Edge computing-based industrial field wear compliance detection method and system

Abstract

This invention provides a method and system for compliance detection of wearable devices in industrial settings based on edge computing. The method includes: determining an adaptive search region containing the wearable parts of a target person based on the pose geometry information of the target person in the current video frame; extracting a local region image from the current video frame based on the adaptive search region; performing feature inference based on the local region image to obtain a texture confidence index and a spatial location index; generating an instantaneous compliance evidence value and a feature semantic matching index for the current frame based on the texture confidence index and the spatial location index; and performing temporal lag filtering based on the instantaneous compliance evidence value and compliance status information from historical frames to obtain a preliminary compliance status for the current frame. This invention achieves low-latency response and high-accuracy detection of wearable devices in complex and highly interfering industrial work environments.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an industrial field wearable compliance testing method and system based on edge computing. Background Technology

[0002] In industrial safety supervision, the standardized wearing of personal protective equipment (PPE) such as safety helmets, anti-static caps, goggles, and safety shoes by workers is a fundamental control measure to reduce on-site operational risks, minimize casualties, and prevent production accidents. Currently, deep learning-based target detection algorithms are widely used in industrial site PPE compliance testing. The testing process involves: acquiring real-time video streams transmitted from monitoring equipment, extracting images frame by frame, inputting complete image frames into a pre-trained convolutional neural network model, and directly outputting a judgment result of compliance or non-compliance after a single feature inference.

[0003] In complex industrial settings such as chemical production workshops and electronics processing plants, images captured by cameras are often affected by various instantaneous strong interference factors, such as strong light reflections, rapidly moving shadows in the work environment, brief obstructions from personnel or equipment, lens stains, or cluttered backgrounds. However, existing detection solutions rely solely on global image feature extraction and one-way inference from the model, lacking feature verification and adaptive adjustment feedback mechanisms for complex interference scenarios. This makes it difficult for existing detection methods to simultaneously meet the dual requirements of anti-interference stability and low-latency real-time performance. They also cannot accurately distinguish between genuine violations of protective equipment wearing regulations and false abnormal signals caused by environmental interference, leading to frequent false alarms and high false negative rates. Consequently, it is difficult to achieve highly accurate, stable, and reliable protective equipment compliance detection applications in complex industrial environments. Summary of the Invention

[0004] This invention provides an edge computing-based method and system for compliance detection of wearable devices in industrial settings. It addresses the problem that existing technologies struggle to balance anti-interference capabilities and real-time response speed in complex dynamic interference scenarios, reduces false alarm and missed alarm rates caused by the inability to distinguish between genuine violations and false signals, and achieves low-latency response and high-accuracy detection of wearable devices in complex and highly interfering industrial environments.

[0005] In a first aspect, the present invention provides an industrial field wearable compliance testing method based on edge computing, comprising: Based on the pose geometry information of the target person in the current video frame, an adaptive search region containing the target's clothing parts is determined, and the current video frame is extracted based on the adaptive search region to obtain a local region image; Based on the local region image, feature reasoning is performed to obtain texture confidence index and spatial location index. Based on the texture confidence index and spatial location index, the instantaneous compliance evidence value and feature semantic matching index of the current frame are generated. Based on the instantaneous compliance evidence value and the compliance status information of historical frames, a time-series lag filtering process is performed to obtain the preliminary compliance status of the current frame. Based on the temporal stability characteristics of the preliminary compliance status within a preset time window and the mapping relationship with the feature semantic matching index, the status correction parameters are determined. The initial compliance status is corrected based on the status correction parameters to obtain the target compliance status, and a structured compliance detection result is generated based on the target compliance status.

[0006] Secondly, the present invention also provides an edge computing-based industrial field wearable compliance testing system, applied to the edge computing-based industrial field wearable compliance testing method described in the first aspect; the edge computing-based industrial field wearable compliance testing system includes: An adaptive region extraction module is used to determine an adaptive search region containing the target's clothing parts based on the pose geometry information of the target person in the current video frame, and to extract the local region image based on the adaptive search region in the current video frame. The single-frame dual-index generation module is used to perform feature reasoning based on the local region image to obtain the texture confidence index and the spatial location index, and to generate the instantaneous compliance evidence value and feature semantic matching index of the current frame based on the texture confidence index and the spatial location index. The time lag filtering module is used to perform time lag filtering based on the instantaneous compliance evidence value and the compliance status information of historical frames to obtain the preliminary compliance status of the current frame. Based on the time stability characteristics of the preliminary compliance status within a preset time window and the mapping relationship with the feature semantic matching index, the status correction parameters are determined. The semantic joint correction module is used to correct the preliminary compliance state based on the state correction parameters to obtain the target compliance state, and generate a structured compliance detection result based on the target compliance state.

[0007] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the edge computing-based industrial field wearable compliance testing method as described above.

[0008] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the edge computing-based industrial field wearable compliance testing method as described above.

[0009] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the edge computing-based industrial field wearable compliance testing method as described above.

[0010] The edge computing-based industrial field wearable compliance detection method provided in this invention determines the adaptive search region for the protective equipment wearing area based on the pose geometry information of the target worker in the current video frame, and extracts the corresponding local image. This avoids the influence of complex backgrounds and instantaneous strong interference factors on feature detection. Feature inference operations are performed on the local image to extract the texture confidence index and spatial position index of the protective equipment. These two indices are then fused to generate the instantaneous compliance evidence value and feature semantic matching core evaluation index for the current video frame, achieving dual-dimensional quantitative verification in the single-frame detection stage. Furthermore, by combining the compliance status temporal information of historical continuous video frames, temporal lag filtering is applied to the instantaneous compliance evidence value of the current frame to eliminate interference caused by single-frame instantaneous interference. The system analyzes data fluctuations and outputs a preliminary compliance status. Then, based on the temporal stability characteristics and semantic matching indicators of the preliminary compliance status within a preset fixed time window, it matches and determines the status correction parameters adapted to the current on-site interference conditions in real time, and calibrates the preliminary compliance status to obtain the final target compliance status. Finally, it automatically generates structured compliance detection results that match the on-site regulatory needs based on the target compliance status. Through the closed-loop cooperation of precise verification of single-frame features and collaborative correction of temporal status, it solves the problem that existing technologies cannot balance anti-interference capability and real-time response speed in complex dynamic interference scenarios. At the same time, it reduces the false alarm rate and false alarm rate caused by the inability to distinguish between real violations and false signals, and achieves low-latency response and high-precision detection of wearable compliance detection in complex and highly interference-prone industrial work sites. Attached Figure Description

[0011] Figure 1 This is a schematic flowchart of the industrial field wearable compliance testing method based on edge computing provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the industrial field wearable compliance testing system based on edge computing provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0012] 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 embodiments of the present invention, and not all embodiments. 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.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] See Figure 1 , Figure 1 This is a flowchart illustrating the edge computing-based industrial wearable compliance testing method provided by the present invention. In this embodiment, the execution entity of the edge computing-based industrial wearable compliance testing method is the compliance testing system. Therefore, the edge computing-based industrial wearable compliance testing method includes: Step 10: Based on the pose geometry information of the target person in the current video frame, determine the adaptive search region containing the target's clothing parts, and extract the current video frame based on the adaptive search region to obtain a local region image.

[0016] Optionally, the compliance detection system acquires real-time transmitted video frames from monitoring equipment deployed in the industrial site (such as surveillance cameras, network cameras, etc.). These current video frames are single-frame images from a continuously acquired video stream currently undergoing detection processing, specifically containing one or more target personnel within the industrial site's work area, as well as the work environment background. Subsequently, the system performs target personnel detection on the acquired current video frames, identifying all target personnel within the current video frame and extracting the posture geometry information of each target personnel. This posture geometry information is a set of geometric parameters characterizing the target personnel's body posture, limb joint positions, and the relative spatial relationships of various body parts. Specifically, it includes the contour coordinates, center point coordinates, length and width dimensions of key body parts such as the head, neck, torso, and limbs, as well as the connection relationships and angle information of each limb joint.

