Safety protection equipment state evaluation method and device, electronic equipment and storage medium

Abstract

The application discloses a kind of safety protection equipment state evaluation method, device, electronic equipment and storage medium, it is related to safety monitoring technical field, specifically can be applied to financial and medical field, can make the insurance enterprise that provides engineering risk and other risk can master construction safety dynamic in time, make medical institution can discover whether the protective clothing is damaged in time, provide strong guarantee for the safety of construction workers or medical staff.The method comprises: real-time acquisition is carried out safety protection equipment state evaluation video stream, feature point extraction is carried out to safety protection equipment in video stream, and effective feature point set is generated;The posture of target object is identified to the object posture data that is obtained in video stream wearing safety protection equipment, and effective feature point set is carried out three-dimensional feature fusion processing, and equipment real-time three-dimensional feature point set is obtained;Difference evaluation is carried out to equipment real-time three-dimensional feature point set and equipment standard three-dimensional feature, and state evaluation result is obtained and output.

Description

Technical Field

[0001] This invention relates to the field of security monitoring technology, and can be specifically applied to the financial and medical fields. In particular, it relates to a method, device, electronic equipment, and storage medium for assessing the status of security protection equipment. Background Technology

[0002] In the financial, insurance, and medical fields, the testing of safety protective equipment is crucial. For example, in the financial and insurance industry, the condition of construction workers' safety clothing is a key factor in assessing site safety risks and determining compensation amounts during the underwriting and claims process for engineering and liability insurance. Testing the integrity of safety clothing helps reduce the insurance company's payout risk and costs. In the medical industry, the protective clothing worn by medical personnel is an important barrier against pathogens and ensures their own safety. Its quality and condition directly affect the health of medical personnel and the smooth operation of medical work. Timely and effective testing of protective clothing is a vital part of medical safety management.

[0003] In related technologies, when assessing the status of safety protection equipment, a combination of manual inspection and image comparison or sensor monitoring is used. Manual inspection relies on safety personnel to check regularly, while image comparison involves comparing real-time images of the safety protection equipment with preset standard images to determine its status. Sensor monitoring involves installing sensors on the safety protection equipment to capture physical changes.

[0004] In the process of implementing the relevant technology, the applicant recognized that the relevant technology has at least the following technical problems: Manual inspections are inefficient and subjective, while image comparison cannot cope with the large deformations of safety equipment caused by changes in human posture during operation, which can easily lead to misjudgments. Sensor monitoring may be affected by factors such as installation location and environmental interference, which can easily lead to missed detection of critical damage, posing a safety hazard to construction workers or medical personnel. Summary of the Invention

[0005] In view of this, the present invention provides a method, device, electronic device and storage medium for assessing the condition of safety protection equipment. The main purpose is to solve the problem that current methods are prone to missing critical damage, which poses a safety hazard to construction workers or medical personnel.

[0006] According to a first aspect of the present invention, a method for assessing the condition of safety protection equipment is provided, the method comprising: The system acquires video streams of the safety protection equipment to be assessed in real time, extracts feature points from the safety protection equipment in the video stream, and generates a set of effective feature points. The target object wearing the safety protection equipment in the video stream is subjected to posture recognition to obtain object posture data, and the effective feature point set is subjected to three-dimensional feature fusion processing in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. Determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

[0007] According to a second aspect of the present invention, a safety protection equipment status assessment device is provided, the device comprising: The feature point extraction module is used to acquire video streams of security protection equipment status assessment in real time, extract feature points of the security protection equipment in the video stream, and generate a set of effective feature points. The three-dimensional feature fusion module is used to perform posture recognition on the target object wearing the safety protection equipment in the video stream to obtain object posture data, and to perform three-dimensional feature fusion processing on the effective feature point set in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. The evaluation module is used to determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

[0008] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0009] According to a fourth aspect of the present invention, a storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the first aspects above.

[0010] By employing the above technical solutions, this invention provides a method, device, electronic device, and storage medium for assessing the condition of safety protection equipment. Targeting the financial insurance and medical fields, this invention, through dynamic acquisition of effective feature points, can avoid misjudgments caused by lighting, stains, or normal deformation, ensuring stable capture of safety protection equipment features. Furthermore, by combining three-dimensional feature fusion processing of the feature points, it overcomes the limitations of two-dimensional vision, accurately distinguishing wrinkles from actual damage to safety protection equipment, improving the reliability of the assessment. This allows insurance companies providing engineering insurance and other types of insurance to promptly grasp the dynamics of construction safety, and enables medical institutions to promptly detect whether protective clothing has been damaged, providing strong protection for the safety of construction workers or medical personnel.

[0011] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0012] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an application environment for a safety protection equipment status assessment method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for assessing the status of safety protection equipment according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating a specific implementation of step S10; Figure 4 This is a flowchart illustrating a specific implementation of step S12; Figure 5 This is a flowchart illustrating a specific implementation of step S20; Figure 6 This is a flowchart illustrating a specific implementation of step S30; Figure 7 This is a schematic diagram of the technical solution architecture of the safety protection equipment status assessment method in one embodiment of the present invention; Figure 8 This is a schematic diagram of a safety protection equipment status assessment device according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention; Figure 10 This is another structural schematic diagram of an electronic device according to one embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] The safety protection equipment status assessment method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can acquire a video stream of the security equipment to be assessed in real time using a camera component. It then extracts feature points from the security equipment in the video stream to generate a set of valid feature points. Next, it performs posture recognition on the target object wearing the security equipment in the video stream to obtain object posture data. Combining this object posture data with the valid feature point set, it performs 3D feature fusion processing to obtain a real-time 3D feature point set for the equipment. Finally, it determines a preset standard 3D feature for the equipment, evaluates the difference between the real-time 3D feature point set and the standard 3D feature, and obtains a status assessment result for the security equipment in the video stream, which is then output to the client so that the client user is aware of the real-time status of the security equipment.

[0015] In this invention, targeting the financial insurance and medical fields, dynamic acquisition of effective feature points avoids misjudgments caused by lighting, stains, or normal deformation, ensuring stable capture of safety equipment features. Furthermore, combined with three-dimensional feature fusion processing of these feature points, it overcomes the limitations of two-dimensional vision, accurately distinguishing wrinkles from actual damage to safety equipment, improving the reliability of assessments. This allows insurance companies providing engineering insurance and other types of insurance to promptly monitor construction safety, and enables medical institutions to promptly detect damage to protective clothing, providing strong protection for the safety of construction workers and medical personnel. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a dedicated server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0016] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the safety protection equipment status assessment method provided in this embodiment of the invention includes the following steps: S10: Acquire the video stream of the security protection equipment to be evaluated in real time, extract feature points of the security protection equipment in the video stream, and generate a set of effective feature points.

[0017] The safety protection equipment status assessment method provided by this invention can be applied to risk warning systems in various application scenarios. Risk warning systems are usually implemented through a server. The server can acquire video streams of the safety protection equipment to be assessed, so as to assess the status of the safety protection equipment worn by relevant objects in the video stream in real time.

