A method and system for precise association of a construction worker's personal protective equipment with the worker

Abstract

The present application relates to a kind of construction personnel personal protective equipment and personnel accurate association method and system, method includes: obtaining the image that has defined multiple personnel target frame and protective equipment target frame;For each protective equipment target frame, calculate its with each personnel target frame The ratio of intersection, and filter candidate associated personnel based on this;Obtain the human key point position corresponding to each personnel target frame;Calculate the minimum distance from the center point of protective equipment target frame to each candidate personnel relevant key point;The ratio of intersection and key point distance is mapped to percentile scale and fused to generate association quality score;Finally, based on association quality score, and follow single personnel standard wearing quantity constraint Global optimization distribution is carried out, the accurate association of personal protective equipment and personnel is realized.The present application can realize the accurate determination of personal protective equipment attribution in complex and dense scene, and provides core technical support for intelligent and personalized safety compliance inspection.

Description

Technical Field

[0001] This invention relates to the fields of intelligent construction and computer vision technology, and in particular to a method and system for accurately associating personal protective equipment of construction workers with personnel, which is used to improve the intelligence and precision of safety management in high-risk operation scenarios such as building construction. Background Technology

[0002] In high-risk industries such as construction and factory operations, the mandatory and correct use of personal protective equipment (PPE) is the last line of defense for ensuring personnel safety and preventing accidents. Traditional supervision methods relying on manual inspections by safety officers suffer from inefficiency, narrow coverage, and the inability to achieve continuous 24 / 7 monitoring. In recent years, automated construction site safety monitoring technology based on computer vision has developed rapidly, and automatically identifying PPE wearing status by analyzing surveillance video streams has become a hot topic in research and application.

[0003] Current automated personal protective equipment (PPE) compliance inspection methods typically focus on PPE identification, neglecting the differentiated needs arising from variations in job types and scenarios in real-world applications. Accurately associating detected PPE (such as helmets, safety belts, and protective gloves) with specific workers in the scenario is a crucial prerequisite for achieving individualized and refined compliance assessments, and also a major bottleneck currently facing the technology. Existing methods suffer from the following shortcomings: First, the association accuracy is limited by simple spatial measurements. Most existing solutions use the intersection-union ratio (IUGR) as the basis for spatial association between personnel and PPE. However, in real-world, complex work scenarios, personnel are densely packed, mutually obstructing each other, and their work postures are diverse, with significant differences in the size of different PPEs. The IUGR calculation is significantly affected by the size of the personnel detection frame. For example, a small glove close to multiple workers may have very low and similar IUGR values ​​with multiple worker detection frames, making it impossible to reliably determine its affiliation and easily leading to erroneous associations. Second, they ignore real-world physical constraints. Existing methods often fail to consider basic physical wearing logic, such as "a worker normally wears at most one safety helmet and two gloves," when performing associations. This can lead to algorithms outputting association results that defy common sense, reducing system credibility. Finally, there is a lack of data foundation for refined evaluation and optimization. Most publicly available personal protective equipment datasets only label object categories and locations, lacking fine-grained labeling of "equipment-wearer" ownership relationships. This makes it impossible to accurately quantify and compare the performance of association algorithms, hindering the development of this technology towards higher accuracy and greater robustness.

[0004] Therefore, there is an urgent need for a core method that can adapt to complex, crowded, and ever-changing real-world working environments and achieve a high-precision and robust correlation between personal protective equipment and workers. Summary of the Invention

[0005] This invention provides a method and system for accurately associating personal protective equipment (PPE) with construction workers. By introducing a new metric for cross-sectional ratio, integrating key human body information, combining physical constraints, and constructing an image dataset, it solves the problems of inaccurate and unreliable association in complex scenarios, laying a core technical foundation for subsequent automated and personalized safety compliance inspections.