[0017] Simultaneously, the compliance detection system identifies the target wearing parts of each person based on the current video frame. These target wearing parts are the body parts that industrial workers must wear personal protective equipment (PPE), specifically including the head (corresponding to safety helmets and anti-static caps), eyes (corresponding to goggles), and feet (corresponding to safety shoes). After acquiring the target person's posture geometry information and determining the target wearing parts, an adaptive search area containing the target wearing parts is determined. The process is as follows: using the center point coordinates of each target wearing part as a reference, the system expands the pixel range outwards by a preset proportion according to the outline size of the target wearing part. This preset expansion proportion can be dynamically adjusted according to the type of target wearing part (e.g., the expansion proportion for the head is larger than that for the eyes to ensure that the wearing area of ​​safety helmets and anti-static caps is fully included). During the determination process, if there is slight occlusion or posture deviation of the target wearing part, the expansion proportion is automatically increased to ensure that the target wearing part falls completely within the adaptive search area.

[0018] After determining the adaptive search region, the compliance detection system extracts local region images from the current video frame. This involves cropping the image portion from the current video frame that perfectly corresponds to the adaptive search region, resulting in a local region image. Each part of the target's clothing corresponds to an independent local region image; for example, the target person's head corresponds to a local region image, the eyes to a local region image, and the feet to a local region image.

[0019] In one embodiment, assuming a chemical production workshop scenario, monitoring equipment collects real-time video streams of the work area. The compliance detection system extracts the current video frame, which contains two workers (target worker 1 and target worker 2). After detecting the target workers in the current video frame, the posture geometry information of target worker 1 is extracted, specifically including: head contour coordinates (100, 200) to (200, 300), head center point coordinates (150, 250), head contour width 100 pixels, and head length 100 pixels; eye center point coordinates (130, 230) and (170, 230), eye contour size 20 pixels × 15 pixels; foot contour coordinates (120, 500) to (150, 600) and (170, 500) to (200, 600), foot center point coordinates (135, 550) and (185, 550), foot contour width 30 pixels, and foot length 100 pixels.

[0020] Meanwhile, based on the protection requirements of the chemical production workshop, the compliance detection system identified the target wearing parts as the head (where a safety helmet is required), eyes (where goggles are required), and feet (where safety shoes are required). Based on the posture geometry information of target person 1, the adaptive search area for each target wearing part was determined: using the head center point (150, 250) as a reference, and considering the head has no obvious obstruction and a stable posture, a default expansion ratio (1.2 times) was used. After expanding outwards, the contour coordinates of the head adaptive search area were (90, 190) to (210, 310); using the eye center points (130, 230) and (170, 230) as references, and considering the small size of the eyes, the expansion ratio was set to 1.5 times. Two adaptive search regions for the eyes were obtained, with contour coordinates of (115, 217.5) to (145, 242.5) and (155, 217.5) to (185, 242.5), respectively. Based on the center point of the foot (135, 550) and (185, 550), the expansion ratio was set to 1.1 times, resulting in two adaptive search regions for the feet, with contour coordinates of (117, 495) to (153, 605) and (167, 495) to (203, 605), respectively.

[0021] Finally, the compliance detection system crops the image portions corresponding to the six adaptive search regions from the current video frame, resulting in six local region images: one for the head, two for the eyes, and two for the feet. Similarly, the process of determining the adaptive search region and extracting local region images for target person 2 is the same as that for target person 1, and will not be described again.

[0022] Step 20: Perform feature reasoning based on local region images to obtain texture confidence index and spatial location index, and generate instantaneous compliance evidence value and feature semantic matching index for the current frame based on texture confidence index and spatial location index.

[0023] Optionally, the compliance detection system performs feature inference on each local area image corresponding to each target wearing part. This involves calling a pre-trained convolutional neural network (CNN) model, inputting the local area image into the CNN model, and using the CNN model to perform parallel feature inference on the local area image. On one hand, the model extracts texture features from the local area image and outputs probability values ​​representing the possibility of the material, color, and pattern of the wearable (e.g., a helmet) through a classification head, using these as texture confidence indicators. On the other hand, the model predicts the geometric outline bounding box and centroid coordinates of the wearable through a regression head, and combines this with the head center point coordinates determined in step 10 to calculate the offset and inclusion relationship of the wearable's centroid relative to the head center, generating a spatial position indicator representing whether the wearable is located within the reasonable wearing area of ​​the human head. This spatial position indicator is a set of structured data, including basic coordinate information (such as wearable outline centroid coordinates, head center point coordinates, etc.) and a quantitative evaluation value representing the compliance of the wearable's position. The wearable outline centroid coordinates are the coordinates corresponding to the geometric centroid of the personal protective equipment outline detected in the local area image.

[0024] Continuing with the embodiment based on step 10, the compliance detection system acquires six local area images of the target person 1 (one head, two eye images, and two foot images), and performs feature inference processing on each local area image. For the head local area image, after the convolutional neural network model extracts the image texture features, it determines that a safety helmet exists in the area, generating a texture confidence index of 0.92; at the same time, by analyzing the positional distribution of the safety helmet, it determines that the safety helmet completely covers the head and is worn correctly, generating a spatial position index of 0.88, the centroid coordinates of the wearer (safety helmet) outline as (152, 248), and the coordinates of the head center point as (150, 250).

[0025] The convolutional neural network model is a lightweight neural network model trained on a large number of industrial field wearable compliance and non-compliance sample images. It can accurately extract texture features (such as material texture, color texture, edge texture and other feature information) and spatial location features (such as the geometric contour bounding box coordinates of the wearable) in local area images.

[0026] Optionally, after obtaining the texture confidence index and spatial location index, the compliance detection system performs dual verification based on the texture confidence index and spatial location index, and finally generates the instantaneous compliance evidence value and feature semantic matching index of the current frame, as described in steps 201 to 207. The feature semantic matching index represents the inherent logical consistency between texture features and spatial features, and reflects the credibility or degree of conflict of the detection result of the current frame. The instantaneous compliance evidence value represents the final judgment strength of whether the current frame complies with the wearability specification; specifically, it includes static zero evidence level, pending confirmation evidence level, or dynamic high evidence level.

[0027] Step 30: Perform time-lag filtering based on instantaneous compliance evidence value and compliance status information of historical frames to obtain the preliminary compliance status of the current frame. Based on the mapping relationship between the temporal stability features of the preliminary compliance status within a preset time window and the feature semantic matching index, determine the status correction parameters.

[0028] Optionally, the compliance detection system retrieves compliance status information from historical frames. These historical frames are a preset number of video frames preceding and consecutive to the current video frame. This preset number can be adjusted based on the detection needs and interference conditions in the industrial setting (e.g., a preset number of 10 frames, meaning retrieving the 10 video frames preceding the current frame). The compliance status information of these historical frames represents the compliance status (including compliant, non-compliant, and pending confirmation) obtained by the compliance detection system after detecting each historical frame, along with the frame's instantaneous compliance evidence value, feature semantic matching indicators, and other relevant detection data. This compliance status information is stored in the historical status cache queue built into the compliance detection system.

[0029] The compliance detection system performs time-lag filtering based on the level and type of instantaneous compliance evidence value and the compliance status information of retrieved historical frames to obtain the preliminary compliance status of the current frame, as detailed in steps 301 to 305.

[0030] After acquiring the preliminary compliance status, the compliance detection system first calculates the temporal stability characteristics of this preliminary compliance status within a preset time window. This preset time window is a set of continuous video frames, including the current frame, and its length can be adjusted according to the interference frequency and detection accuracy requirements of the industrial site (e.g., a preset time window length of 5 frames, including the current frame and the previous 4 historical frames). The temporal stability characteristics represent the degree of temporal stability of the preliminary compliance status within the preset time window, calculated based on the cumulative number of state transitions within the time window. Then, based on the mapping relationship between the temporal stability characteristics and the feature semantic matching index, state correction parameters are determined, as detailed in steps 306 to 309. These state correction parameters are a set of parameters used to correct the preliminary compliance status, eliminate the impact of instantaneous interference, and improve the accuracy of the detection results. Specifically, they include the time lag compensation step size, conflict suppression level indicator, and state lock flag.

[0031] Step 40: Correct the preliminary compliance status based on the status correction parameters to obtain the target compliance status, and generate structured compliance detection results based on the target compliance status.