[0018] In this embodiment of the invention, after acquiring the video stream of the safety protection equipment to be assessed in real time, the risk warning system employs an adaptive region sampling method to extract feature points from the safety protection equipment in the video stream to generate a set of effective feature points. The adaptive region sampling focuses on key protective areas of the safety protection equipment; for example, in construction worker safety clothing, it focuses on reflective strips at the elbows and abrasion-resistant layers at the shoulders; in medical personnel protective clothing, it focuses on the collar seal and cuff protection areas. The risk warning system utilizes a multi-scale feature extraction algorithm, which can analyze images at different scales to adapt to complex lighting conditions (such as direct sunlight and shadow occlusion) and potential dirt interference. Furthermore, when extracting feature points, the risk warning system prioritizes retaining feature points in high-contrast areas, such as the edges of reflective strips and seam interfaces. These areas have distinct features and can better reflect the inherent attributes of safety protection equipment. At the same time, it filters out temporary texture points caused by equipment wrinkles to avoid these unstable points affecting subsequent evaluations. It also scores the sampling points for repetition detection rate and retains only points that can be stably identified in multiple consecutive frames as valid feature points to obtain a set of valid feature points. This reduces the number of invalid points involved in the calculation due to image blurring or momentary occlusion, laying a reliable foundation for subsequent matching.

[0019] In this way, by using adaptive region sampling and multi-scale feature extraction algorithms, key areas of safety protective equipment can be accurately located, effectively avoiding interference from lighting, stains, or normal deformation, ensuring that the extracted feature points are stable and reliable, and providing solid data support for subsequent accurate assessment of the condition of safety protective equipment. For example, when assessing the safety protective equipment worn by construction workers in construction scenarios under insurance policies such as engineering insurance, a camera can be used to capture real-time video streams of construction workers wearing safety clothing, and feature points can be extracted from key areas such as the reflective strips at the elbows and the abrasion-resistant layers on the shoulders to generate a valid set of feature points. As another example, in the medical field, when assessing the protective clothing worn by medical personnel, a camera can capture real-time images of medical personnel wearing protective clothing, and feature points can be extracted from key areas such as the collar seal and the cuff protection area to generate a valid set of feature points for accurate assessment of the condition of the protective clothing.

[0020] Among them, such as Figure 3 As shown, step S10, which involves extracting feature points from the security equipment in the video stream to generate a set of valid feature points, includes the following steps: S11: Based on the preset security protection equipment design standards, locate critical protection areas and non-critical protection areas on the security protection equipment in the video stream.

[0021] After acquiring the video stream of the safety equipment to be evaluated, the critical and non-critical protection areas are first determined based on the pre-set safety equipment design standards. These standards are developed through extensive practical experience and research, comprehensively considering factors such as the equipment's protective needs in actual use, stress conditions, and common injury sites. For example, for construction worker safety clothing, critical protection areas typically include reflective strips at the elbows and abrasion-resistant layers at the shoulders. The elbows frequently come into contact with objects during work, and reflective strips improve visibility in low-light conditions, reducing the risk of collisions. The shoulders bear the pressure of carrying tools and during work, and abrasion-resistant layers enhance their wear resistance. Non-critical protection areas are areas such as the clothing body that are relatively less susceptible to serious damage or have minimal impact on protective performance.

[0022] In this way, by identifying critical and non-critical protection areas based on design standards, feature point extraction can be performed in a targeted manner, avoiding wasting computational resources in irrelevant areas. Simultaneously, it ensures that the features of critical protection areas receive sufficient attention, laying the foundation for accurate assessment of the status of safety protective equipment and helping to promptly identify potential safety issues. For example, when assessing the safety protective equipment worn by construction workers in construction scenarios under engineering insurance and other types of insurance, the reflective strips at the elbows and the abrasion-resistant layer on the shoulders are identified as critical protection areas based on the design standards for construction worker safety clothing. Similarly, in the medical field, when assessing protective clothing worn by medical personnel, the collar seal and cuff protection areas are identified as critical protection areas based on the protective clothing design standards.

[0023] S12: Using a preset multi-scale feature extraction algorithm, feature points are extracted from the critical protection area to obtain multiple first effective feature points, and feature points are extracted from the non-critical protection area to obtain multiple second effective feature points.

[0024] After locating critical and non-critical protection areas, a pre-defined multi-scale feature extraction algorithm is used to extract feature points. This algorithm is a technique capable of analyzing images at different scales, adapting to features of varying sizes and levels of detail. Specifically, for critical protection areas, the algorithm focuses on high-contrast regions, such as the edges of reflective strips and seam interfaces, to better reflect the inherent properties of the security equipment. The algorithm performs convolution and downsampling operations on the image at multiple scales to extract stable and representative feature points as the first effective feature points. For non-critical protection areas, the same algorithm is used to extract feature points, resulting in second effective feature points, but with a slightly lower precision and focus compared to critical areas.

[0025] In this way, the multi-scale feature extraction algorithm can adapt to image features at different scales and effectively extract stable feature points in both critical and non-critical protection areas. For critical protection areas, high-precision feature point extraction can accurately capture their state changes and promptly detect potential security risks. For non-critical protection areas, appropriate feature point extraction can provide comprehensive equipment information while avoiding over-computation, which helps improve the accuracy and efficiency of the assessment and provides reliable data support for subsequent assessment and processing.

[0026] For example, when assessing the safety protective equipment worn by construction workers in construction scenarios under insurance policies such as engineering insurance, a multi-scale feature extraction algorithm is used for key protective areas of the safety clothing (such as reflective strips on the elbows). This algorithm analyzes the image features of the reflective strip edges at different scales, extracting multiple stable first effective feature points. For non-critical areas (such as the body of the garment), corresponding second effective feature points are also extracted. This provides data for a comprehensive assessment of the safety clothing's condition, helping to promptly identify wear and tear and take appropriate measures to ensure worker safety. Similarly, in the medical field, when assessing protective clothing worn by healthcare workers, a multi-scale feature extraction algorithm is used for key protective areas (such as the collar seal). This algorithm analyzes features such as the seam interfaces at the collar seal at multiple scales, extracting first effective feature points. For non-critical areas (such as the back of the protective clothing), second effective feature points are extracted. This provides a basis for assessing the overall condition of the protective clothing, helping to promptly identify problems and ensure the safety of healthcare workers.

[0027] Among them, such as Figure 4 As shown, in step S12, a preset multi-scale feature extraction algorithm is used to extract feature points from the key protection area to obtain multiple first effective feature points, including the following steps: S121: A multi-scale feature extraction algorithm is used to extract feature points in the key protection area, resulting in multiple original feature points, and multiple candidate feature points are selected from these original feature points.

[0028] After acquiring the video stream data of the critical protection area, a multi-scale feature extraction algorithm is first used to extract feature points from the critical protection area, resulting in multiple raw feature points. These raw feature points contain various local feature information of the critical protection area, but may contain invalid feature points caused by changes in lighting, dirt interference, or normal deformation. Therefore, in order to select more reliable candidate feature points, the following two methods can be used for processing; One approach is to calculate the local contrast of the local region where each original feature point is located in the video stream. Local contrast measures the brightness difference between a region and its surrounding areas, reflecting the prominence of the feature. Then, original feature points whose local contrast meets a preset contrast condition are selected as candidate feature points. The preset contrast condition indicates whether to extract a specified number of feature points with the highest local contrast or to extract feature points with a local contrast greater than a contrast threshold. For example, in the critical protective area of ​​a construction worker's safety clothing, the edges of reflective strips typically have high local contrast. This method can accurately extract feature points from the edges of reflective strips while filtering out low-contrast feature points caused by stains or normal wrinkles.

