0061] Under the detection window, the system records the image data and corresponding processing time at each time point, and the image acquisition time is calibrated as t i , the time when the system issues anomaly detection results is t d,i ;

[0062] For each detection event, the detection time delay is calculated, that is, the time difference from image acquisition to detection result. The calculation expression is as follows:

[0063] Δt i =t d,i -t i ;

[0064] In the formula, Δt i Indicates time delay;

[0065] Collect all time delays Δt within the detection window i , and perform weighted processing on it. The calculation expression is as follows:

[0066] Δt w,i =ω i ·Δt i ;

[0067] In the formula, Δt w,i represents the weighted delay time, ω i represents the weighting factor;

[0068] It should be noted that determining the weighting factors generally involves the following steps and considerations to ensure the accuracy and usefulness of the Anomaly Perceived Latency Index:

[0069] 1. Importance analysis:

[0070] First, the importance of each abnormal detection event needs to be analyzed, which includes evaluating the impact of the event on the workflow or safety management. For example, for a safety helmet wearing detection system for power construction, it may be necessary to consider the risk level of not wearing a safety helmet correctly, such as the degree of injury that may be caused in a high-risk working environment.

[0071] 2. Event complexity:

[0072] Consider the complexity and scope of the event. Events with high complexity may require a higher weighting factor to ensure that the system can detect and respond in a timely manner. Complexity can usually be assessed by factors such as environmental conditions and the number of people involved in the event.

[0073] 3. Historical data analysis:

[0074] Based on past data analysis, evaluate the frequency of similar events and their impact on system operation. Historical data can provide the frequency distribution of events and the actual impact, which helps determine the reasonable range of weighting factors.

[0075] 4. Real-time monitoring and feedback:

[0076] If possible, real-time monitoring and feedback is recommended. By monitoring the system operation in real time, the weighting factors can be adjusted to reflect the actual impact of events and system performance under different conditions.

[0077] 5. System optimization and iteration:

[0078] The determination of weighting factors is a dynamic process and should be adjusted as system performance is optimized and the environment changes. The weighting factors should be evaluated and iterated regularly to ensure that the system can provide a stable and efficient abnormal perception delay index under different circumstances.

[0079] In summary, determining the weighting factor requires comprehensive consideration of factors such as the importance, complexity, historical data, expert opinions, and real-time monitoring of events. Reasonable weighting factors can improve the accuracy and practicality of the abnormal perception delay index, thereby effectively helping to improve the security and management efficiency of the system;

[0080] The weighted delay time is processed by a nonlinear function to simulate the system's sensitivity to different delays. The Logistic function is used to process the nonlinear impact. The calculation expression is as follows:

[0081]

[0082] In the formula, Φ(Δt w,i ) represents the delay impact function, which is used to simulate the response sensitivity of the system to different delays, θ represents the threshold, and σ is the sensitivity parameter of the delay impact function, which determines the smoothness of the function;

[0083] It should be noted that the Logistic function is a common S-shaped function with the following characteristics:

[0084] S-shaped curve characteristics: The graph of the Logistic function shows a typical S-shaped curve, that is, in (Δt w,i -θ) / σ is close to 0, the function value is close to 0.5, and as (Δt w,i -θ) / σ increases or decreases, the function value gradually approaches 0 or 1;

[0085] Range constraint: The function value range is between 0 and 1, which makes the logistic function often used to represent the cumulative distribution function (CDF) of probability or probability distribution;

[0086] Smoothness: The function has good smoothness properties, and the derivative can be used to represent the slope and rate of change, which is particularly useful for modeling and optimization problems;

[0087] In the process of generating the abnormal perception delay index, the Logistic function is mainly used to simulate the system's response to the abnormal perception delay. The specific functions include:

[0088] Response function: The form of the Logistic function can be used to describe the system's sensitivity to the abnormal perception delay time. By adjusting the parameters in the function, such as the threshold and slope, the system's response degree under different delay conditions can be controlled. For example, when the delay time is small, the function value is close to 1, and the system responds quickly. When the delay time increases, the function value gradually approaches 0, and the system's response becomes slower.

[0089] Smooth adjustment: The smoothness of the logistic function enables it to better adapt to complex actual situations and to adapt to different system requirements and environmental conditions by adjusting parameters. This feature is very useful in simulating the complex response of actual systems to abnormal situations.