[0032] Optionally, the compliance detection system corrects the initial compliance status based on the time lag compensation step, conflict suppression level identifier, and status lock flag in the status correction parameters to eliminate detection deviations caused by instantaneous environmental interference and obtain a target compliance status that can truly reflect the wear compliance status of the target personnel, as described in steps 401 to 405.

[0033] After obtaining the target compliance status, the compliance testing system generates structured compliance testing results based on the target compliance status. That is, according to the preset format and specifications, the key information in the testing process and the final testing results are sorted and summarized to obtain structured compliance testing results containing testing time, testing area, target personnel identification, protective equipment type, compliance judgment result and confidence level information, so that industrial safety production supervisors can view, analyze and archive them.

[0034] The embodiments of the present invention solve the problem that the existing technology is difficult to balance anti-interference capability and real-time response speed in complex dynamic interference scenarios. At the same time, it reduces the false alarm rate and false alarm rate caused by the inability to distinguish between real violations and false signals, and realizes low latency response and high accuracy detection of wearable compliance detection in complex and highly interference-prone industrial operations.

[0035] Optionally, steps 201 to 207 include: Step 201: Based on the relative vector direction of the centroid coordinates of the wearable outline and the center point coordinates of the head in the spatial position index, and combined with the preset human head anatomical structure parameters, determine the boundary of the spatial tolerance area, obtain the geometric projection interval that characterizes the expected distribution range of the wearable, and divide the local area image based on the geometric projection interval to obtain the semantic attention sub-region corresponding to the physiological wearing position of the wearable.

[0036] Optionally, the compliance inspection system extracts the coordinates of the wearable's outline centroid and the head center point from the spatial location indicators, and calculates the relative vector direction between the wearable's outline centroid and the head center point. This relative vector direction is the direction pointed to by the vector starting from the head center point and ending at the wearable's outline centroid, used to characterize the wearable's positional offset direction relative to the head center point. Simultaneously, it retrieves preset human head anatomical structural parameters, which are a standardized set of parameters preset based on a large amount of human head anatomical data. These parameters specifically include the length, width, and height of the human head, as well as the positional range and distance parameters of key head parts (such as the top of the head, forehead, back of the head, and both temples) relative to the head center point.

[0037] Based on relative vector direction and human head anatomical parameters, the compliance detection system uses the head center point as a reference to construct a geometrically closed region as the boundary of the spatial tolerance region. This spatial tolerance region boundary represents the maximum allowable offset of the wearable's geometric center of gravity under normal wearing conditions. Subsequently, the spatial tolerance region boundary is mapped onto the pixel coordinate system of the local area image, forming a geometric projection interval characterizing the expected distribution range of the wearable.

[0038] Finally, the compliance inspection system divides the local image based on the geometric projection interval. Using the geometric projection interval as a mask template, it logically divides the local image and extracts pixel data within that interval to form a semantically relevant sub-region. This semantically relevant sub-region is a local image area in the local image that corresponds to the physiological wearing position of the wearable device on the human head. Its range is consistent with the geometric projection interval, excluding the background area.

[0039] In one embodiment, the compliance detection system obtains spatial position indicators of a local area image of the head of the target person 1, showing that the centroid coordinates of the helmet outline are (152, 248) and the coordinates of the head center point are (150, 250). The relative vector direction is calculated: starting from the head center point (150, 250) and ending at the centroid of the clothing outline (152, 248), a slight offset to the upper right is obtained (2 pixels horizontally to the right and 2 pixels vertically upward).

[0040] The system calls the preset human head anatomy parameters, which specify that the reasonable area for wearing a safety helmet is 30 to 50 pixels above the center point of the head and 20 pixels to the left and right, i.e., 130 to 170 pixels on the horizontal axis and 200 to 220 pixels on the vertical axis.

[0041] Combining relative vector direction and anatomical parameters, the spatial tolerance zone boundary is determined to be 128-172 pixels x-coordinate and 198-222 pixels y-coordinate. This boundary extends 2 pixels outwards from the preset reasonable wearing range to accommodate slight positional offsets. Based on this boundary, a geometric projection region (a rectangular area with x-coordinates of 128-172 pixels x-coordinate and 198-222 pixels y-coordinate) is obtained. The system then uses this region to divide the head local area image (contour coordinates 90, 190 to 210, 310), resulting in a semantically relevant sub-region of 128-172 pixels x-coordinate and 198-222 pixels y-coordinate, corresponding to the physiological wearing position of the helmet (above the head).

[0042] Step 202: Based on the pixel coverage of the semantic interest sub-region in the local region image, spatial masking is performed on the texture feature response map corresponding to the texture confidence index to obtain an effective texture feature set that is only retained within the semantic interest sub-region. Based on the effective texture feature set, the connected domain topology that represents the continuity of the surface material of the wearable is extracted.

[0043] Optionally, the compliance inspection system retrieves the texture feature response map corresponding to the texture confidence index obtained in step 20. This texture feature response map is the response intensity distribution map of each pixel in the representation image generated by the convolutional neural network model after extracting features from the local region image, indicating that each pixel belongs to the texture feature of the wearable object. Based on the semantically concerned sub-region, the system determines the pixel coverage range of the sub-region in the local region image, that is, the pixel region defined by the contour coordinates of the semantically concerned sub-region.

[0044] The compliance inspection system performs spatial masking on the texture feature response map based on the pixel coverage of the semantically concerned sub-region. Using masking techniques, it retains only the portion of the response map that overlaps with the semantically concerned sub-region, discarding pixel response information outside this range to obtain an effective texture feature set. This set contains pixel response information representing the texture features of the wearable device within the semantically concerned sub-region. Subsequently, based on the effective texture feature set, it extracts the connected component topology representing the continuity of the wearable device's surface material. This connected component topology is the geometric structure formed by all interconnected pixels in the effective texture feature set, including the number of connected components, their contour shape, area, perimeter, and the relative positional relationships between them.

[0045] Continuing with the embodiment based on step 201, the pixel coverage range of the semantic interest sub-region is from 128 to 172 pixels on the horizontal axis and from 198 to 222 pixels on the vertical axis, containing a total of (172-128+1)×(222-198+1)=45×25=1125 pixels. The pixel range of the texture feature response map is consistent with that of the local area image of the head (90 to 210 pixels on the horizontal axis and 190 to 310 pixels on the vertical axis). After spatial masking based on the pixel coverage range of the semantic interest sub-region, the effective texture feature set contains texture response information of 1125 pixels, and the response intensity of all pixels is above 0.7. Connectivity analysis is performed on the effective texture feature set to extract the topology of the connected components: there is only one connected component, the outline shape is close to a circle, the area is 1080 pixels, the perimeter is 120 pixels, and the outline of the connected component is complete and unbroken, indicating that the surface material of the safety helmet has good continuity.

[0046] Step 203: Based on the geometric circumscribed rectangle boundary of the connected domain topology, compare the overlapping area with the geometric outer frame boundary of the wearable body derived from the centroid coordinates of the wearable body outline in the spatial position index to obtain the geometric overlap rate, which characterizes the degree of overlap between texture distribution and spatial positioning.

[0047] Optionally, the compliance detection system takes the geometric circumscribed rectangle boundary of the connected component topology, which is the smallest rectangular edge boundary that can completely enclose all connected components in the connected component topology. The four sides of this rectangle are parallel to the horizontal and vertical directions of the local region image, and the contour coordinates are determined by the coordinates of the leftmost, rightmost, topmost, and bottommost pixels of the connected components. Simultaneously, the system retrieves the centroid coordinates of the wearable's contour, which are included in the spatial position indicators. Based on these coordinates, the geometric outline boundary of the wearable is derived. This is a rectangular boundary centered on the centroid coordinates of the wearable's contour, pre-defined according to the wearable's standard dimensions, to characterize the expected spatial distribution range of the wearable in the local region image. After determining the geometric circumscribed rectangle boundary and the wearable's geometric outline boundary, the overlapping area of ​​the geometric circumscribed rectangle boundary of the connected component topology and the wearable's geometric outline boundary is compared. The pixel area of ​​the overlapping portion of the two rectangular boundaries is calculated, and then the ratio of this overlapping pixel area to the area of ​​the smaller rectangle boundary is calculated to obtain the geometric overlap rate.