[0029] Another approach is to calculate the texture descriptor of the local region where each original feature point is located in the video stream. The texture descriptor indicates the texture rules and structure of the local region, reflecting information such as texture roughness and directionality. Then, based on the texture descriptor corresponding to each original feature point, temporary texture feature points generated by the texture of the local region where the original feature point is located are identified from among the multiple original feature points. Temporary texture feature points are usually generated due to normal wrinkles or texture changes in the safety clothing and are not real damage features. Therefore, temporary texture feature points are filtered out from the multiple original feature points, and the remaining original feature points are used as multiple candidate feature points.

[0030] In this way, by calculating local contrast and texture descriptors, invalid feature points caused by lighting, stains, or normal deformation can be effectively filtered out. This ensures that candidate feature points better reflect the true state of the safety equipment, helping to improve the accuracy of subsequent assessments and avoid misjudgments caused by interference from invalid feature points. This provides insurance companies and medical institutions with more reliable safety assessment data. For example, when assessing the safety equipment worn by construction workers in construction scenarios in insurance products such as engineering insurance, feature points are extracted from key protective areas of the construction worker's safety clothing (such as reflective strips on the elbows). After obtaining multiple original feature points through a multi-scale feature extraction algorithm, the local contrast of each original feature point is calculated. Since the edge of the reflective strip has high contrast under sunlight, while the contrast of the surrounding stains or normal wrinkles is low, candidate feature points at the edge of the reflective strip can be accurately screened. At the same time, temporary texture feature points caused by wrinkles are filtered out by texture descriptors, ensuring that candidate feature points better reflect the actual state of the safety clothing and providing reliable data for insurance companies to assess construction safety. For example, when evaluating protective clothing worn by medical personnel in the medical field, feature points are extracted from key protective areas of the protective clothing (such as the collar seal). After obtaining the original feature points using a multi-scale feature extraction algorithm, the local contrast is calculated. Key features such as the seam interfaces at the collar seal usually have high contrast and can be accurately screened as candidate feature points. At the same time, temporary texture feature points caused by normal wrinkles in the protective clothing are filtered out by texture descriptors to ensure that the candidate feature points reflect the true state of the protective clothing, providing a basis for medical institutions to detect damage to the protective clothing in a timely manner.

[0031] S122: Count the number of consecutive frames corresponding to each candidate feature point in the video stream to extract multiple first effective feature points from multiple candidate feature points.

[0032] After selecting candidate feature points, the number of consecutive frames corresponding to each candidate feature point in the video stream is counted. The number of consecutive frames indicates the number of image frames in which the corresponding candidate feature point is continuously identified in the video stream. Since workers move continuously during the operation, the images in the video stream will change to some extent. However, the features of real safety protection equipment (such as tears or wear on safety clothing) will remain relatively stable in consecutive frames. Meanwhile, some feature points caused by accidental factors (such as brief occlusion or sudden changes in light) will not appear in consecutive frames.

[0033] Furthermore, the risk warning system has a preset threshold, which will extract candidate feature points with a number of consecutive frames greater than the threshold as the first valid feature points. For example, if the preset threshold is 3 frames, if a candidate feature point can be stably identified in 3 or more consecutive frames, it is considered a real and reliable first valid feature point. This further filters out invalid feature points caused by accidental factors, ensuring that the first valid feature points better reflect the actual status of the security equipment.

[0034] In this way, by counting the number of consecutive frames and setting a threshold, invalid feature points generated by accidental factors can be effectively filtered out, improving the reliability and stability of the first valid feature point. This helps to more accurately assess the status of security protection equipment, providing insurance companies and medical institutions with more timely and accurate security information, and ensuring the safety of construction workers and medical staff.

[0035] For example, in assessing the safety protective equipment worn by construction workers in construction scenarios under insurance policies such as engineering insurance, after screening candidate feature points for the key protective areas of the construction worker's safety clothing, the number of consecutive frames for each candidate feature point is counted. Assuming a candidate feature point at the reflective strip on the elbow of the safety clothing can be stably identified in 8 consecutive images, while another candidate feature point caused by a brief occlusion only appears in 2 images, by setting a preset threshold of 3 frames, the candidate feature point appearing in 8 consecutive frames is extracted as the first effective feature point, providing a reliable basis for insurance companies to accurately assess the condition of the safety clothing. As another example, in the medical field, when assessing the protective clothing worn by medical personnel, after screening candidate feature points for the key protective areas of the protective clothing, the number of consecutive frames is counted. Assuming a candidate feature point at the cuff of the protective clothing appears stably in 6 consecutive images, while another candidate feature point caused by a sudden change in light only appears in 1 image, by setting a preset threshold of 3 frames, the candidate feature point appearing in 6 consecutive frames is extracted as the first effective feature point, providing accurate information for medical institutions to promptly detect problems with the protective clothing.

[0036] S13: Integrate multiple first effective feature points and multiple second effective feature points to obtain a set of effective feature points.

[0037] In this embodiment of the application, after obtaining multiple first effective feature points and multiple second effective feature points, the multiple first effective feature points and multiple second effective feature points are integrated to obtain a set of effective feature points.

[0038] S20: Perform posture recognition on the target object wearing safety protective equipment in the video stream to obtain the object posture data, and combine the object posture data to perform three-dimensional feature fusion processing on the effective feature point set to obtain the real-time three-dimensional feature point set of the equipment.

[0039] In this embodiment, pose recognition is performed on a target object wearing protective equipment in a video stream. Specifically, a pose estimation model can be used for pose recognition. This model can be a lightweight human pose estimation model, which can reduce computational resource consumption and improve processing speed while maintaining a certain level of accuracy, thus meeting the needs of real-time evaluation.

[0040] After posture recognition, object posture data can be obtained. This data can specifically include a posture depth map or a posture normal map. The posture depth map reflects the distance between the target object and the camera, while the posture normal map describes the normal direction of the object's surface at various points, aiding in understanding the object's three-dimensional structure. Next, by combining the physical properties of the safety equipment fabric—for example, the stiffness coefficient of the elbow pad area in construction workers' safety clothing is higher than that of the body, and the elasticity coefficient of key protective parts in medical personnel's protective clothing—depth deviations caused by postures such as bending, climbing, and rapid movement are corrected, making the spatial information more closely match the actual shape of the safety equipment. During three-dimensional feature fusion, the object posture data needs to be combined, and the weights are dynamically adjusted based on the posture estimation confidence level corresponding to the object posture data. For example, the weight of depth information is reduced when the arm swings rapidly to avoid the influence of posture estimation errors on the accuracy of feature fusion. This upgrades two-dimensional feature points into three-dimensional perception features, resulting in a real-time three-dimensional feature point set for the equipment.

[0041] In this way, by combining object posture data with three-dimensional feature fusion processing, the limitations of two-dimensional vision in capturing the spatial structure and real damage of safety protection equipment can be made up for. Moreover, by dynamically adjusting the weights based on the physical properties of the fabric and the confidence level of posture estimation, the actual state of the safety protection equipment can be reflected more accurately, effectively distinguishing wrinkles from real damage and improving the reliability of the assessment.

[0042] For example, in engineering insurance and other types of insurance, when assessing the safety protective equipment worn by construction workers in construction scenarios, workers are constantly changing postures while working at heights, such as bending over to carry materials or climbing scaffolding. By identifying the worker's posture data and performing 3D feature fusion, a real-time set of 3D feature points of the safety clothing is obtained, accurately reflecting the actual state of the safety clothing in the current posture. As another example, in the medical field, when assessing the protective clothing worn by medical personnel, their postures frequently change as they move around in wards and bend over to care for patients. By identifying the medical personnel's posture data and performing 3D feature fusion, a real-time set of 3D feature points of the protective clothing is obtained, providing a basis for accurately assessing the condition of the protective clothing.