[0090] In addition to the Logistic function, there are other functions that can be used for similar response simulations, such as:

[0091] Hyperbolic tangent function (Tanh function): The Tanh function is similar in form to the Logistic function. The Tanh function also has an S-shaped curve, but its function value range is between -1 and 1. It can be normalized and adapted according to specific needs. Gaussian function: The Gaussian function is famous for its bell-shaped curve and is suitable for describing continuous and smooth response situations.

[0092] In summary, the Logistic function, as a common S-shaped function, is used to simulate the response speed and sensitivity of the system in the generation of the abnormal perception delay index. It has good adaptability and smoothness. According to the specific system requirements and response characteristics, other similar functions can be considered to achieve a more accurate and realistic simulation.

[0093] The final abnormal perception delay index is calculated by weighting the delay impact function. The calculation expression is as follows:

[0094]

[0095] Where APD represents the abnormal perception delay index, φ i is the weight factor of the delay impact function, and N represents the total number of anomaly detection events in the dataset;

[0096] Under the detection window, it can be seen from the calculation expression of the abnormal perception delay index that the larger the performance value of the abnormal perception delay index generated when analyzing image data based on image recognition technology, the more serious the delay of the system's response to abnormalities under the detection window, that is, when the system analyzes image data using image recognition technology, there may be a large time difference or delay, resulting in a weakened ability to detect and handle abnormal situations in a timely manner. Therefore, a high abnormal perception delay index indicates that the system faces a higher risk, that is, the probability of abnormalities occurring when analyzing images based on image recognition technology is high. On the contrary, if the abnormal perception delay index is small, it means that the system responds to abnormalities more promptly and can detect and handle abnormalities in image analysis in a shorter time, thereby reducing the probability of abnormalities occurring. The recognition confidence fluctuation index is an important parameter in the algorithm stability information, which is used to describe the accuracy variation of the safety helmet wearing detection system for power construction in identifying the wearing status of workers under different environmental conditions (such as light changes, camera angles, etc.);

[0097] When image recognition technology analyzes image data, the fluctuation of recognition confidence will have a direct impact on the recognition accuracy and stability of the system. First, the increase in recognition confidence fluctuation means that the system's judgment of the workers' helmet wearing status in different scenarios is unstable, and may cause misjudgment due to changes in factors such as light, angle or workers' posture. This instability may cause the system to occasionally mistake workers who are not wearing helmets for those who are wearing them, or vice versa, resulting in an increase in missed or false positives.

[0098] Secondly, the increase in recognition confidence fluctuations may cause the system to improperly handle abnormal situations. When the recognition confidence fluctuates greatly, the system may not be able to accurately determine the worker's helmet wearing status, especially in a rapidly changing working environment. This uncertainty will cause the system to fail to detect and handle potential safety risks in a timely manner at critical moments, increasing the risk of accidents.

[0099] Finally, the increase in recognition confidence fluctuations may also affect managers' trust and reliance on the safety status of the construction site. If managers find that the system recognition results are unstable, they may reduce the importance of system alarm information and mistakenly believe that the system has many false alarms and ignore the real safety hazards. This lack of trust will further weaken the effectiveness of safety management measures and increase the risk of serious injuries or even fatal accidents caused by insufficient head protection for workers in high-risk environments;

[0100] In summary, the change of recognition confidence fluctuation directly affects the stability and accuracy of the safety helmet wearing detection system for power construction, which may cause abnormal hidden dangers in the system when identifying and processing the safety helmet wearing status, thereby increasing the risk of safety accidents at the construction site;

[0101] The logic for obtaining the recognition confidence fluctuation index RCF is as follows:

[0102] Under the detection window, a series of continuous image frame data is obtained, which is expressed as:

[0103] I={I x}={I1,I2,I3,…,I D}; where D represents the total number of image frames;

[0104] It should be noted that image frame data refers to continuously acquired images, usually captured by digital image sensors. In image processing and computer vision, images are decomposed into a series of continuous still images, each of which is called a frame. These frames are arranged in time order to form a continuous image sequence, namely image frame data.