[0048] The geometric overlap rate characterizes the degree of overlap between texture distribution and spatial positioning, and its value ranges from 0 to 1. The closer the value is to 1, the higher the overlap between the actual spatial distribution of the effective texture feature set and the expected spatial distribution of the wearer; the closer the value is to 0, the lower the overlap, which may indicate false detection or abnormal wear position.

[0049] Continuing with the embodiment based on step 201, the compliance detection system extracts the geometric bounding rectangle boundary of the connected domain topology: the leftmost pixel x-coordinate is 130, the rightmost pixel x-coordinate is 170, the topmost pixel y-coordinate is 212, and the bottommost pixel y-coordinate is 232. Therefore, the boundary coordinates are (130, 222) to (170, 242), and the area is (170-130+1)×(242-222+1)=41×21=861 pixels. The centroid coordinates of the helmet outline are extracted as (152, 210). Based on these coordinates, the geometric outline boundary of the helmet is derived: centered at the centroid coordinates (152, 210), extending 20 pixels horizontally to the left and right, and 20 pixels vertically up and down, with boundary coordinates from (132, 190) to (172, 230). The area is (172-132+1)×(230-190+1)=41×41=1681 pixels. The overlapping areas of the two rectangles are as follows: X-axis overlap interval: the intersection of [130, 170] and [132, 172] is [132, 170], with an overlap width of 170-132+1=39 pixels. Y-axis overlap interval: the intersection of [222, 232] and [190, 230] is [222, 230], with an overlap height of 230-222+1=9 pixels. The overlapping area is 39 × 9 = 351 pixels. Therefore, the geometric overlap rate = overlapping area ÷ area of ​​the smaller rectangle = 351 ÷ 861 ≈ 0.41.

[0050] Step 204: Based on the geometric overlap rate, determine whether the effective texture feature set is contained within the geometric outline of the wearable in spatial location, and generate a feature semantic matching index that represents the consistency between texture and spatial logic.

[0051] Optionally, the compliance detection system determines whether the effective texture feature set is spatially contained within the geometric outline of the wearable based on the geometric overlap rate, i.e., comparing the geometric overlap rate with a preset consistency judgment logic. Based on the numerical range of the geometric overlap rate, the inherent logical relationship between the texture features and the spatial location is determined, thereby generating a feature semantic matching index. Specifically, the judgment process is as follows: if the geometric overlap rate is greater than or equal to a first preset threshold (e.g., 0.7), the texture is determined to be well contained within the spatial boundary, and its consistency level is "high coordination"; if the geometric overlap rate is less than the first preset threshold but greater than a second preset threshold (e.g., 0.3), it is determined to be partially contained, and its consistency level is "moderate transition"; if the geometric overlap rate is less than the second preset threshold, the texture is determined to be severely separated from the spatial boundary, and its consistency level is "high conflict". The first preset threshold and the second preset threshold are both empirical parameters determined based on a large amount of historical video data from industrial sites, through offline statistical analysis and ROC curve (Responder Operating Characteristic) evaluation.

[0052] Continuing with the embodiment based on step 204, the first preset threshold is set to 0.7 and the second preset threshold is set to 0.3. Since the geometric overlap rate calculated in step 203 is 0.41, which is less than 0.7 and greater than 0.3, a feature semantic matching index is generated, and its consistency level is marked as "moderate transition", with a value recorded as 0.41.

[0053] Step 205: Based on the geometric inclusion relationship between the centroid coordinates of the wearable outline in the spatial location index and the spatial tolerance area, a judgment is made. If the centroid coordinates of the wearable outline do not fall within the spatial tolerance area, it is determined to be an absolute violation of spatial location, a spatial inactivation signal is generated, and the instantaneous compliance evidence value is mapped to a static zero evidence level that characterizes the determination of the violation state.

[0054] Optionally, based on the spatial tolerance region obtained in step 201 and the centroid coordinates of the wearable outline contained in the spatial location index in step 20, the compliance detection system checks whether the centroid coordinates of the wearable outline in the spatial location index are located inside the boundary of the spatial tolerance region determined in step 201. If it is determined that the centroid coordinates of the wearable outline are outside the boundary of the spatial tolerance region (for example, the centroid of the detection box falls on the worker's shoulder or in the background), then the current frame is directly determined to be an absolute violation in spatial location. At this time, no matter how high the texture confidence is (it may be that there happens to be a similar color object in the background), it is considered invalid evidence, and a spatial disabling signal is generated to block the subsequent transmission of texture validity, forcibly assigning the instantaneous compliance evidence value to a static zero evidence level. This indicates that the detection result of the current frame is a clear violation, with the highest judgment strength, and can be determined as a violation without subsequent temporal verification.

[0055] Continuing with the embodiment based on step 201, the compliance detection system obtains the spatial tolerance region boundary as x-coordinate 128 to 172 pixels, y-coordinate 198 to 222 pixels, and the centroid coordinates of the wearable (helmet) outline as (152, 248). Geometric inclusion is determined: the x-coordinate 152 of the wearable outline centroid coordinates falls within the 128 to 172 pixel range, but the y-coordinate 248 exceeds the 198 to 222 pixel range. Therefore, it is determined that the wearable outline centroid coordinates do not fall within the spatial tolerance region. Based on this determination, it is judged as an absolute spatial position violation, indicating that the helmet's wearing position is severely deviated from the standard position (in the embodiment, the helmet is worn too low, with the center of gravity below the head, exceeding the reasonable tolerance range). Subsequently, a spatial disabling signal is generated to prohibit subsequent redundant texture feature verification operations, and the instantaneous compliance evidence value corresponding to the local area of ​​the head in the current frame is set to a static zero evidence level, characterizing that the target wearing part is clearly in violation.

[0056] Step 206: If the centroid coordinates of the wearable outline fall within the spatial tolerance area, a spatial enabling signal is generated. Based on the spatial enabling signal, the geometric overlap rate in the feature semantic matching index is judged. If the geometric overlap rate is less than the preset consistency threshold, it is determined that there is a semantic conflict between texture distribution and spatial positioning in the current frame. A semantic conflict identifier is generated, the texture confidence index is marked as a state to be verified, and the instantaneous compliance evidence value is mapped to the level of unconfirmed evidence that characterizes the uncertainty of observation.

[0057] Optionally, when the compliance detection system receives a judgment result indicating that the centroid coordinates of the wearable's outline fall within the spatial tolerance area, it generates a spatial enabling signal, indicating that the position is basically compliant, and continues with subsequent texture and spatial logic consistency verification operations. Based on this signal, the geometric overlap rate obtained in step 203 is retrieved and compared with a preset consistency threshold. This consistency threshold is a pre-set critical value used to determine whether there is a semantic conflict between texture distribution and spatial positioning, and is set according to industrial field wearable specifications and detection accuracy requirements (e.g., 0.7, the same as or slightly lower than the first preset threshold). If the geometric overlap rate is less than the preset consistency threshold, it is determined that there is a semantic conflict between texture distribution and spatial positioning in the current frame, that is, the actual distribution of the effective texture feature set is inconsistent with the expected spatial distribution of the wearable, and it cannot be determined whether the wearable is actually worn in compliance (e.g., the wearable outline is clear but the texture features are incomplete, or the texture features exist but the spatial position is offset, etc.).

[0058] After determining that a semantic conflict exists, a semantic conflict identifier is generated to indicate that there is a conflict between texture and spatial semantics in the current frame. Simultaneously, the texture confidence index is marked as pending verification, indicating that the reliability of this index is uncertain. The instantaneous compliance evidence value is then mapped to a pending confirmation evidence level, representing the uncertainty of the observation. This level is one of three levels of instantaneous compliance evidence value, used to indicate that the detection result of the current frame is uncertain and cannot be directly determined as compliant or non-compliant, requiring further confirmation.

[0059] Continuing with the embodiments based on steps 201 and 203, the compliance detection system obtains the spatial tolerance region boundary as x-coordinates from 128 to 172 pixels and y-coordinates from 198 to 222 pixels. The centroid coordinates of the wearable (helmet) outline in the current frame are (152, 210). Geometric inclusion is determined: x-coordinate 152 is within the range of 128 to 172 pixels; y-coordinate 210 is within the range of 198 to 222 pixels. Therefore, it is determined that the centroid coordinates of the wearable outline fall within the spatial tolerance region, and a spatial enable signal is generated.