[0043] Among them, such as Figure 5As shown, step S20, which involves performing posture recognition on the target object wearing safety protective equipment in the video stream to obtain object posture data, and combining the object posture data to perform three-dimensional feature fusion processing on the effective feature point set to obtain the real-time three-dimensional feature point set of the equipment, includes the following steps: S21: Obtain a pre-trained pose estimation model, use the pose estimation model to perform pose recognition on the target object, and obtain the object pose data.

[0044] Object pose data includes pose depth maps or pose normal maps; In this embodiment, the pre-trained pose estimation model is trained based on a large amount of human pose data and safety equipment related data. This model can extract human pose-related information by analyzing the image of the target object in the input video stream. The risk warning system can input the video stream into the pose estimation model so that the pose estimation model can identify the pose of the target object and obtain the object pose data. The object pose data includes a pose depth map or a pose normal map. The pose depth map is an image that reflects the distance information between the target object and the camera. Each pixel value in the map represents the distance from the corresponding object to the camera, and it can intuitively show the depth distribution of the target object in space. The pose normal map describes the normal direction of the object's surface at various points. The normal direction is important for understanding the three-dimensional shape and surface structure of the object.

[0045] In this way, obtaining posture depth maps or posture normal maps through pre-trained posture estimation models can provide accurate depth and normal information for subsequent 3D feature fusion, compensating for the shortcomings of planar vision in capturing spatial information. This makes the assessment more closely resemble the actual shape of safety protective equipment and helps to accurately distinguish between wrinkles and actual damage. For example, in engineering insurance and other types of insurance, when assessing the safety protective equipment worn by construction workers in construction scenarios, cameras capture video streams of construction workers in real time. The video frames are input into a pre-trained posture estimation model, which outputs a posture depth map of the construction worker. The posture depth map shows the distance between different parts of the worker's body and the camera, as well as the spatial distribution of the safety clothing, providing an important spatial information basis for subsequent assessment of the safety clothing's condition. As another example, in the medical field, when assessing protective clothing worn by medical personnel, cameras capture video streams of medical personnel wearing protective clothing. After being input into a posture estimation model, the model outputs a posture normal map of the medical personnel. The normal map reflects the normal direction of the protective clothing surface, helping to understand the 3D shape and surface structure of the protective clothing and providing a basis for accurately assessing whether the protective clothing is damaged.

[0046] S22: Obtain the attitude estimation confidence score of the attitude estimation model output for the object's attitude data, and determine the feature fusion weights that match the attitude estimation confidence score.

[0047] In this embodiment, the risk warning system obtains the attitude estimation confidence score of the attitude estimation model outputting the object's attitude data. The attitude estimation confidence score is a measure of the accuracy of the attitude estimation model's output of the object's attitude data. During model training, it learns accuracy criteria for different attitude conditions based on a large amount of labeled data. Therefore, when the attitude estimation model performs attitude recognition on the input video frame and outputs an attitude depth map or attitude normal map, it simultaneously provides an attitude estimation confidence score. Simultaneously, the risk warning system determines feature fusion weights that match the attitude estimation confidence score. These weights are dynamically adjusted based on the attitude estimation confidence score to balance the importance of different information sources in subsequent 3D feature fusion processes. For example, when the attitude estimation confidence score is high, indicating relatively accurate attitude data, the weight of depth information in feature fusion can be appropriately increased. Conversely, when the attitude estimation confidence score is low, such as due to rapid worker movement or medical personnel turning around causing inaccurate attitude estimation, the weight of depth information is reduced to minimize the impact of inaccurate attitude data on the evaluation results.

[0048] In this way, dynamically adjusting the feature fusion weights based on the attitude estimation confidence level makes the subsequent 3D feature fusion process more flexible and accurate, adapting to different attitude situations, avoiding inaccurate evaluation results due to attitude estimation errors, and improving the reliability of the evaluation. For example, in assessing the safety protection equipment worn by construction workers in construction scenarios in insurance types such as engineering insurance, when the construction worker is stationary or moving slowly, the attitude estimation model outputs a higher attitude estimation confidence level. The risk warning system determines a higher depth information feature fusion weight according to preset rules, allowing depth information to play a greater role in subsequent feature fusion and more accurately reflect the spatial state of the safety clothing. Conversely, when the worker is running quickly, the attitude estimation confidence level decreases, and the risk warning system correspondingly reduces the depth information weight to reduce interference caused by inaccurate attitude estimation. For example, in the medical field, when evaluating the protective clothing worn by medical staff, the confidence level of the posture estimation model is high when the medical staff are walking normally, and the risk warning system determines the appropriate feature fusion weights. However, when the medical staff quickly turn around to take care of patients, the confidence level of the posture estimation decreases, and the risk warning system adjusts the feature fusion weights in a timely manner to reduce the influence of depth information and ensure that the evaluation of the protective clothing status is not greatly interfered with by the posture estimation error.

[0049] S23: Extract multiple valid feature points from the set of valid feature points, and determine the corresponding depth information and noise information for each valid feature point based on the object pose data.

[0050] In this embodiment of the application, the set of effective feature points is the set that has been extracted in the previous steps and includes stable feature points of critical and non-critical protection areas of the security protection equipment. That is, the multiple effective feature points include multiple first effective feature points and multiple second effective feature points, and the first effective feature points and the second effective feature points correspond to the feature points of the critical and non-critical protection areas, respectively.

[0051] After acquiring the object's pose data (pose depth map or pose normal map), for each valid feature point, the risk warning system retrieves the corresponding depth value from the pose depth map based on its position in the image. This depth information reflects the feature point's distance from the camera in space and is crucial for understanding the three-dimensional structure of the safety protection equipment. Simultaneously, considering the various interference factors in real-world scenarios, such as changes in lighting and dirt, which introduce noise, this embodiment of the application estimates noise information by analyzing the changes in pixels surrounding the feature point. This noise information is used to measure the reliability of the feature point data; the greater the noise, the more severe the interference with the feature point data.

[0052] In this way, by determining depth and noise information for each effective feature point through the above process, the spatial state and reliability of the feature point can be more comprehensively described. This provides richer information for subsequent accurate 3D feature fusion, helps improve the accuracy of the assessment, and effectively distinguishes feature changes caused by normal factors and actual damage. For example, when assessing the safety protective equipment worn by construction workers in construction scenarios in insurance types such as engineering insurance, for the first effective feature point at the reflective strip on the elbow of the safety suit, the corresponding depth value is found based on its position in the posture depth map to determine its depth information. At the same time, the brightness changes of the surrounding pixels are analyzed to estimate the noise information. If there is dirt or other interference in the surrounding area, the noise information will indicate that the data of this feature point is less reliable, providing a reference for subsequent assessments. As another example, when assessing the protective clothing worn by medical personnel in the medical field, for the first effective feature point at the collar seal of the protective clothing, the depth information is obtained from the posture depth map, and the noise information is determined by analyzing the surrounding pixels. If the collar is wet with sweat, causing the image to be blurry, the noise information will reflect that the data of this feature point is interfered with, helping the system to more accurately assess the condition of the protective clothing.

[0053] S24: Combining feature fusion weights, calculate the two-dimensional texture features, corresponding depth information, and noise information of each effective feature point to perform three-dimensional feature fusion processing on each effective feature point, and obtain multiple real-time three-dimensional feature points of the equipment corresponding to multiple effective feature points.