[0105] Image frame data can be obtained in a variety of ways, depending on the application scenario and system design:

[0106] 1. Camera or video camera acquisition: The most common way is to use a camera or professional video camera to capture images in real time in the scene; these devices can be fixed or mobile for different application needs;

[0107] 2. Image sensor capture: Digital image sensors are widely used in modern digital cameras. They can convert optical signals into digital image data. These sensors can continuously output image frame data at a specific frame rate depending on the design and requirements of the device;

[0108] 3. Storage device or real-time transmission: The captured image frame data can be transmitted to a computer or data storage device in real time for subsequent processing and analysis. In some applications, the image can also be stored on the local device or transmitted to a remote server via the network;

[0109] 4. Data acquisition for specific applications: In some specific applications, such as medical imaging, security monitoring, machine vision, etc., specially designed image acquisition devices or sensors may be used to meet specific data needs and processing requirements;

[0110] In summary, the acquisition of image frame data is the capture and conversion of scenes in the real world by camera devices or sensors so that they can be further analyzed and processed in computer systems;

[0111] For each frame image I x Perform feature extraction and obtain the feature vector f x , expressed as:

[0112] f x =[f xu ]=[f x1 ,f x2 ,...,f xm ] T ;

[0113] Where m represents the dimension of the feature vector, T represents the feature vector f x Perform a transpose operation to transform it into a column vector;

[0114] It should be noted that feature extraction for each frame of image is an important step in computer vision and image processing. Its purpose is to extract numerical information from the image that can describe its content and features. The specific method of feature extraction can select different technologies and algorithms according to the application scenario and requirements. Common features include:

[0115] 1. Color features: describe the distribution and properties of colors in an image, which can be represented by statistics such as color histogram, color mean, and variance;

[0116] 2. Texture features: describe the texture structure of the image, such as Local Binary Patterns (LBP), Gray Level Co-occurrence Matrix (GLCM), etc.

[0117] 3. Shape features: describe the shape information of objects or regions in the image, such as edge detection, contour extraction, etc.

[0118] 4. Structural features: Features that describe the spatial relationship between objects in an image, such as Scale-Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), etc.

[0119] 5. Deep learning features: Use deep learning models such as Convolutional Neural Networks (CNNs) to extract high-level semantic features from images, such as recognition of specific objects and scene understanding;

[0120] Calculate the familiarity between adjacent frames to measure the feature vector f x and the eigenvector f x The difference between +1 is calculated as follows:

[0121]

[0122] Where, Sim(I x ,I x+1 ) represents the degree of familiarity, that is, the x-th frame image I x and the next adjacent frame image I x+1 The similarity between 2 is the weight of the eigenvector difference;

[0123] The recognition confidence fluctuation index is generated by taking a weighted average of the similarities of adjacent image frames. The calculation expression is as follows:

[0124]

[0125] RCF represents the recognition confidence fluctuation index, α x is the weight of the x-th frame image;

[0126] Under the detection window, it can be seen from the calculation expression of the recognition confidence fluctuation index that the larger the recognition confidence fluctuation index performance value generated when analyzing image data based on image recognition technology, the greater the probability of abnormality when analyzing image data based on image recognition technology. This is because a high fluctuation index reflects the significant difference in features between adjacent frame images, which may increase the inconsistency of recognition results due to environmental changes, object movement, or stability issues of the algorithm itself. On the contrary, the smaller the recognition confidence fluctuation index performance value generated when analyzing image data based on image recognition technology, the relatively stable recognition process, so the probability of abnormality is relatively low;

[0127] The hidden danger analysis module establishes a data analysis model with the processed system response performance information and algorithm stability information, generates a data evaluation index, and transmits the data evaluation index to the comparison module;

[0128] The hidden danger analysis module obtains the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, establishes a data analysis model, and generates the data evaluation index Data As π, based on the following formula:

[0129]

[0130] Where r1 and r2 are preset proportional coefficients of the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, respectively, and both r1 and r2 are greater than 0.

[0131] It can be seen from the data evaluation index calculation expression that the larger the abnormal perception delay index performance value generated when analyzing image data based on image recognition technology, the larger the recognition confidence fluctuation index performance value, that is, the larger the data evaluation index performance value, the greater the probability of abnormality when analyzing image data based on image recognition technology. The smaller the abnormal perception delay index performance value generated when analyzing image data based on image recognition technology, the smaller the recognition confidence fluctuation index performance value, that is, the smaller the data evaluation index performance value, the smaller the probability of abnormality when analyzing image data based on image recognition technology.