[0060] The geometric overlap rate obtained in step 203 is 0.41, and the preset consistency threshold is 0.7. Since 0.41 < 0.7, it is determined that there is a semantic conflict between texture distribution and spatial positioning in the current frame (i.e., the geometric bounding rectangle of the effective texture feature set has a low degree of overlap with the geometric outline of the wearable object). A semantic conflict identifier is generated to mark that the detection results of the local area of ​​the head in the current frame have a semantic conflict; at the same time, the texture confidence index of 0.92 obtained in step 20 is marked as a state to be verified; finally, the instantaneous compliance evidence value corresponding to the local area of ​​the head in the current frame is mapped to the level of evidence to be confirmed, which indicates that there is uncertainty in the detection results of the target wearable part.

[0061] Step 207: If the geometric overlap rate is greater than or equal to the preset consistency threshold, the texture and spatial height of the current frame are determined to be highly coordinated, a semantic coordination identifier is generated, the texture confidence index is determined as a valid and credible state, and the instantaneous compliance evidence value is mapped to a dynamic high evidence level that represents the determined compliance state.

[0062] Optionally, if the compliance detection system receives a geometric overlap rate greater than or equal to a preset consistency threshold in step 206, it determines that the texture and space of the current frame are highly coordinated. This means the actual distribution of the effective texture feature set is highly consistent with the expected spatial distribution of the wearable, the texture features are complete, and the spatial position is compliant, clearly indicating that the wearable is being worn compliantly. After determining the high coordination between texture and space, a semantic coordination identifier is generated to mark the semantic consistency between the texture and space of the current frame. Simultaneously, the texture confidence index is determined to be in a valid and credible state, indicating that the index has high credibility and can be directly used as the basis for compliance determination without further verification. The instantaneous compliance evidence value is mapped to a dynamic high evidence level characterizing the determined compliance state. This level characterizes the current frame detection result as a clear compliance state with high judgment strength, serving as the core basis for compliance determination.

[0063] Continuing with the embodiment based on step 206, assuming the geometric overlap rate obtained in step 203 is 0.85 and the preset consistency threshold is 0.7. Since 0.85 ≥ 0.7, the texture and spatial height of the current frame are determined to be highly coordinated (the geometric bounding rectangle of the effective texture feature set has a high degree of overlap with the geometric outline of the wearable object, the texture distribution is complete, and the spatial position is compliant). A semantic coordination identifier is generated to mark that the detection result of the local area of ​​the head of the current frame has consistent texture and spatial height; at the same time, the texture confidence index of 0.92 obtained in step 20 is determined as a valid and credible state; finally, the instantaneous compliance evidence value corresponding to the local area of ​​the head of the current frame is mapped to a dynamic high evidence level, characterizing that the target wearable part is clearly compliant.

[0064] This invention achieves deep fusion verification of texture and spatial features in single-frame detection through the dual collaboration of spatial location verification and texture feature verification. It effectively distinguishes between real wear violations and false abnormal signals caused by environmental interference, solves the misjudgment problem caused by relying on only a single feature in the prior art, and further improves the accuracy of single-frame detection.

[0065] Optionally, the processes of steps 301 to 305 include: Step 301: Based on the level type identifier of the instantaneous compliance evidence value and the preset time sensitivity configuration table, a search is performed to obtain the minimum consecutive confirmation frame threshold that matches the current evidence level. Based on the minimum consecutive confirmation frame threshold, a dynamic time consistency verification standard for the current frame is constructed.

[0066] Optionally, the compliance detection system extracts the level type identifier based on the instantaneous compliance evidence value. This level type identifier corresponds one-to-one with the three levels of the instantaneous compliance evidence value: static zero evidence level identifier, pending confirmation evidence level identifier, and dynamic high evidence level identifier. It then retrieves a preset time-sensitivity configuration table, which stores the correspondence between different level type identifiers and the minimum consecutive confirmation frame threshold. The minimum consecutive confirmation frame threshold is the minimum number of consecutive frames that the target state pointed to by the current frame's instantaneous compliance evidence value needs to appear, used to avoid misjudgments caused by single-frame instantaneous interference. Based on the level type identifier, the system searches the configuration table to find the unique corresponding minimum consecutive confirmation frame threshold. After obtaining the minimum consecutive confirmation frame threshold, the compliance detection system constructs a dynamic temporal consistency verification standard for the current frame. The standard is as follows: the target state pointed to by the instantaneous compliance evidence value of the current frame must appear consecutively in the historical sequence at least a number of times reaching this threshold in order to be considered a temporally valid state.

[0067] In one embodiment, it is assumed that the instantaneous compliance evidence value of the current frame (the local area of ​​the head of target person 1) is a pending evidence level, and the corresponding level type identifier is "pending evidence level identifier". The time sensitivity configuration table specifies: static zero evidence level identifier → minimum consecutive confirmation frame threshold of 1 frame; pending evidence level identifier → minimum consecutive confirmation frame threshold of 3 frames; dynamic high evidence level identifier → minimum consecutive confirmation frame threshold of 2 frames. Based on the "pending evidence level identifier", the threshold of 3 frames is obtained. Therefore, the dynamic time consistency verification standard is constructed as follows: the target state (pending confirmation state) of the current frame must appear consecutively for 3 or more frames at the end of the historical frame state sequence to meet the time consistency requirement.

[0068] Step 302: Based on the dynamic temporal consistency verification standard and the state sequence of multiple historical frames in the compliance status information of historical frames, determine the longest consecutive run length of the target state pointed to by the instantaneous compliance evidence value of the current frame at the end of the state sequence.

[0069] Optionally, the compliance detection system extracts the state sequence of multiple historical frames from the compliance state information of historical frames from the historical state cache queue. This sequence consists of the compliance state information corresponding to a preset number of consecutive video frames preceding the current frame, arranged in chronological order (from earliest to latest). Subsequently, the target state pointed to by the instantaneous compliance evidence value of the current frame is determined. The correspondence is as follows: static zero evidence level → violation state; pending confirmation evidence level → pending confirmation state; dynamic high evidence level → compliance state. Combining the dynamic temporal consistency verification standard, the longest consecutive run length of the target state at the end of the state sequence is determined. This involves starting from the end of the historical frame state sequence and checking frame by frame against the target state until a different state is encountered. The number of consecutive consistent frames is the longest consecutive run length. If the state of the first frame at the end of the state sequence is inconsistent with the target state, the run length is 0 frames.

[0070] Continuing with the embodiment based on step 301, the preset number of historical frames is 5 frames. The state sequence arranged in chronological order is: pending confirmation state, pending confirmation state, compliant state, pending confirmation state, pending confirmation state (the last frame is the historical frame closest to the current frame). The instantaneous compliance evidence value of the current frame is the pending confirmation evidence level, and the target state is pending confirmation state. Counting backwards from the end of the state sequence, the last frame is pending confirmation state, the second to last frame is pending confirmation state, and the third to last frame is compliant state (stop counting). The longest consecutive run length of the pending confirmation state at the end of the state sequence is 2 frames.

[0071] Step 303: Compare the length of the longest consecutive run with the minimum consecutive acknowledgment frame threshold to obtain the temporal continuity compliance indicator.

[0072] Optionally, the compliance detection system compares the minimum consecutive acknowledgment frame threshold obtained in step 301 with the longest consecutive run length obtained in step 302, and generates a temporal continuity compliance flag based on the comparison result. That is, if the longest consecutive run length is greater than or equal to the minimum consecutive acknowledgment frame threshold, the temporal continuity compliance flag is true; if the longest consecutive run length is less than the minimum consecutive acknowledgment frame threshold, the temporal continuity compliance flag is false.

[0073] Continuing with the embodiment based on step 302, the minimum consecutive confirmation frame threshold obtained by the compliance detection system is 3 frames, and the longest consecutive run length is 2 frames. Comparison result: 2 frames < 3 frames, temporal continuity does not meet the standard, and the generated temporal continuity compliance flag is false.