[0006] The technical solution adopted by the present invention to solve the above problems is: A method for accurately linking personal protective equipment (PPE) to construction workers includes the following steps: Step S1: Acquire an image containing personnel and personal protective equipment, and acquire multiple predefined personnel target boxes and multiple protective equipment target boxes in the image; Step S2: For each of the protective equipment target frames, calculate its intersection ratio with each of the personnel target frames, and based on the intersection ratio, filter out candidate associated personnel target frames for the protective equipment target frame. The protective equipment target frame and each candidate associated personnel target frame form a candidate association pair; wherein, the intersection ratio is the ratio of the intersection area of ​​the protective equipment target frame and the personnel target frame to the area of ​​the protective equipment target frame itself. Step S3: Obtain a set of key point positions of the human body corresponding to the target box of each person according to the human pose estimation algorithm; Step S4: For each protective equipment target box and each candidate associated person target box, calculate the minimum Euclidean distance from the center point of the protective equipment target box to a set of related human key points corresponding to the person target box; Step S5: For each candidate association pair, map the intersection ratio and the minimum Euclidean distance to the same comparison scale to obtain the intersection ratio mapping value and the minimum Euclidean distance mapping value, and generate the association quality score of the candidate association pair based on the intersection ratio mapping value and the minimum Euclidean distance mapping value. Step S6: Based on the association quality scores of all candidate association pairs and combined with the single-person standard wearing quantity constraint of each type of personal protective equipment, perform global optimization allocation to uniquely associate each of the protective equipment target boxes with one of the personnel target boxes.

[0007] In step S2 of this invention, candidate associated personnel target boxes are selected for the protective equipment target box through the following steps: (1) Set a cross-sectional ratio threshold for each type of personal protective equipment; (2) For a target bounding box of a protective device, calculate the intersection ratio between it and all target bounding boxes of personnel in the image; (3) The personnel target boxes with a cross-sectional ratio greater than or equal to the threshold corresponding to the personal protective equipment are determined as candidate associated personnel target boxes of the protective equipment target box, and the protective equipment target box and each candidate associated personnel target box form a candidate association pair.

[0008] Step S4 of this invention specifically includes: For a target bounding box of protective equipment, calculate the coordinates of its center point, and for a corresponding candidate associated person target bounding box, obtain the subset of relevant human body key points for that person corresponding to that type of protective equipment. K t ; Calculate the subset of critical points on the human body from the center point of the protective equipment. K t Calculate the Euclidean distance of each key point in the equation, and take the minimum value as the minimum Euclidean distance.

[0009] In step S5 of this invention, a percentile mapping method is used to map the intersection ratio and the minimum Euclidean distance to the same comparison scale: For the crossover ratio, based on the pre-constructed empirical distribution of the crossover ratio of this type of personal protective equipment, the percentile of the current crossover ratio value in the distribution is calculated and used as the crossover ratio mapping value; For the minimum Euclidean distance, based on the pre-constructed empirical distribution of key point distances for this type of personal protective equipment, the percentile of the current minimum Euclidean distance value in the distribution is calculated and used as the minimum Euclidean distance mapping value.

[0010] The empirical distribution of cross-sectional proportions and empirical distribution of key point distances of various personal protective equipment described in this invention are obtained by statistical analysis of image datasets containing ownership labels of personnel and personal protective equipment.

[0011] The associated quality score described in this invention employs the following strategy: , in, q To correlate quality scores, p dist This is the minimum Euclidean distance mapping value. p IoP Let Ø be the intersection ratio mapping value, where Ø represents the empty set.

[0012] The constraint on the standard number of personal protective equipment worn by a single person in step S6 of this invention specifically refers to: during the association allocation process, ensuring that the number of target frames of the same type of protective equipment allocated to the same person's target frame does not exceed the standard wearing limit; the standard wearing limit is set based on the common sense of physical wearing of personal protective equipment.

[0013] In step S6 of this invention, a greedy algorithm is used to perform the global optimization allocation, including the following steps: (1) Sort all candidate association pairs in descending order of their association quality scores; (2) Traverse the sorted list of candidate association pairs in order; (3) For the candidate association pair being traversed, if the protective equipment target box has not yet been assigned, and the number of personal protective equipment of this type currently assigned to the personnel target box in the candidate association pair does not reach the standard quantity constraint, then assign the personal protective equipment to the personnel and update the status; otherwise, remove the candidate association pair. (4) Repeat the previous step until the traversal is complete.

[0014] The image dataset described in this invention is constructed through the following steps: (1) Acquire images containing personnel and various personal protective equipment; (2) The images are labeled, and a unique personnel identifier is assigned to each detected person, and an owner identifier is assigned to each detected personal protective equipment; wherein, if a personal protective equipment is worn by a person, its owner identifier is the same as the personnel identifier of that person; if it is not worn by any person, a specific unassociated identifier is assigned. (3) Divide the labeled images into training set, validation set and test set to form the image dataset.