[0132] The comparison module compares the data evaluation index generated when analyzing the image data based on the image recognition technology with the pre-set data evaluation index reference threshold, generates a high-risk data signal or a low-risk data signal, and transmits the signal to the comprehensive analysis module;

[0133] The comparison module compares the data evaluation index generated when analyzing the image data based on the image recognition technology with the pre-set data evaluation index reference threshold. If the data evaluation index is greater than or equal to the pre-set data evaluation index reference threshold, a high-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module. If the data evaluation index is less than the pre-set data evaluation index reference threshold, a low-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module.

[0134] The comprehensive analysis module, after receiving the high-risk data information, conducts a comprehensive analysis on a data set of several data evaluation indexes subsequently generated when analyzing the image data based on the image recognition technology, generates an accidental signal or a non-accidental signal, and transmits the signal to the early warning module, through which the early warning module issues or does not issue an early warning signal;

[0135] After receiving the high-risk data information, the comprehensive analysis module establishes a data set for several data evaluation indexes subsequently generated when analyzing the image data based on image recognition technology, and marks the data set as H, then:

[0136] H={Data As π}={Data As1, Data As2, Data As3,...Data As f}, then π=1, 2, 3, ..., f, f is a positive integer;

[0137] Calculate the mean and standard deviation of several data evaluation indexes in the data set, and calibrate the mean and standard deviation of the data evaluation index to E1 and E2 respectively, then:

[0138]

[0139] The average values ​​E1 and standard deviations E2 of several data evaluation indices in the data set are compared with the pre-set average value reference threshold value and standard deviation reference threshold value of the data evaluation indices, respectively. The comparison results are as follows:

[0140] If the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is less than the preset standard deviation reference threshold of the data evaluation index, an accidental signal is generated through the comprehensive analysis module and the signal is transmitted to the early warning module, and an early warning signal is not issued through the early warning module;

[0141] It should be noted that when the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is less than the preset standard deviation reference threshold of the data evaluation index, it indicates that the situation in which the data evaluation index in the data set is greater than or equal to the data evaluation index reference threshold is accidental;

[0142] If the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is greater than or equal to the preset standard deviation reference threshold of the data evaluation index, or the average value E1 of several data evaluation indexes in the data set is greater than the preset average value reference threshold of the data evaluation index, a non-accidental signal is generated through the comprehensive analysis module, and the signal is transmitted to the early warning module, and an early warning prompt is issued through the early warning module to prompt the background staff that the system has abnormal hidden dangers, and the system operation status is checked in time to prevent safety accidents caused by untimely system early warning;

[0143] It should be noted that when the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is greater than or equal to the preset standard deviation reference threshold of the data evaluation index, or the average value E1 of several data evaluation indexes in the data set is greater than the preset average value reference threshold of the data evaluation index, it indicates that the situation in which the data evaluation index in the data set is greater than or equal to the data evaluation index reference threshold is not accidental;

[0144] The present invention collects various parameter information based on image recognition technology analysis through a data acquisition module, including system response performance information and algorithm stability information. The abnormal perception delay index is used as system response performance information to evaluate the delay degree of abnormal detection in the image analysis process. The recognition confidence fluctuation index is used as algorithm stability information to evaluate the stability of the image recognition result. This information is used to establish a data analysis model through a hidden danger analysis module to generate a data evaluation index, which is transmitted to a comparison module for comparison with a preset threshold value, and then a high-risk or low-risk signal is generated. After comprehensive analysis, it is transmitted to an early warning module. This process ensures that the system can detect abnormal conditions in the image analysis process in real time and warn managers in time, thereby avoiding workers who do not wear safety helmets correctly from being missed, reducing the risk of safety accidents at construction sites, and avoiding the problem of lax safety supervision and expansion of management blind spots due to managers' failure to detect system abnormalities.

[0145] The present invention generates accidental or non-accidental signals by comparing the mean value and standard deviation of the data evaluation index in the data set with a preset reference threshold through a comprehensive analysis module, and then decides whether to trigger a warning signal. If the evaluation index is within the preset threshold, the system generates an accidental signal and sends a warning signal to the manager to help respond to potential safety hazards in a timely manner. Through the implementation of the above scheme, not only the efficiency and accuracy of safety management are improved, but also the legal liability, economic losses, and the double blow of project progress and reputation that may be caused by safety accidents to the enterprise are effectively avoided, which helps to ensure the normal operation and long-term development of the enterprise.