[0074] Step 304: Based on the temporal continuity compliance flag and the level type flag of the instantaneous compliance evidence value, determine the state inertial unlocking, and generate an instruction based on the judgment result to determine the state transition permission of the current frame; wherein, when the instantaneous compliance evidence value is a static zero evidence level, a forced flip instruction is generated to skip the inertial constraint; when the temporal continuity compliance flag is true, an inertial unlocking instruction is generated; when the temporal continuity compliance flag is false, an inertial hold instruction is generated.

[0075] Optionally, the compliance detection system performs state inertia unlocking judgment based on the temporal continuity compliance indicator and the level type indicator of the instantaneous compliance evidence value of the current frame. That is, it judges whether the target state of the current frame has the authority to break through the compliance state inertia of the previous frame and realize state transition. Based on the judgment result, it generates the corresponding instruction to determine the state transition authority. The specific determination process is as follows: When the instantaneous compliance evidence value is at the static zero evidence level, it indicates that the target state of the current frame is a clear violation state, which is a serious violation. There is no need to consider the temporal continuity, and a forced flip instruction is generated to skip the inertial constraint. When the temporal continuity compliance indicator is true and the instantaneous compliance evidence value is not at the static zero evidence level, it indicates that the target state has sufficient temporal continuity and can eliminate the influence of instantaneous interference. An inertial unlock instruction is generated, that is, flipping is allowed, and the inertial constraint of the previous frame's compliant state is released. When the temporal continuity compliance indicator is false and the instantaneous compliance evidence value is not at the static zero evidence level, an inertial hold instruction is generated, indicating that the temporal continuity of the target state is insufficient and may be affected by instantaneous interference. An inertial hold instruction is generated, that is, flipping is prohibited, and the inertia of the previous frame's compliant state is maintained.

[0076] Continuing with the embodiment based on step 303, if the temporal continuity compliance indicator obtained by the compliance detection system is false and the instantaneous compliance evidence value of the current frame is a pending confirmation evidence level (not a static zero evidence level), which meets the judgment condition that the temporal continuity compliance indicator is false and the instantaneous compliance evidence value is not a static zero evidence level, an inertia maintenance instruction is generated to maintain the compliance state inertia of the previous frame and prohibit the target state (pending confirmation state) of the current frame from transitioning to a new state.

[0077] Step 305: Based on the state transition permission and the compliance status of the previous frame, and combined with the preset industrial safety state machine transition graph, reachability verification is performed. When the state transition permission allows flipping and the target state has a direct directed connection with the previous frame state in the state machine transition graph, the target state is confirmed as the preliminary compliance status of the current frame. When there is no direct directed connection or the state transition permission prohibits flipping, the compliance status of the previous frame is determined as the preliminary compliance status of the current frame.

[0078] Optionally, the industrial safety state machine transition map is a pre-defined standardized map used to regulate the transition rules between compliant states. This map records all the directed connections between the three compliant states (compliant, non-compliant, and pending confirmation), that is, it clearly defines which states (current frame target state) a certain state (previous frame state) can directly transition to, and which states are not allowed to transition directly, in order to ensure the rationality and standardization of state transitions.

[0079] The compliance detection system, based on the obtained state transition permissions (i.e., the instructions generated in step 304), retrieves the compliance state of the previous frame as the benchmark for determining the preliminary compliance state of the current frame, and performs reachability verification by retrieving a preset industrial safety state machine transition graph. When the instruction is a forced flip instruction or an inertial unlock instruction, the state transition permission is allowed to flip. Subsequently, the graph is queried to determine whether there is a direct directed connection between the target state of the current frame and the compliance state of the previous frame. If a direct directed connection exists, the target state is confirmed as the preliminary compliance state of the current frame; if no direct directed connection exists, the compliance state of the previous frame is determined as the preliminary compliance state of the current frame, and unauthorized transitions are prohibited. When the instruction is an inertial hold instruction, the state transition permission is prohibited to flip, and the compliance state of the previous frame is directly determined as the preliminary compliance state of the current frame.

[0080] Continuing with the embodiment based on step 304, the current frame instruction obtained by the compliance detection system is an inertia-maintaining instruction, and the state transition permission is "prohibited from flipping". The compliance state of the previous frame is "pending confirmation", and the target state of the current frame is also "pending confirmation". According to the reachability verification rules, when the permission is "prohibited from flipping", the pending confirmation state of the previous frame is directly determined as the preliminary compliance state of the current frame.

[0081] In another embodiment, the instruction is an inertial unlock instruction (allowing flipping), the previous frame is in a pending confirmation state, the target state is a compliant state, and a query of the graph reveals a direct directed connection between the two. In this case, the compliant state is confirmed as the preliminary compliant state of the current frame.

[0082] The embodiments of the present invention achieve deep fusion of instantaneous evidence values ​​and temporal information. By filtering single-frame interference through temporal continuity verification and maintaining state stability through state inertia control and transition rule constraints, the invention solves the problems of weak anti-interference ability and large state fluctuation caused by relying solely on single-frame detection in the prior art. This enables the generated preliminary compliance state to accurately reflect the temporal trend of the target person's wearing status.

[0083] Optionally, the process of steps 306 to 309 includes: Step 306: Based on the historical state sequence of the initial compliance state within a preset time window, determine the cumulative number of state transitions that occurred in the initial compliance state within the time window, and compare the cumulative number with a preset stability grading threshold range to obtain the temporal stability level identifier of the current frame.

[0084] Optionally, the compliance detection system retrieves the preliminary compliance status of the current frame obtained in step 305, and simultaneously retrieves a preset time window (such as the last 5 frames), and counts the number of state transitions (i.e. changes) of the preliminary compliance status generated within the preset time window. That is, it compares two adjacent frames one by one, and if they are inconsistent, it is counted as 1 transition. Finally, the cumulative number is obtained, which is the temporal stability feature in step 30.

[0085] Subsequently, the compliance detection system retrieves a preset stability grading threshold range. This range is a pre-defined set of thresholds used to grade the temporal stability of the initial compliance state. It contains multiple consecutive and non-overlapping threshold ranges, each corresponding to a temporal stability level identifier. The fewer the cumulative number of state transitions, the higher the temporal stability level; conversely, the more cumulative the number of transitions, the lower the temporal stability level. Therefore, the cumulative number of state transitions is compared with the stability grading threshold range to determine which range it falls into, thus obtaining the corresponding temporal stability level identifier.

[0086] In one embodiment, assuming the initial compliance status of the current frame is "pending confirmation," the preset time window length is 5 frames, and the historical state sequence within the window is: pending confirmation, pending confirmation, compliance, pending confirmation, pending confirmation (the last frame is the current frame). The cumulative number of state transitions is counted as follows: Frame 1 and 2: both are pending confirmation → no transition; Frame 2 and 3: pending confirmation vs. compliance → 1 transition; Frame 3 and 4: compliance vs. pending confirmation → 1 transition; Frame 4 and 5: both are pending confirmation → no transition. The cumulative number is 2. The stability grading threshold range is defined as follows: 0 cumulative times → high stability level; 1-2 cumulative times → medium stability level; 3 or more cumulative times → low stability level. Therefore, 2 transitions fall within the 1-2 range, resulting in a medium stability level.

[0087] Step 307: Based on the geometric overlap rate in the feature semantic matching index, a comparison is made with the preset semantic consistency classification threshold range to obtain the semantic consistency level identifier of the current frame.

[0088] Optionally, the compliance detection system retrieves the geometric overlap rate from the feature semantic matching index obtained in step 203, and retrieves a preset semantic consistency grading threshold range. This semantic consistency grading threshold range is a pre-defined set of threshold ranges used to grade the semantic consistency between texture features and spatial positioning. It contains multiple continuous and non-overlapping threshold ranges, each corresponding to a semantic consistency level identifier. The higher the geometric overlap rate, the higher the semantic consistency level; the lower the geometric overlap rate, the lower the semantic consistency level. Therefore, the geometric overlap rate value is compared with the preset semantic consistency grading threshold range to determine the range it falls into, thereby obtaining the corresponding semantic consistency level identifier.

[0089] In one embodiment, assume that the geometric overlap rate extracted from the feature semantic matching index is 0.41. The semantic consistency classification threshold range is defined as follows: geometric overlap rate of 0.7 and above → high consistency level; geometric overlap rate of 0.4-0.7 → medium consistency level; geometric overlap rate below 0.4 → low consistency level. Since 0.41 falls into the 0.4-0.7 range, the semantic consistency level is thus determined to be medium consistency level.