[0015] A system for precisely linking personal protective equipment (PPE) with personnel, used to achieve the aforementioned method for precisely linking construction workers' PPE with personnel, includes: The input module is used to acquire an image containing personnel and personal protective equipment, as well as multiple predefined personnel bounding boxes and protective equipment bounding boxes. The candidate filtering module is used to calculate the intersection ratio between the target box of each protective device and the target box of each person, and to filter candidate related personnel target boxes based on the intersection ratio; The key point acquisition module is used to acquire the positions of key points on the human body corresponding to the target box of each person; The distance calculation module is used to calculate the minimum Euclidean distance from the center point of each protective device target frame to the relevant human key points of each candidate associated person target frame; The quality scoring module is used to map the intersection ratio and the minimum Euclidean distance to the same comparison scale and fuse them to generate an association quality score for each candidate association pair. The global allocation module is used to perform global optimization allocation based on the quality score and the standard number of personal protective equipment worn by a single person for each type of personal protective equipment, and output the associated results.

[0016] Compared with the prior art, the present invention has the following advantages and effects: This invention effectively overcomes the scale sensitivity of traditional cross-union ratios by proposing the intersection ratio (IU), a geometric metric insensitive to the size of personal protective equipment (PPE). By fusing precise positioning information provided by key human body points and utilizing percentile mapping to fairly compare and fuse features of different dimensions, a high-discrimination quality score is generated. Furthermore, global optimization allocation is performed in conjunction with physical wearing constraints, ensuring that the association results are both accurate and consistent with real-world logic, thereby achieving accurate determination of PPE ownership in complex and intensive work scenarios. The accompanying image dataset construction method provides an indispensable data foundation for model training, key parameter statistics, and performance evaluation, forming a closed-loop technology.

[0017] The precise "personnel-equipment" correlation achieved by this invention is a crucial link in the intelligent and refined personal protective equipment (PPE) compliance inspection process. Its direct value lies in generating a unique "actual equipment list" for each on-site worker. This allows safety management to move from a broad, group-based inspection of "whether there are personnel not wearing safety helmets in the area" to a refined, personalized inspection of "whether steelworker Zhang San is wearing a safety helmet, safety belt, protective gloves, and safety shoes as required on the foundation pit working surface." By automatically comparing this precise correlation result with the standard wearing list, real-time and accurate compliance judgments and early warnings can be achieved, significantly improving the inherent safety level of the work site. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall steps in an embodiment of the present invention; Figure 2 This is a detailed flowchart illustrating the steps of an embodiment of the present invention; Figure 3 The following are histograms comparing the cross-proportion and cross-union ratios of different categories of personal protective equipment according to embodiments of the present invention; wherein, a is a histogram comparing the cross-proportion and cross-union ratios of protective gloves, b is a histogram comparing the cross-proportion and cross-union ratios of safety helmets, c is a histogram comparing the cross-proportion and cross-union ratios of protective masks, d is a histogram comparing the cross-proportion and cross-union ratios of safety belts, e is a histogram comparing the cross-proportion and cross-union ratios of protective glasses, f is a histogram comparing the cross-proportion and cross-union ratios of work shoes, g is a histogram comparing the cross-proportion and cross-union ratios of reflective vests, and h is a histogram comparing the cross-proportion and cross-union ratios of welding protective masks. Figure 4These are histograms showing the distribution of the minimum distance from the center point of different categories of personal protective equipment to relevant key points of the human body in embodiments of the present invention; wherein, a is the histogram of the distribution of the minimum distance from the center point of the protective glove to relevant key points of the human body, b is the histogram of the distribution of the minimum distance from the center point of the safety helmet to relevant key points of the human body, c is the histogram of the distribution of the minimum distance from the center point of the protective mask to relevant key points of the human body, d is the histogram of the distribution of the minimum distance from the center point of the safety belt to relevant key points of the human body, e is the histogram of the distribution of the minimum distance from the center point of the protective glasses to relevant key points of the human body, f is the histogram of the distribution of the minimum distance from the center point of the safety shoes to relevant key points of the human body, g is the histogram of the distribution of the minimum distance from the center point of the reflective vest to relevant key points of the human body, and h is the histogram of the distribution of the minimum distance from the center point of the welding protective mask to relevant key points of the human body. Figure 5 This is a structural block diagram of a system for precisely linking personal protective equipment and personnel, according to an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] The implementation principle of the method for accurately associating personal protective equipment (PPE) with personnel provided in this invention is as follows: First, preliminary screening is performed using the cross-sectional area ratio. This metric, by using the area of ​​the PPE itself as a normalization benchmark, effectively eliminates the influence of differences in the size of the personnel detection frame on the overlap rate calculation, thus ensuring stable extraction of high-probability candidate personnel even in scenarios with dense crowds and severe occlusion. Then, key point information obtained from human pose estimation is introduced. By calculating the minimum Euclidean distance from the center of the device's target frame to the corresponding key point of the personnel, the geometric proximity relationship between the device and the personnel is evaluated at a more refined spatial scale. To address the issues of inconsistent dimensions and large distribution differences between the cross-sectional area ratio and the key point distance, a percentile mapping method is used to transform them to a unified and fairly comparable scale and then merge them to generate a comprehensive association quality score. Finally, based on this score and combined with real-world physical constraints, a global optimization algorithm completes the final allocation, thereby achieving high-precision and robust accurate association between PPE and personnel in complex scenarios with dense crowds and mutual occlusion. This effectively supports intelligent and refined supervision of PPE wearing compliance in complex scenarios.