[0054] In this embodiment, the risk warning system combines feature fusion weights to calculate the two-dimensional texture features, corresponding depth information, and noise information of each effective feature point, thereby performing three-dimensional feature fusion processing on each effective feature point to obtain multiple real-time three-dimensional feature points of the equipment corresponding to multiple effective feature points. The two-dimensional texture features include information such as the color and texture roughness of the effective feature points on the image plane, reflecting the characteristics of the surface of the safety protection equipment. For each effective feature point, the risk warning system fuses its two-dimensional texture features, depth information, and noise information according to the feature fusion weights determined in the preceding steps. Specifically, the fusion formula shown in Formula 1 below can be used to calculate the real-time three-dimensional feature points of the equipment corresponding to each effective feature point: Formula 1:

[0055] in, For the first The equipment's real-time 3D feature points are obtained by fusing 3D features from 1 effective feature points. For the first Two-dimensional texture features of 1 effective feature point, including information such as color and texture roughness; For the first Depth information corresponding to each valid feature point; The depth information weights are dynamically adjusted based on the confidence level of the pose estimation. The deep information fusion coefficient is used to balance the contribution of deep information to the overall fusion. For the first Noise information corresponding to each effective feature point; The noise information weights are dynamically adjusted based on the attitude estimation confidence. The noise information fusion coefficient is used to balance the influence of noise information in the overall fusion. Using Formula 1, two-dimensional texture features, depth information, and noise information are fused according to certain weights to obtain three-dimensional feature points that reflect the spatial structure and true state of the safety protection equipment.

[0056] In this way, by combining feature fusion weights to calculate various types of information, three-dimensional feature fusion of effective feature points can be achieved. This comprehensively considers the planar and spatial information of feature points as well as data reliability, accurately reflects the real-time status of safety protection equipment, effectively distinguishes wrinkles from actual damage, and improves the reliability of the assessment. For example, when assessing the safety protection equipment worn by construction workers in construction scenarios in insurance types such as engineering insurance, for an effective feature point on the safety suit, based on its two-dimensional texture features (such as the color and texture of reflective strips), depth information (distance from the camera), and noise information (degree of interference from stains), combined with feature fusion weights, the real-time three-dimensional feature point corresponding to the effective feature point is calculated according to Formula 1 above. As another example, when assessing the protective clothing worn by medical personnel in the medical field, for an effective feature point on the protective clothing, comprehensively considering its two-dimensional texture features (such as the color and texture of the protective clothing fabric), depth information (position in space), and noise information (interference from sweat, etc.), the three-dimensional feature point is calculated using Formula 1.

[0057] S25: Generate a set of real-time 3D feature points of the equipment, including multiple real-time 3D feature points of the equipment.

[0058] In this embodiment of the application, after completing the three-dimensional feature fusion processing of all effective feature points to obtain multiple real-time three-dimensional feature points of the equipment, these feature points are grouped together to form a real-time three-dimensional feature point set of the equipment. The real-time three-dimensional feature point set contains the three-dimensional information of feature points of various key and non-key protection areas on the safety protection equipment, which can comprehensively and accurately reflect the real-time status of the safety protection equipment.

[0059] In this way, the real-time 3D feature point set of the equipment generated through the above process provides a complete and accurate data foundation for subsequent safety protection equipment status assessment. This enables the assessment to be based on comprehensive 3D information analysis, improving the accuracy and reliability of the assessment and providing more valuable information for insurance companies and medical institutions, thus ensuring the safety of construction workers and medical personnel. For example, when assessing the safety protection equipment worn by construction workers in construction scenarios under insurance policies such as engineering insurance, the generated real-time 3D feature point set includes 3D information of feature points from various key areas (such as elbows and shoulders) and non-key areas of the safety clothing. As another example, when assessing protective clothing worn by medical personnel in the medical field, the real-time 3D feature point set covers 3D information of feature points from key areas such as the collar and cuffs, as well as other areas of the protective clothing.

[0060] S30: Determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, obtain the status evaluation result of the security protection equipment in the video stream and output it.

[0061] In this embodiment of the invention, the preset standard three-dimensional features of the equipment are obtained by periodically collecting and analyzing the three-dimensional features of safety protective equipment in good condition. These standard features incorporate natural aging factors; for example, after a period of normal use, the fabric of a safety suit will show a certain degree of wear, but this wear is within the normal range and will not affect its protective performance. After acquiring the real-time three-dimensional feature point set of the equipment, the risk warning system will evaluate the difference between it and the preset standard three-dimensional features of the equipment. Different weights are assigned to the critical protective areas of the safety protective equipment (such as the elbows and shoulders of construction workers' safety suits, and the critical sealing parts of medical personnel's protective suits) and non-critical areas (such as the body of the garment), amplifying the weight of damage to critical areas. This makes the evaluation results more reflective of the actual safety condition of the safety protective equipment, thereby obtaining a status evaluation result for the safety protective equipment in the video stream, such as whether the safety suit has slight wear, whether there are tears or reflective strips falling off, and whether the protective suit is damaged and affects its protective performance, and then outputting this result.

[0062] In this way, by pre-setting standard three-dimensional features of equipment that incorporate natural aging factors and employing a layered loss design for difference assessment, the difference between the real-time state and the standard state of safety protective equipment can be accurately quantified, ensuring the accuracy of the assessment results. This allows insurance companies to promptly grasp the dynamics of construction safety, and medical institutions to promptly detect whether protective clothing has been damaged, providing strong protection for the safety of construction workers or medical personnel. For example, in engineering insurance and other types of insurance, when assessing the safety protective equipment worn by construction workers in construction scenarios, the three-dimensional features of safety clothing in good condition are periodically collected as a standard. Real-time video streams of construction workers' safety clothing are acquired, and when comparing the differences with the preset standard, a layered loss design is used, focusing on key areas such as the elbows and shoulders. When it is found that the difference between the elbow feature point of a construction worker's safety clothing and the standard exceeds a threshold, it is assessed as slight wear, and the status assessment result is output in a timely manner to remind the worker to pay attention to protection. At the same time, insurance companies can use this information to grasp the dynamics of construction safety and adjust risk management strategies accordingly. For example, in the medical field, when evaluating the protective clothing worn by medical staff, the three-dimensional features of brand-new and standard-compliant protective clothing are collected periodically as preset standards. The video stream of the protective clothing is acquired in real time and processed to obtain a set of real-time three-dimensional feature points. When evaluating the differences with the preset standards, if it is found that the feature points of the collar sealing area of ​​a medical staff member's protective clothing differ too much from the standard, it is determined after evaluation that the seal is damaged. The status evaluation result is output, and the medical staff member is notified to replace the protective clothing to ensure their safety. At the same time, the hospital can strengthen the management of protective measures based on this.

[0063] Among them, such as Figure 6 As shown, step S30, which involves determining the preset standard three-dimensional features of the equipment, evaluating the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtaining and outputting the status evaluation result of the security protection equipment in the video stream, includes the following steps: S31: Determine the preset evaluation weights.