[0090] Step 308: Based on the combined encoding of the temporal stability level identifier and the semantic consistency level identifier, retrieve the state correction strategy instruction that uniquely corresponds to the current combined encoding from the preset two-dimensional state correction mapping table.

[0091] Optionally, the compliance testing system combines the temporal stability level identifier obtained in step 306 with the semantic consistency level identifier obtained in step 307 to form a two-dimensional combined code. This combined code is a unique encoded string formed according to preset rules. For example, if the high stability level identifier is encoded as "G", the medium stability level identifier is encoded as "Z", and the low stability level identifier is encoded as "D"; if the high consistency level identifier is encoded as "G", the medium consistency level identifier is encoded as "Z", and the low consistency level identifier is encoded as "D", then the combined code of medium stability + medium consistency is "ZZ".

[0092] The two-dimensional state correction mapping table is a pre-defined, standardized mapping table used to store the correspondence between combined codes and state correction policy instructions. This mapping table stores the correspondence between combined codes and state correction policy instructions, with each combined code uniquely corresponding to one state correction policy instruction. This state correction policy instruction guides the system in generating state correction parameters, containing specific logic and requirements. Different combined codes correspond to different policies. For example, high timing stability and high semantic consistency → light correction policy; low timing stability and low semantic consistency → heavy correction policy; medium timing stability and medium semantic consistency → medium correction policy. Therefore, by using the combined code as the key, a unique corresponding state correction policy instruction is obtained by searching the mapping table. This instruction contains guidance on system behavior for the current specific operating condition (such as high noise + low quality, or low noise + high quality).

[0093] In one embodiment, assume the temporal stability level identifier is a medium stability level identifier, and the semantic consistency level identifier is a medium consistency level identifier. According to the encoding rules: the medium stability level identifier is encoded as "Z", the medium consistency level identifier is encoded as "Z", and the combined encoding is "ZZ". Based on the two-dimensional state correction mapping table, the following are included but not limited to: "GG" (high stability + high consistency) → mild correction policy instruction; "ZZ" (medium stability + medium consistency) → medium correction policy instruction; "DD" (low stability + low consistency) → severe correction policy instruction; "GZ" (high stability + medium consistency) → mild correction policy instruction. The state correction policy instruction corresponding to the combined encoding "ZZ" is retrieved as a medium correction policy instruction.

[0094] Step 309: Based on the state correction strategy instruction, parse it to generate state correction parameters including time lag compensation step size, conflict suppression level identifier and state lock flag.

[0095] Optionally, the compliance detection system parses the received state correction strategy instructions to extract specific configuration information of the state correction parameters, including the time lag compensation step size, conflict suppression level identifier, and state lock flag. The time lag compensation step size is used to compensate for the detection delay caused by time lag filtering, and its value is a positive integer (e.g., 1, 2, 3). The smaller the step size, the smaller the compensation delay and the faster the response speed; the larger the step size, the greater the compensation delay and the stronger the anti-interference capability. The conflict suppression level identifier is used to identify the conflict suppression strength, and its value is also a positive integer (e.g., 1, 2, 3). The higher the level, the stronger the conflict suppression strength; the lower the level, the weaker the conflict suppression strength. The state lock flag indicates whether the current preliminary compliance state is locked, specifically including two states: on / off. On indicates locking to avoid subsequent instantaneous interference leading to erroneous corrections; off indicates not locking, allowing corrections based on subsequent detection results. After completing the instruction parsing, a complete state correction parameter containing these three items is generated.

[0096] Continuing with the embodiment based on step 308, the status correction strategy instruction obtained by the compliance detection system is a medium correction strategy instruction, and the configuration information obtained by parsing is: the time lag compensation step size is 2, the conflict suppression level identifier is 2, and the status lock flag is off.

[0097] Based on this, the generated complete state correction parameters include: a time lag compensation step size of 2 (moderate compensation for delay, balancing response speed and anti-interference capability), a conflict suppression level flag of 2 (moderate conflict suppression strength, effectively suppressing semantic conflict interference while avoiding over-suppression), and a state lock flag that is off (not locking the initial compliance state, allowing fine-tuning based on subsequent detection results).

[0098] The embodiments of the present invention achieve dual-dimensional collaborative management of temporal stability and semantic consistency. It can dynamically adjust the state correction parameters according to the current detection conditions (interference intensity, detection reliability), avoiding the drawback of fixed correction parameters being unable to adapt to different interference scenarios, and ensuring the pertinence and accuracy of state correction.

[0099] Optionally, the processes of steps 401 to 405 include: Step 401: Perform state freeze determination based on the state lock flag bit included in the state correction parameters. When the state lock flag bit indicates forced lock, determine the target compliance state of the previous frame as the target compliance state of the current frame.

[0100] Optionally, the compliance detection system performs a state freeze determination based on the state lock flag included in the state correction parameters. Specifically, it determines whether the state lock flag locks the current preliminary compliance state. When the state lock flag indicates a forced lock, it means that the current detection environment is experiencing extreme interference or is in system protection mode, and any state transition is considered unreliable. Therefore, a state freeze is performed, ignoring the preliminary compliance state of the current frame, and the final target compliance state determined in the previous frame is copied and assigned to the target compliance state of the current frame.

[0101] In one embodiment, the compliance detection system obtains the status lock flag as "forced lock" and determines that the target compliance status of the current frame needs to be frozen. At this time, the target compliance status of the previous frame is retrieved as the pending confirmation status, and the target compliance status (pending confirmation status) of the previous frame is directly determined as the target compliance status of the current frame, thus completing the determination of the target compliance status of the current frame without needing to proceed to subsequent correction steps.

[0102] In another embodiment, assuming the state lock flag indicates "unlocked", the compliance detection system will not freeze the state and will directly proceed to the subsequent correction process.

[0103] Step 402: When the state lock flag indicates that it is not locked, a dynamic backtracking window is constructed with the current video frame as the endpoint and the time lag compensation step size as the length, based on the time lag compensation step size contained in the state correction parameters. Based on all historical compliant state records and their corresponding stability flags in the dynamic backtracking window, the time-series state slice sequence is determined.

[0104] Optionally, when the compliance detection system receives a non-locked status lock flag, it extracts the time lag compensation step size from the status correction parameters and traces back a time lag compensation step size units, using the current video frame as the end point of the time axis and the time lag compensation step size as the length, to form a dynamic backtracking window. This dynamic backtracking window contains a set of consecutive video frames, including the current frame and historical frames, and the number of video frames in the window is equal to the value of the time lag compensation step size.

[0105] After constructing the dynamic backtracking window, the compliance detection system extracts the compliance status records of all historical frames within the scope of the dynamic backtracking window from the historical status cache queue. This includes all historical compliance status records and their corresponding stability markers. The historical compliance status records represent the compliance status of each video frame within the dynamic backtracking window (including the target compliance status of the previous and earlier historical frames, and the preliminary compliance status of the current frame). The stability markers represent the stability level of the corresponding compliance status and correspond to the temporal stability level identifiers generated in step 306, namely high stability, medium stability, and low stability. All historical compliance status records and their corresponding stability markers within the dynamic backtracking window are arranged in the chronological order of video frame acquisition (from earliest to latest) to form a temporal status slice sequence, where each sequence element contains the compliance status of one frame and its stability marker.

[0106] Continuing with the embodiments based on steps 309 and 401, the compliance detection system obtains that the state lock flag indicates that it is unlocked, and extracts a time lag compensation step size of 2 (a positive integer) from the state correction parameters. Based on this time lag compensation step size, a dynamic backtracking window is constructed: with the current frame as the endpoint, the window length is 2, including the current frame and the previous frame (a total of 2 video frames). The historical compliance status records and their corresponding stability markers within the dynamic backtracking window are retrieved to obtain the following: the previous frame: the target compliance status is pending confirmation, and the corresponding stability marker is a medium stability marker; the current frame: the preliminary compliance status is pending confirmation, and the corresponding stability marker is a medium stability marker. Arranging the above combined information in chronological order (previous frame, current frame), the temporal state slice sequence is obtained as: [(pending confirmation status, medium stability marker), (pending confirmation status, medium stability marker)].