[0146] The present invention provides Figure 2 A method for detecting the wearing of a safety helmet for electric power construction is shown, comprising the following steps:

[0147] Collect various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and process the collected system response performance information and algorithm stability information;

[0148] Establish a data analysis model with the processed system response performance information and algorithm stability information to generate a data evaluation index;

[0149] Compare the data evaluation index generated when analyzing the image data based on the image recognition technology with a preset data evaluation index reference threshold to generate a high-risk data signal or a low-risk data signal;

[0150] After receiving high-risk data information, a data set is established for comprehensive analysis of a number of data evaluation indices subsequently generated when analyzing image data based on image recognition technology, generating accidental signals or non-accidental signals, and issuing or not issuing warnings for the signals;

[0151] A method for detecting the wearing of a safety helmet for electric power construction provided in an embodiment of the present invention is realized by the above-mentioned system for detecting the wearing of a safety helmet for electric power construction. The method and process for detecting the wearing of a safety helmet for electric power construction are detailed in the embodiment of the above-mentioned system for detecting the wearing of a safety helmet for electric power construction, which will not be repeated here.

[0152] The above only describes some exemplary embodiments of the present invention by way of illustration. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A safety helmet wearing detection system for electric power construction, characterized in that: Including data collection module, hidden danger analysis module, comparison module, comprehensive analysis module and early warning module; The data acquisition module collects various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and transmits the system response performance information and algorithm stability information to the abnormality analysis module after processing; The hidden danger analysis module establishes a data analysis model with the processed system response performance information and algorithm stability information, generates a data evaluation index, and transmits the data evaluation index to the comparison module; The comparison module compares the data evaluation index generated when analyzing the image data based on the image recognition technology with the pre-set data evaluation index reference threshold, generates a high-risk data signal or a low-risk data signal, and transmits the signal to the comprehensive analysis module; Comprehensive analysis module. After receiving high-risk data information, it conducts comprehensive analysis on a data set of several data evaluation indexes subsequently generated when analyzing image data based on image recognition technology, generates accidental signals or non-accidental signals, and transmits the signals to the early warning module, which issues or does not issue early warning signals.

2. A safety helmet wearing detection system for electric power construction according to claim 1, characterized in that: The system response performance information includes the anomaly perception delay index. After collection, the data acquisition module calibrates the anomaly perception delay index as APD. The algorithm stability information includes the recognition confidence fluctuation index. After collection, the data acquisition module calibrates the recognition confidence fluctuation index as RCF.

3. A safety helmet wearing detection system for electric power construction according to claim 2, characterized in that: The logic for obtaining the abnormal perception delay index APD is as follows: Under the detection window, the system records the image data and corresponding processing time at each time point, and the image acquisition time is calibrated as t i , the time when the system issues anomaly detection results is t d,i ; For each detection event, the detection time delay is calculated, that is, the time difference from image acquisition to detection result. The calculation expression is as follows: Δt i =t d,i -t i ; In the formula, Δt i Indicates time delay; Collect all time delays Δt within the detection window i , and perform weighted processing on it. The calculation expression is as follows: Δt w,i =ω i ·Δt i ; In the formula, Δt w,i represents the weighted delay time, ω i represents the weighting factor; The weighted delay time is processed by a nonlinear function to simulate the system's sensitivity to different delays. The Logistic function is used to process the nonlinear impact. The calculation expression is as follows: In the formula, Φ(Δt w,i ) represents the delay impact function, which is used to simulate the response sensitivity of the system to different delays, θ represents the threshold, and σ is the sensitivity parameter of the delay impact function, which determines the smoothness of the function; The final abnormal perception delay index is calculated by weighting the delay impact function. The calculation expression is as follows: Where APD represents the abnormal perception delay index, φ i is the weight factor of the delay impact function, and N represents the total number of anomaly detection events in the dataset.