[0021] The embodiments of the present invention specifically include the following steps: Step S1: Obtain an image containing personnel and personal protective equipment, and obtain multiple personnel target boxes and multiple protective equipment target boxes defined in the image.

[0022] In this step, the input is an image containing multiple people and their personal protective equipment (PPE), with multiple pre-defined bounding boxes for people and PPE. These bounding boxes can be obtained through reasoning on the input image using a mature object detection model, or they can be directly derived from a finely annotated image dataset. The bounding boxes are typically represented by boundary coordinates, such as the x-coordinate of the top-left corner, the y-coordinate of the top-left corner, the x-coordinate of the bottom-right corner, and the y-coordinate of the bottom-right corner.

[0023] Step S2: For each protective equipment target box, calculate its intersection ratio with each personnel target box, and based on the intersection ratio, filter out candidate associated personnel target boxes for the protective equipment target box. The protective equipment target box and each candidate associated personnel target box form a candidate association pair.

[0024] In this step, the intersection ratio is defined as the ratio of the intersection area of ​​the protective equipment target frame and the personnel target frame to the area of ​​the protective equipment target frame itself. The intersection ratio directly measures the proportion of personal protective equipment covered by the personnel target frame. Since its denominator is the area of ​​the equipment frame, it is not sensitive to changes in equipment size and provides a more stable spatial correlation measure compared to the traditional intersection-union ratio.

[0025] Specifically, as shown in the attached document Figure 3 As shown, based on statistical analysis of the image dataset, the crossover ratio of personal protective equipment (PPE) is generally highly concentrated in the range of 0.9-1.0, exhibiting a significantly right-skewed distribution, indicating that the vast majority of worn devices are almost entirely within the bounding box of the person. In contrast, the crossover ratio (CR) shows a severely left-skewed distribution, easily affected by the size of the person's bounding box. This verifies the superiority of the crossover ratio as a correlation metric.

[0026] The candidate selection process based on crossover ratio includes the following steps: First, an overlap threshold is set for each type of personal protective equipment. This threshold is derived from statistical analysis of image datasets containing explicit ownership labels. For example, the overlap threshold for safety helmets could be set to 0.67, and for protective gloves, it could be 0.20.

[0027] Then, for a protective equipment target bounding box, calculate its intersection ratio with the target bounding boxes of all personnel in the image.

[0028] Finally, the personnel target boxes with a cross-sectional area ratio greater than or equal to the corresponding threshold for that type of equipment are initially identified as candidate associated personnel target boxes for the protective equipment target box. If a certain equipment has no candidate associated personnel, it is determined to be personal protective equipment that is not being worn.