[0064] In this embodiment, the risk warning system determines preset evaluation weights. These weights are pre-set based on the importance of different areas of the safety equipment, indicating the corresponding weights for critical and non-critical protection areas. Specifically, in safety equipment, critical protection areas (such as the elbows and shoulders of construction workers' safety suits, and key sealing parts of medical personnel's protective clothing) play a crucial role in protective performance. Damage to these areas will severely affect the protective effect, thus they are assigned higher weights. Non-critical protection areas (such as the body of construction workers' safety suits, and non-critical parts of medical personnel's protective clothing) have a relatively smaller impact on protective performance and are assigned lower weights. For example, in construction workers' safety suits, the elbows frequently come into contact with objects during work and are prone to wear and impact, so the evaluation weight for the elbow area is set higher. The shoulders bear the pressure of carrying tools and working, so their evaluation weight is also relatively high. The body of the suit is relatively less susceptible to serious damage, so its evaluation weight is lower. In this way, the importance of different areas in the evaluation can be reasonably allocated, making the evaluation results more consistent with actual safety requirements.

[0065] By setting appropriate assessment weights, the importance of critical protection areas in the assessment can be highlighted, allowing the assessment results to more accurately reflect the actual safety status of safety equipment. This provides insurance companies and medical institutions with more targeted information, helping to promptly identify safety hazards in critical areas and ensuring the safety of construction workers and medical personnel. For example, when assessing the safety equipment worn by construction workers in construction scenarios under insurance policies such as engineering insurance, the assessment weight for critical protection areas such as the elbows and shoulders is set at 0.7, while the assessment weight for non-critical protection areas such as the body of the garment is set at 0.3. This ensures that changes in the condition of critical areas have a greater impact on the assessment results in subsequent assessments, allowing insurance companies to pay closer attention to the safety status of these critical areas and take timely risk control measures. For example, when evaluating protective clothing worn by medical personnel in the medical field, the evaluation weight of key protective areas such as the collar seal and cuff protection areas is set at 0.8, while the evaluation weight of other non-critical areas of the protective clothing is set at 0.2. This allows problems in key areas to be more prominently reflected in the evaluation results, enabling medical institutions to identify problems in a timely manner and ensure the safety of medical personnel.

[0066] S32: By using a preset hierarchical temporal loss function and combining it with evaluation weights, the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment is quantitatively calculated to obtain the loss value, and the loss value is used as the state evaluation result and output.

[0067] In this embodiment, the risk warning system uses a preset hierarchical temporal loss function, combined with evaluation weights, to quantify the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, obtains a loss value, and outputs the loss value as the state evaluation result. The hierarchical temporal loss function is designed to quantify the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and to ensure dynamic stability; specifically, it can be the following formula 2: Formula 2:

[0068] in, It is the base loss value, used to measure the overall feature differences; For the feature point set of the critical protection area, This is a set of feature points for non-critical protection areas; and These parameters are set in conjunction with the previously determined evaluation weights, and > This is to amplify the weight of damage in key areas; For the equipment's real-time 3D feature point set, the first Three-dimensional features of a feature point; For the equipment standard three-dimensional features corresponding to the first A standard three-dimensional feature of a feature point.

[0069] In addition, in practical applications, Formula 2 above can also be combined with temporal constraints. During the calculation process, Formula 3 below can be used to calculate the displacement difference of feature points in consecutive frames. Formula 3:

[0070] in, This represents the displacement difference of feature points in consecutive frames. Indicates the feature point at time t. coordinates Indicates the feature point at time t. The coordinates. When the change When the threshold is exceeded (such as due to abrupt changes caused by motion blur), the risk warning system can use interpolation to correct the feature point coordinates of the current frame, ensuring the continuity of the feature point trajectory and avoiding evaluation confusion in dynamic scenes.

[0071] In this way, through the above process, the hierarchical temporal loss function, combined with evaluation weights, can accurately quantify the difference between the real-time state and the standard state of safety protection equipment. Simultaneously, it ensures the continuity and stability of the evaluation in dynamic scenarios, enabling timely detection of abnormalities in safety protection equipment. This provides insurance companies and medical institutions with reliable state assessment results, facilitating timely measures to ensure safety. For example, in engineering insurance and other types of insurance, when evaluating the safety protection equipment worn by construction workers in construction scenarios, the hierarchical temporal loss function calculates the loss value of the worker's safety clothing. Assuming that at a certain moment, slight wear occurs in the critical area of ​​the safety clothing's elbow, causing a difference between the real-time 3D features and the standard 3D features, the calculated loss value will be significantly increased due to the larger weight of the critical area. Based on the output loss value, the changes in the safety clothing's state can be understood in a timely manner, determining whether there are any safety hazards and thus adjusting risk management strategies. For example, in the medical field, when evaluating the protective clothing worn by medical staff, a hierarchical temporal loss function is used to assess the condition of the protective clothing. If the collar of the protective clothing is damaged, the difference between the real-time feature and the standard feature increases. Combined with the high weight of the critical area, the loss value will increase significantly. Based on the output loss value, the damage to the protective clothing can be detected in time, and medical staff can be notified to replace the protective clothing to ensure their work safety.

[0072] Optionally, in this embodiment of the invention, the method further includes: To further translate technical assessments into practical business actions and build a complete business loop from early warning to handling, the risk warning system acquires multiple preset risk level thresholds. These thresholds are pre-set based on different degrees of damage and potential risks to the safety equipment. For example, a level one threshold corresponds to minor wear, while a level two threshold corresponds to more serious damage such as tearing or detachment of reflective strips. The risk warning system compares the loss value with these preset risk level thresholds to determine at least one specified level threshold. The threshold with the largest value among these specified level thresholds is then selected as the target level threshold. Each target level threshold is associated with a corresponding risk warning strategy. For example, when the target level threshold is Level 1 (minor wear), the associated risk warning strategy might be to push a reminder to the worker's terminal, prompting the worker to pay attention to protection. When the target level threshold is Level 2 (tears, reflective strips falling off, etc.), the associated risk warning strategy might be to simultaneously push a reminder to the safety officer and mark the worker's real-time location, initiating an on-site verification process to ensure that the risk can be handled in a timely manner. This allows for targeted warning and handling measures to be taken based on the actual degree of damage to the safety equipment, improving the efficiency and accuracy of risk response.

[0073] Simultaneously, to achieve full lifecycle management of safety protection equipment, the risk warning system records the generated real-time 3D feature point set and loss values ​​of the equipment. Periodically, it uses the recorded set of all real-time 3D feature points to generate feature point change trajectories. These trajectories visually demonstrate the state changes of each feature point on the safety protection equipment at different times, reflecting the wear, deformation, and other development trends of the equipment. Similarly, it periodically uses all recorded loss values ​​to generate loss value curves. These curves clearly show the overall state of the safety protection equipment over time, helping to understand the rate and extent of equipment degradation. Based on the feature point change trajectories and loss value curves, the risk warning system generates and outputs equipment assessment information for the safety protection equipment. This assessment information may include the equipment's remaining service life, replacement recommendations, etc., providing data support for optimizing replacement cycles and evaluating supplier quality.

[0074] For example, in scenarios where insurance policies like construction insurance assess the safety equipment worn by construction workers, a notification is sent to the worker's mobile phone when the safety suit shows slight wear and the loss value exceeds the first-level threshold. Upon receiving the notification, the worker will pay closer attention to the use and protection of their safety suit. If the safety suit is torn and the loss value exceeds the second-level threshold, a notification is sent not only to the safety officer but also to the worker's real-time location. The safety officer can quickly locate the worker, initiate on-site verification procedures, and address the risk promptly. By recording the real-time 3D feature point set and loss value of the safety suit over a long period, the generated feature point change trajectory and loss value curve show that the elbow feature points of the safety suit wear relatively quickly. Based on the equipment assessment information, insurance companies can recommend that construction companies shorten the replacement cycle of the elbow protection components of the safety suit, while also evaluating the product quality of the safety suit supplier to provide a reference for future cooperation.