4. A safety helmet wearing detection system for electric power construction according to claim 2, characterized in that: The logic for obtaining the recognition confidence fluctuation index RCF is as follows: Under the detection window, a series of continuous image frame data is obtained, which is expressed as: I={I x }={I1,I2,I3,…,I D }, where D represents the total number of image frames; Perform feature extraction on each frame image Ix to obtain the feature vector f x , expressed as: f x =[f xu ]=[f x1 ,f x2 ,...,f xm ] T ; Where m represents the dimension of the feature vector, T represents the feature vector f x Perform a transpose operation to transform it into a column vector; Calculate the familiarity between adjacent frames to measure the feature vector f x and the eigenvector f x The difference between +1 is calculated as follows: Where, Sim(I x ,I x+1 ) represents the degree of familiarity, that is, the x-th frame image I x and the next adjacent frame image I x+1 The similarity between 2 is the weight of the eigenvector difference; The recognition confidence fluctuation index is generated by taking a weighted average of the similarities of adjacent image frames. The calculation expression is as follows: RCF represents the recognition confidence fluctuation index, α x is the weight of the x-th frame image.

5. A safety helmet wearing detection system for electric power construction according to claim 4, characterized in that: The hidden danger analysis module obtains the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, establishes a data analysis model, and generates the data evaluation index Data As π , based on the following formula: Where r1 and r2 are preset proportional coefficients of the abnormal perception delay index calibrated as APD and the recognition confidence fluctuation index calibrated as RCF, respectively, and both r1 and r2 are greater than 0.

6. A safety helmet wearing detection system for electric power construction according to claim 5, characterized in that: The comparison module will compare the data evaluation index generated when analyzing the image data based on the image recognition technology with the pre-set data evaluation index reference threshold. If the data evaluation index is greater than or equal to the pre-set data evaluation index reference threshold, a high-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module. If the data evaluation index is less than the pre-set data evaluation index reference threshold, a low-risk data signal is generated through the comparison module and the signal is transmitted to the comprehensive analysis module.

7. A safety helmet wearing detection system for electric power construction according to claim 6, characterized in that: After receiving the high-risk data information, the comprehensive analysis module establishes a data set for several data evaluation indexes subsequently generated when analyzing the image data based on image recognition technology, and marks the data set as H, then: H={Data As π }={Data As1, Data As2, Data As3,...Data As f }, then π=1, 2, 3, ..., f, f is a positive integer; Calculate the mean and standard deviation of several data evaluation indexes in the data set, and calibrate the mean and standard deviation of the data evaluation index to E1 and E2 respectively, then: The average values ​​E1 and standard deviations E2 of several data evaluation indices in the data set are compared with the pre-set average value reference threshold value and standard deviation reference threshold value of the data evaluation indices, respectively. The comparison results are as follows: If the average value E1 of several data evaluation indexes in the data set is less than the preset average value reference threshold of the data evaluation index and the standard deviation E2 of the data evaluation index is less than the preset standard deviation reference threshold of the data evaluation index, an accidental signal is generated through the comprehensive analysis module and the signal is transmitted to the early warning module, and an early warning signal is not issued through the early warning module; If the average value E1 of several data evaluation indexes in the data set is less than the preset reference threshold value of the average value of the data evaluation index and the standard deviation E2 of the data evaluation index is greater than or equal to the preset reference threshold value of the standard deviation of the data evaluation index, or the average value E1 of several data evaluation indexes in the data set is greater than the preset reference threshold value of the average value of the data evaluation index, a non-accidental signal is generated by the comprehensive analysis module, and the signal is transmitted to the early warning module, and an early warning prompt is issued through the early warning module.

8. A method for detecting the wearing of a safety helmet for electric power construction, implemented by a system for detecting the wearing of a safety helmet for electric power construction as claimed in any one of claims 1 to 7, characterized in that: The following steps are involved: Collect various parameter information generated when analyzing image data based on image recognition technology, including system response performance information and algorithm stability information, and process the collected system response performance information and algorithm stability information; Establish a data analysis model with the processed system response performance information and algorithm stability information to generate a data evaluation index; Compare the data evaluation index generated when analyzing the image data based on the image recognition technology with a preset data evaluation index reference threshold to generate a high-risk data signal or a low-risk data signal; After receiving high-risk data information, a data set is established for comprehensive analysis of several data evaluation indices subsequently generated when analyzing image data based on image recognition technology, generating accidental signals or non-accidental signals, and issuing or not issuing an early warning for the signal.

Images