[0029] The above image dataset was constructed using the following steps: (1) Acquire images containing personnel and various personal protective equipment; (2) The images are labeled, and a unique personnel identifier is assigned to each detected person, and an owner identifier is assigned to each detected personal protective equipment; wherein, if a personal protective equipment is worn by a person, its owner identifier is the same as the personnel identifier of that person; if it is not worn by any person, a specific unassociated identifier is assigned.

[0030] (3) Divide the labeled images into training set, validation set and test set to form the image dataset.

[0031] Step S3: Obtain a set of key points of the human body corresponding to the target box of each person according to the human pose estimation algorithm.

[0032] In this step, an open-source human pose estimation algorithm is used to process the input image. This algorithm can output structured keypoint information for each human body in the image, typically including 17 keypoints: nose, eyes, ears, shoulders, elbows, wrists, hips, knees, and ankles. Each keypoint information includes its horizontal and vertical pixel coordinates and a visibility confidence score.

[0033] Step S4: For each protective equipment target frame and each candidate associated personnel target frame, calculate the minimum Euclidean distance from the center point of the protective equipment target frame to a set of related human key points corresponding to the personnel target frame.

[0034] Different types of personal protective equipment (PPE) are worn on the human body at different locations corresponding to different key points on the body. Therefore, this embodiment predefines a subset of relevant key points for each type of equipment. K t For example: the key points related to a safety helmet are the left ear and the right ear; the key points related to protective gloves are the left wrist and the right wrist; the key points related to a reflective vest are the left shoulder, the right shoulder, the left hip, and the right hip; and the key points related to safety shoes are the left ankle and the right ankle.

[0035] Specifically, for a target bounding box of protective equipment, calculate the coordinates of its center point, and for a target bounding box of a candidate associated person, obtain a subset of relevant human body key points corresponding to that person and the protective equipment. K t Calculate the subset of critical points on the human body from the center point of the protective equipment. K t The Euclidean distance of each key point is calculated, and the minimum value is taken as the minimum Euclidean distance. This distance reflects the spatial proximity between the device and the part of the body most likely to be worn by the person.

[0036] Step S5: For each candidate association pair, map the crossover ratio and minimum Euclidean distance to the same comparison scale to obtain the crossover ratio mapping value and the minimum Euclidean distance mapping value, and generate the association quality score of the candidate association pair based on the crossover ratio mapping value and the minimum Euclidean distance mapping value.

[0037] In this embodiment, it should be noted that the intersection ratio and the minimum Euclidean distance differ in their dimensions and numerical distribution, making direct comparison or simple weighting impossible. Therefore, this embodiment employs a percentile mapping method to normalize both to a unified comparison scale.

[0038] like Figure 3 and Figure 4 As shown, based on the image dataset, the empirical distribution of cross-sectional proportion and keypoint distance for each type of personal protective equipment can be pre-analyzed and constructed. For a specific cross-sectional proportion value of the current candidate pair, the percentile of the value in the empirical distribution of cross-sectional proportion of its category is calculated by querying the empirical distribution of cross-sectional proportion of its category, thus obtaining the cross-sectional proportion mapping value. p IoP Similarly, for the candidate pair, the minimum Euclidean distance value calculated is used to query the empirical distance distribution of its category to obtain the minimum Euclidean distance mapping value. p dist The percentile values ​​range from [0, 1], with higher values ​​indicating better performance of the current indicator across all possible scenarios within that category. These two mapping values ​​are then combined to generate the final association quality score. q This embodiment employs the following strategy for fusion: , In the formula, K t When the set is not empty, the association quality score uses the minimum Euclidean distance mapping value; otherwise, the intersection ratio mapping value is used. This quality score integrates coarse-grained bounding box overlap information and fine-grained human pose space information. The higher the score, the better the matching quality of the "person-equipment" candidate pair.

[0039] Step S6: Based on the association quality scores of all candidate association pairs and combined with the single-person standard wearing quantity constraint of each type of personal protective equipment, perform global optimization allocation to uniquely associate each of the protective equipment target boxes with one of the personnel target boxes.

[0040] To ensure that the allocation results conform to real-world physical logic, this embodiment introduces a standard constraint on the number of items a single person can wear. This constraint is set based on common sense about the physical wearing of personal protective equipment, for example: a person can wear a maximum of 1 safety helmet, 1 reflective vest, 2 protective gloves, and 2 safety shoes.