[0075] For example, in the medical field, when evaluating protective clothing worn by healthcare workers, if a minor abnormality occurs at the collar seal of the protective clothing, exceeding the first-level threshold, a notification is sent to the smart device worn by the healthcare worker, reminding them to pay attention to the condition of the protective clothing. If the protective clothing is severely damaged, exceeding the second-level threshold, a notification is simultaneously sent to the hospital's safety management department, marking the location of the healthcare worker. The safety management department then promptly arranges for personnel to verify and handle the situation. By recording the real-time three-dimensional feature point set and loss value of the protective clothing over a long period, the generated feature point change trajectory and loss value curve can help hospitals understand the lifespan and performance changes of the protective clothing. Based on the equipment evaluation information, hospitals can rationally plan the procurement and replacement of protective clothing, ensuring the safety of healthcare workers and achieving a complete closed loop from technical evaluation to operational actions, providing comprehensive and effective support for the management of safety protective equipment.

[0076] In summary, in practical applications, such as Figure 7As shown, the technical solution of this invention is based on the need to construct a real-time and accurate safety clothing status assessment system to address the problems of low efficiency, delayed early warning, and ambiguous standards in traditional manual inspections. The technical solution of this invention revolves around dynamic feature point sampling. In terms of dynamic feature point sampling, adaptive region sampling is adopted, focusing on key protective areas and extracting high-contrast feature points at multiple scales. It also features a dynamic filtering mechanism that retains points that are stably identified for three consecutive frames to ensure matching reliability. Simultaneously, three-dimensional perception enhancement is performed. Through depth / normal maps associated with posture, combined with human posture and fabric properties, depth deviation is corrected. Furthermore, three-dimensional feature fusion is performed, fusing two-dimensional texture and three-dimensional spatial information to distinguish between wrinkles and tears. A layered temporal loss function is used, giving higher weight to damage in key areas and incorporating natural aging factors. At the same time, temporal smoothing constraints correct instantaneous feature jumps to ensure the continuity of dynamic assessment, ultimately achieving real-time early warning and closed-loop management.

[0077] The method provided in this invention, targeting the financial insurance and medical fields, avoids misjudgments caused by lighting, stains, or normal deformation by dynamically collecting effective feature points, ensuring stable capture of safety protection equipment features. Furthermore, combined with three-dimensional feature fusion processing of feature points, it overcomes the limitations of two-dimensional vision, accurately distinguishing wrinkles from actual damage to safety protection equipment, improving the reliability of assessments. This allows insurance companies providing engineering insurance and other types of insurance to promptly grasp construction safety dynamics, and enables medical institutions to promptly detect whether protective clothing has been damaged, providing strong protection for the safety of construction workers or medical personnel.

[0078] Furthermore, as Figure 1 In a specific implementation of the method, this embodiment of the invention provides a safety protection equipment status assessment device, such as... Figure 8 As shown, the device includes: a feature point extraction module 801, a three-dimensional feature fusion module 802, and an evaluation module 803.

[0079] The feature point extraction module 801 is used to acquire the video stream of the security protection equipment to be evaluated in real time, extract feature points of the security protection equipment in the video stream, and generate a set of effective feature points. The three-dimensional feature fusion module 802 is used to perform posture recognition on the target object wearing the safety protection equipment in the video stream to obtain object posture data, and to perform three-dimensional feature fusion processing on the effective feature point set in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. The evaluation module 803 is used to determine the preset equipment standard three-dimensional features, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the equipment standard three-dimensional features, and obtain and output the status evaluation result of the security protection equipment in the video stream.

[0080] In a specific application scenario, the feature point extraction module 801 is used to locate critical protection areas and non-critical protection areas on the security protection equipment in the video stream according to a preset security protection equipment design standard; to extract feature points from the critical protection areas using a preset multi-scale feature extraction algorithm to obtain multiple first effective feature points, and to extract feature points from the non-critical protection areas to obtain multiple second effective feature points; and to integrate the multiple first effective feature points and the multiple second effective feature points to obtain the effective feature point set.

[0081] In a specific application scenario, the feature point extraction module 801 is used to extract feature points from the key protection area using the multi-scale feature extraction algorithm to obtain multiple original feature points, and to filter multiple candidate feature points from the multiple original feature points; and to count the number of consecutive frames corresponding to each candidate feature point in the video stream, so as to extract the multiple first effective feature points from the multiple candidate feature points, wherein the number of consecutive frames is used to indicate the number of image frames in which the corresponding candidate feature point is continuously identified in the video stream, and the number of consecutive frames corresponding to the multiple first effective feature points is greater than a preset number threshold.

[0082] In a specific application scenario, the feature point extraction module 801 is used to calculate the local contrast of the local region where each of the original feature points is located in the video stream, and to select the original feature points whose local contrast satisfies a preset contrast condition as the multiple candidate feature points. The preset contrast condition is used to indicate the extraction of a specified number of feature points with the highest local contrast or to indicate the extraction of feature points with a local contrast greater than a contrast threshold. And / or, it calculates the texture descriptor of the local region where each of the original feature points is located in the video stream, and based on the texture descriptor corresponding to each original feature point, determines temporary texture feature points generated by the texture of the local region where each of the original feature points is located, and filters out the temporary texture feature points from the multiple original feature points. The remaining original feature points after filtering are then used as the multiple candidate feature points. The texture descriptor is used to indicate the texture rules and structure of the local region.

[0083] In a specific application scenario, the 3D feature fusion module 802 is used to acquire a pre-trained attitude estimation model, use the attitude estimation model to perform attitude recognition on the target object, and obtain the object attitude data, which includes an attitude depth map or an attitude normal map; acquire the attitude estimation confidence score output by the attitude estimation model for the object attitude data, and determine the feature fusion weights matching the attitude estimation confidence score; extract multiple effective feature points included in the effective feature point set, and determine the corresponding depth information and noise information for each effective feature point based on the object attitude data, wherein the multiple effective feature points include multiple first effective feature points and multiple second effective feature points; combine the feature fusion weights to calculate the 2D texture features, corresponding depth information, and noise information of each effective feature point, so as to perform 3D feature fusion processing on each effective feature point, and obtain multiple real-time 3D feature points of the equipment corresponding to the multiple effective feature points; and generate a set of real-time 3D feature points of the equipment including the multiple real-time 3D feature points of the equipment.

[0084] In a specific application scenario, the evaluation module 803 is used to determine a preset evaluation weight, which indicates the area weights corresponding to the critical protection area and non-critical protection area on the safety protection equipment; by using a preset hierarchical time-series loss function, combined with the evaluation weight, the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment is quantitatively calculated to obtain a loss value, and the loss value is used as the state evaluation result and output.

[0085] In specific application scenarios, the device further includes: An execution module is configured to acquire multiple preset risk level thresholds, compare the loss value with the multiple preset risk level thresholds to determine at least one specified level threshold among the multiple risk level thresholds, wherein the loss value is greater than each of the specified level thresholds, determine the specified level threshold with the largest value among the at least one specified level threshold as a target level threshold, query the risk warning strategy associated with the target level threshold, and execute the risk warning strategy; and / or, The recording module is used to record the generated real-time three-dimensional feature point set of the equipment and the loss value, periodically use the recorded real-time three-dimensional feature point set of the equipment to generate feature point change trajectories, and periodically use the recorded loss values ​​to generate loss value curves. Based on the feature point change trajectories and the loss value curves, the module generates and outputs equipment evaluation information for the safety protection equipment.