[0041] This embodiment uses a greedy algorithm to perform global optimization allocation, including the following steps: First, the candidate association pairs are categorized by personal protective equipment type and sorted from high to low according to their association quality score q.

[0042] Then, iterate through the sorted list of candidate association pairs in order.

[0043] Next, for the currently traversed candidate association pairs, if the protective equipment target box has not yet been assigned, and the number of personal protective equipment of this type currently assigned to the personnel target box in the candidate association pair does not reach the standard quantity constraint, then the personal protective equipment is assigned to the personnel and the status is updated; otherwise, the candidate association pair is removed.

[0044] Taking a candidate association pair containing a helmet target frame and a personnel target frame as an example, before assignment confirmation, it is first checked whether the helmet target frame has already been identified and assigned. If it has, the candidate association pair is deleted. Next, the personnel target frame is checked. For the helmet category, it is checked whether the personnel target frame has been assigned the standard number of helmets required. If the standard number has been exceeded, the candidate association pair is deleted. If not, the personal protective equipment is assigned to the person.

[0045] Finally, repeat the above steps according to the category of personal protective equipment until the process is complete.

[0046] This allocation mechanism, under the dual constraints of "one device to one person" and "each person and each type of device not exceeding the upper limit", prioritizes the association with the highest quality score, thereby outputting the globally optimal association result.

[0047] Reference Figure 5 As shown, this embodiment of the invention also provides a system for accurately associating personal protective equipment with personnel. The system includes: an input module 101, a candidate screening module 102, a key point acquisition module 103, a distance calculation module 104, a quality scoring module 105, and a global allocation module 106.

[0048] Input module 101 is used to acquire an image containing personnel and personal protective equipment, as well as multiple predefined personnel target boxes and protective equipment target boxes therein; The candidate filtering module 102 is used to calculate the intersection ratio between the target box of each protective device and the target box of each person, and to filter the candidate related personnel target boxes based on the intersection ratio; The key point acquisition module 103 is used to acquire the positions of human body key points corresponding to each person's target box; The distance calculation module 104 is used to calculate the minimum Euclidean distance from the center point of each protective device target frame to the relevant human key points of each candidate associated person target frame; The quality scoring module 105 is used to map the intersection ratio and minimum Euclidean distance to the same comparison scale and fuse them to generate an association quality score for each candidate association pair. The global allocation module 106 is used to perform global optimization allocation based on the associated quality score and the single-person standard wearing quantity constraint of each type of personal protective equipment, and output the associated results.

[0049] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for accurately linking personal protective equipment (PPE) to construction workers, characterized in that, Includes the following steps: Step S1: Acquire an image containing personnel and personal protective equipment, and acquire multiple predefined personnel target boxes and multiple protective equipment target boxes in the image; Step S2: For each of the protective equipment target frames, set the corresponding intersection ratio threshold according to the personal protective equipment category to which the protective equipment target frame belongs, calculate the intersection ratio between the protective equipment target frame and each of the personnel target frames, and determine the personnel target frames with an intersection ratio greater than or equal to the intersection ratio threshold corresponding to the category as candidate associated personnel target frames of the protective equipment target frame, and form a candidate association pair with each of the protective equipment target frames and each candidate associated personnel target frame, wherein the intersection ratio is the ratio of the intersection area of ​​the protective equipment target frame and the personnel target frame to the area of ​​the protective equipment target frame itself; Step S3: Obtain a set of key point positions of the human body corresponding to the target box of each person according to the human pose estimation algorithm; Step S4: For each protective equipment target box and each candidate associated person target box, calculate the minimum Euclidean distance from the center point of the protective equipment target box to a set of related human key points corresponding to the person target box; Step S5: For each candidate association pair, use the percentile mapping method to map the intersection ratio and the minimum Euclidean distance to the same comparison scale: For the crossover ratio, based on the pre-constructed empirical distribution of crossover ratios for personal protective equipment, the percentile of the current crossover ratio value in this distribution is calculated and used as the crossover ratio mapping value p. IoP When the relevant human body key point subset K t When not empty, for the minimum Euclidean distance, based on the pre-constructed empirical distribution of key point distances for this type of personal protective equipment, the percentile of the current minimum Euclidean distance value in the distribution is calculated and used as the minimum Euclidean distance mapping value p. dist The association quality score q of the candidate association pair is determined according to the following strategy: when the relevant human keypoint subset K t When not empty, the minimum Euclidean distance mapping value p is used. dist As the associated quality score q, when the relevant subset K of human key points t When empty, the cross-ratio mapping value p is used. IoP As the associated quality score q; Step S6: Based on the association quality score q of all candidate association pairs, and combined with the single-person standard wearing quantity constraint of each type of personal protective equipment, a greedy algorithm is used to perform global optimization allocation according to the personal protective equipment category. The candidate association pairs corresponding to each type of personal protective equipment are sorted from high to low according to their association quality score q, and each protective equipment target box is uniquely associated with a person target box.