[0086] The device provided in this invention, targeting the financial insurance and medical fields, avoids misjudgments caused by lighting, stains, or normal deformation by dynamically collecting effective feature points, ensuring stable capture of safety protection equipment features. Furthermore, by combining three-dimensional feature fusion processing of the feature points, it overcomes the limitations of two-dimensional vision, accurately distinguishing wrinkles from actual damage to safety protection equipment, improving the reliability of assessments. This allows insurance companies providing engineering insurance and other types of insurance to promptly grasp construction safety dynamics, and enables medical institutions to promptly detect whether protective clothing has been damaged, providing strong protection for the safety of construction workers or medical personnel.

[0087] Specific limitations regarding the safety protection equipment status assessment device can be found in the limitations of the safety protection equipment status assessment method described above, and will not be repeated here. Each module in the aforementioned safety protection equipment status assessment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0088] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, the electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a security equipment status assessment method on the server side.

[0089] In one embodiment, an electronic device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 10 As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a security equipment status assessment method on the client side.

[0090] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The system acquires video streams of the safety protection equipment to be assessed in real time, extracts feature points from the safety protection equipment in the video stream, and generates a set of effective feature points. The target object wearing the safety protection equipment in the video stream is subjected to posture recognition to obtain object posture data, and the effective feature point set is subjected to three-dimensional feature fusion processing in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. Determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

[0091] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The system acquires video streams of the safety protection equipment to be assessed in real time, extracts feature points from the safety protection equipment in the video stream, and generates a set of effective feature points. The target object wearing the safety protection equipment in the video stream is subjected to posture recognition to obtain object posture data, and the effective feature point set is subjected to three-dimensional feature fusion processing in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. Determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

[0092] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties.

[0094] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0095] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0096] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for assessing the condition of safety protection equipment, characterized in that, include: The system acquires video streams of the safety protection equipment to be assessed in real time, extracts feature points from the safety protection equipment in the video stream, and generates a set of effective feature points. The target object wearing the safety protection equipment in the video stream is subjected to posture recognition to obtain object posture data, and the effective feature point set is subjected to three-dimensional feature fusion processing in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. Determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

2. The method according to claim 1, characterized in that, The step of extracting feature points from the security equipment in the video stream to generate a set of effective feature points includes: According to the preset design standards for security protection equipment, key protection areas and non-key protection areas are located on the security protection equipment in the video stream. A preset multi-scale feature extraction algorithm is used to extract feature points in the critical protection area to obtain multiple first effective feature points, and feature points are extracted in the non-critical protection area to obtain multiple second effective feature points. The plurality of first effective feature points and the plurality of second effective feature points are integrated to obtain the set of effective feature points.

3. The method according to claim 2, characterized in that, The method employs a preset multi-scale feature extraction algorithm to extract feature points from the critical protection area, obtaining multiple first effective feature points, including: The multi-scale feature extraction algorithm is used to extract feature points in the key protection area to obtain multiple original feature points, and multiple candidate feature points are selected from the multiple original feature points. The number of consecutive frames corresponding to each candidate feature point in the video stream is counted to extract the plurality of first effective feature points from the plurality of candidate feature points. The number of consecutive frames is used to indicate the number of image frames in which the corresponding candidate feature point is continuously identified in the video stream, and the number of consecutive frames corresponding to the plurality of first effective feature points is greater than a preset number threshold.

4. The method according to claim 3, characterized in that, The step of filtering multiple candidate feature points from the plurality of original feature points includes: Calculate the local contrast of the local region where each original feature point is located in the video stream, and select the original feature points whose local contrast satisfies a preset contrast condition as the candidate feature points. The preset contrast condition is used to indicate the extraction of a specified number of feature points with the highest local contrast or to indicate the extraction of feature points with a local contrast greater than a contrast threshold; and / or, Calculate the texture descriptor of the local region where each original feature point is located in the video stream, and based on the texture descriptor corresponding to each original feature point, determine temporary texture feature points generated by the texture of the local region where the original feature point is located among the plurality of original feature points, filter out the temporary texture feature points among the plurality of original feature points, and take the remaining original feature points after filtering as the plurality of candidate feature points, wherein the texture descriptor is used to indicate the texture rules and structure of the local region.

5. The method according to claim 1, characterized in that, The process of performing pose recognition on the target object wearing the safety protection equipment in the video stream to obtain object pose data, and combining the object pose data to perform three-dimensional feature fusion processing on the effective feature point set to obtain a real-time three-dimensional feature point set of the equipment, includes: A pre-trained pose estimation model is obtained, and the pose estimation model is used to perform pose recognition on the target object to obtain the object pose data, which includes a pose depth map or a pose normal map. Obtain the attitude estimation confidence score output by the attitude estimation model for the attitude data of the object, and determine the feature fusion weights that match the attitude estimation confidence score; Extract multiple valid feature points from the set of valid feature points, and determine the corresponding depth information and noise information for each valid feature point based on the object pose data. The multiple valid feature points include multiple first valid feature points and multiple second valid feature points. Combining the feature fusion weights, the two-dimensional texture features, corresponding depth information, and noise information of each effective feature point are calculated to perform three-dimensional feature fusion processing on each effective feature point, thereby obtaining multiple real-time three-dimensional feature points of the equipment corresponding to the multiple effective feature points. Generate the set of real-time three-dimensional feature points of the equipment, which includes the plurality of real-time three-dimensional feature points of the equipment.

6. The method according to claim 1, characterized in that, The process of determining preset standard three-dimensional features of the equipment, evaluating the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, obtaining and outputting the status evaluation result of the security protection equipment in the video stream includes: Determine preset evaluation weights, wherein the evaluation weights indicate the area weights corresponding to critical protection areas and non-critical protection areas on the safety protection equipment; By using a preset hierarchical temporal loss function and combining the evaluation weights, the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment is quantitatively calculated to obtain a loss value, and the loss value is used as the state evaluation result and output.

7. The method according to claim 6, characterized in that, The method further includes: Obtain multiple preset risk level thresholds, compare the loss value with the multiple preset risk level thresholds to determine at least one specified level threshold among the multiple risk level thresholds, wherein the loss value is greater than each of the specified level thresholds, determine the specified level threshold with the largest value among the at least one specified level threshold as the target level threshold, query the risk warning strategy associated with the target level threshold, and execute the risk warning strategy; and / or, The generated set of real-time 3D feature points of the equipment and the loss value are recorded. The recorded set of all real-time 3D feature points of the equipment is periodically used to generate feature point change trajectories, and the recorded loss values ​​are periodically used to generate loss value curves. Based on the feature point change trajectories and the loss value curves, equipment evaluation information is generated and output for the safety protection equipment.

8. A safety protection equipment status assessment device, characterized in that, include: The feature point extraction module is used to acquire video streams of security protection equipment status assessment in real time, extract feature points of the security protection equipment in the video stream, and generate a set of effective feature points. The three-dimensional feature fusion module is used to perform posture recognition on the target object wearing the safety protection equipment in the video stream to obtain object posture data, and to perform three-dimensional feature fusion processing on the effective feature point set in combination with the object posture data to obtain the real-time three-dimensional feature point set of the equipment. The evaluation module is used to determine the preset standard three-dimensional features of the equipment, evaluate the difference between the real-time three-dimensional feature point set of the equipment and the standard three-dimensional features of the equipment, and obtain and output the status evaluation result of the security protection equipment in the video stream.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images