2. The method for accurately associating personal protective equipment (PPE) with personnel according to claim 1, characterized in that, Step S4 specifically involves: For a target bounding box of protective equipment, calculate the coordinates of its center point, and for a corresponding candidate associated person target bounding box, obtain the subset K of relevant human key points of the protective equipment for that person. t Calculate the subset K of key points on the human body from the center point of the protective equipment. t Calculate the Euclidean distance of each key point in the equation, and take the minimum value as the minimum Euclidean distance.

3. The method for accurately associating personal protective equipment (PPE) with personnel according to claim 1, characterized in that, The pre-constructed empirical distributions of the cross-sectional proportions and key point distances of various personal protective equipment are obtained through statistical analysis of image datasets containing ownership labels of personnel and personal protective equipment.

4. The method for accurately associating personal protective equipment (PPE) with personnel according to claim 1, characterized in that, The constraint on the standard number of personal protective equipment worn by a single person mentioned in step S6 specifically refers to ensuring that, during the association allocation process, the number of target frames of the same type of protective equipment allocated to the same person's target frame does not exceed the standard wearing limit, which is set based on the physical wearing common sense of personal protective equipment.

5. The method for accurately associating personal protective equipment (PPE) with personnel according to claim 1, characterized in that, In step S6, the global optimization allocation is performed using a greedy algorithm, including the following steps: (1) Establish a candidate association pair list according to the category of personal protective equipment; (2) Sort the candidate association pairs corresponding to each type of personal protective equipment from high to low according to the association quality score; (3) Traverse the sorted list of candidate association pairs in order; (4) For the candidate association pair being traversed, if the protective equipment target box has not yet been assigned, and the number of similar personal protective equipment currently assigned to the candidate associated personnel target box in the candidate association pair has not reached the upper limit of the standard number of such personal protective equipment worn by a single person, then associate the protective equipment target box with the candidate associated personnel target box and update the status; otherwise, remove the candidate association pair. (5) Repeat the previous step until the traversal is complete.

6. The method for accurately associating personal protective equipment (PPE) with personnel according to claim 1, characterized in that, The image dataset is constructed through the following steps: (1) Acquire images containing personnel and various personal protective equipment; (2) The image is labeled, and a unique personnel identifier is assigned to each detected person, and an owner identifier is assigned to each detected personal protective equipment. If a personal protective equipment is worn by a person, its owner identifier is the same as the personnel identifier of that person. If it is not worn by any person, a specific unassociated identifier is assigned. (3) Divide the labeled images into training set, validation set and test set to form the image dataset.

7. A system for precisely associating personal protective equipment (PPE) with personnel, used to implement the method for precisely associating construction workers' PPE with personnel as described in any one of claims 1-6, characterized in that, include: The input module is used to acquire an image containing personnel and personal protective equipment, as well as multiple predefined personnel bounding boxes and protective equipment bounding boxes. The candidate filtering module is used to calculate the intersection ratio between each protective equipment target box and each personnel target box, and filter candidate related personnel target boxes based on the intersection ratio; The key point acquisition module is used to acquire the positions of key points on the human body corresponding to the target box of each person; The distance calculation module is used to calculate the minimum Euclidean distance from the center point of each protective device target frame to the relevant human key points of each candidate associated person target frame; The quality scoring module is used to map the intersection ratio and the minimum Euclidean distance to the same comparison scale and fuse them to generate an association quality score for each candidate association pair. The global allocation module is used to perform global optimization allocation based on the quality score and the standard number of personal protective equipment worn by a single person for each type of personal protective equipment, and output the associated results.

Images