[0107] Step 403: Based on the equality comparison of each historical compliance status record in the time-series state slice sequence with the preliminary compliance status of the current frame, the state continuity length is obtained, and the time validity identifier is determined based on the relationship between the state continuity length and the time lag compensation step size; wherein, when the state continuity length is greater than or equal to the time lag compensation step size, the time validity identifier is valid; otherwise, the time validity identifier is invalid.

[0108] Optionally, the compliance detection system extracts the preliminary compliance status of the current frame based on the time-series state slice sequence, and performs an item-by-item equality comparison between each historical compliance status record in the time-series state slice sequence and the preliminary compliance status of the current frame. The comparison process is as follows: it checks whether each historical compliance status record in the time-series state slice sequence is completely consistent with the preliminary compliance status; if a certain historical compliance status record is completely consistent with the preliminary compliance status (i.e., both are compliant, both are non-compliant, or both are pending confirmation), it is determined that the comparison is consistent; if they are inconsistent, it is determined that the comparison is inconsistent.

[0109] During the comparison process, the compliance detection system starts from the last frame (the current frame) of the time-series state slice sequence and traces backward frame by frame, counting the number of consecutive historical compliance state records that are completely consistent with the initial compliance state of the current frame, until a historical compliance state record that is inconsistent with the comparison is encountered, at which point the counting stops. The number of consecutive consistent frames counted is the state continuity length; if the historical compliance state record of the last frame (the current frame) of the time-series state slice sequence is inconsistent with the initial compliance state, then the state continuity length is 0.

[0110] The compliance detection system compares the statistically obtained state continuity length with the time lag compensation step size extracted in step 402. When the state continuity length is greater than or equal to the time lag compensation step size, the timing validity is marked as valid, meaning that the preliminary compliance state of the current frame has sufficient timing continuity to eliminate the influence of instantaneous interference. When the state continuity length is less than the time lag compensation step size, the timing validity is marked as invalid, meaning that the preliminary compliance state of the current frame has insufficient timing continuity and may be affected by instantaneous interference.

[0111] Continuing with the embodiment based on step 402, the compliance detection system obtains the time-series state slice sequence as [(state to be confirmed, medium stability marker), (state to be confirmed, medium stability marker)], and the initial compliance state of the current frame is the state to be confirmed. A step-by-step equality comparison reveals that: the historical compliance state record of the previous frame is the state to be confirmed, consistent with the initial compliance state; the historical compliance state record of the current frame is the state to be confirmed, consistent with the initial compliance state; both comparisons are consistent. Tracing back from the last frame (the current frame) of the time-series state slice sequence, the historical compliance state records of both frames are consistent with the initial compliance state, with no inconsistencies. Therefore, the continuous length of the state is 2. The time lag compensation step size extracted in step 402 is 2. The continuous length of the state (2) is compared with the time lag compensation step size (2). 2 equals 2, so the time sequence validity is determined to be valid.

[0112] Step 404: Based on the timing validity identifier and the preliminary compliance status of the current frame, physical reachability is verified by combining the preset industrial safety state machine transition graph to obtain a state permission instruction; wherein, when the timing validity identifier is valid and there is a legal directed path in the state machine transition graph from the target compliance status of the previous frame, a state permission instruction is generated as a state transition permission instruction; otherwise, a state permission instruction is generated as a state maintenance permission instruction.

[0113] Optionally, the compliance detection system, based on the timing validity identifier and the preliminary compliance status of the current frame, retrieves the target compliance status of the previous frame and invokes a preset industrial safety state machine transition graph. Based on the timing validity identifier, the preliminary compliance status of the current frame, the target compliance status of the previous frame, and the industrial safety state machine transition graph, it determines whether the preliminary compliance status of the current frame can legally transition from the target compliance status of the previous frame. This ensures the rationality and legality of state transitions and avoids unauthorized transitions.

[0114] When the time sequence validity is valid and there is a legal directed path from the target compliance state of the previous frame to the initial compliance state, the generated state permission instruction is a state transition permission instruction, which allows the initial compliance state of the current frame to transition to the target compliance state. When the timing validity flag is invalid, or when there is no legitimate directed path from the target compliance state of the previous frame to the initial compliance state, the generated state permission instruction is a state maintenance permission instruction, which prohibits the transition of the initial compliance state of the current frame and maintains the relevant logic of the target compliance state of the previous frame.

[0115] Continuing with the embodiment based on step 403, the compliance detection system obtains the timing validity identifier as valid, the preliminary compliance state of the current frame is a pending confirmation state, retrieves the target compliance state of the previous frame as a pending confirmation state, and invokes the preset industrial safety state machine transition graph. This industrial safety state machine transition graph clearly states that the pending confirmation state can directly transition to the compliance state, the violation state, or itself; that is, there is a legal directed path between the pending confirmation state (the target compliance state of the previous frame) and the pending confirmation state (the preliminary compliance state of the current frame). Therefore, the timing validity identifier is valid, and the preliminary compliance state has a legal directed path from the target compliance state of the previous frame, a state transition permission instruction is generated, allowing the preliminary compliance state (pending confirmation state) of the current frame to transition to the target compliance state.

[0116] Step 405: Based on the state permission instruction and the conflict suppression level identifier contained in the state correction parameters, a state consistency verification is performed to obtain the target compliance state. Specifically, if the state permission instruction is a state transition permission instruction, the current preliminary compliance state is determined as the target compliance state; if the state permission instruction is a state maintenance permission instruction, when the conflict suppression level identifier indicates strong suppression, a search is performed based on the historical state records corresponding to the dynamic backtracking window, and the most recently marked high-stability compliance state is determined as the target compliance state; when the conflict suppression level identifier indicates non-strong suppression, the target compliance state of the previous frame is determined as the target compliance state of the current frame.

[0117] Optionally, the compliance detection system extracts the conflict suppression level identifier from the state correction parameters. The conflict suppression level identifier is a marker used to identify the strength of conflict suppression. Specifically, it is divided into two types: strong suppression and non-strong suppression. Strong suppression requires strong suppression of interference caused by texture and spatial semantic conflicts, and prioritizes historical high stability states, which corresponds to the conflict suppression level identifier greater than 2 in step 309. Non-strong suppression does not require strong suppression of interference and prioritizes maintaining state inertia, which corresponds to the conflict suppression level identifier less than or equal to 2 in step 309.

[0118] The compliance detection system performs state consistency verification based on the state permission instruction and the conflict suppression level identifier contained in the state correction parameters to determine the target compliance state of the current frame, specifically including the following two cases: If the state permission instruction is a state transition permission instruction, it means that the preliminary compliance state of the current frame has legal transition conditions and sufficient temporal continuity, and there is no need to perform abnormal rollback. In this case, the preliminary compliance state of the current frame is directly determined as the target compliance state, and the correction and generation of the target compliance state is completed.

[0119] If the status permission instruction is a status maintenance permission instruction, it indicates that the initial compliance status of the current frame does not meet the transition conditions and an abnormal rollback is required. In this case, further judgment is needed based on the conflict suppression level indicator. That is, when the conflict suppression level indicator indicates strong suppression, the historical status record corresponding to the dynamic backtracking window constructed in step 402 is retrieved, and the most recent compliance status marked as high stability is found. This high-stability compliance status is determined as the target compliance status of the current frame. Here, a high-stability compliance status is a status in the historical status record corresponding to the dynamic backtracking window where the compliance status record itself is "compliant" (distinguished from violation or pending confirmation status), and the stability mark corresponding to this compliance status record is a high-stability mark. This stability mark is a marker used to characterize the stability of the corresponding compliance status. It corresponds to the temporal stability level indicator generated in step 306. The high stability mark corresponds to the high stability level indicator, indicating that the compliance status has high temporal stability within the preset time window, is less affected by instantaneous interference, and has high credibility.

[0120] When the conflict suppression level indicator is non-strong suppression, the target compliance status of the previous frame is retrieved and directly determined as the target compliance status of the current frame to avoid status fluctuations caused by excessive rollback.

[0121] Continuing with the embodiment based on step 404, the compliance detection system obtains the state permission instruction as a state transition permission instruction and extracts the conflict suppression level identifier as 2 (non-strong suppression type) from the state correction parameters. Since the state permission instruction is a state transition permission instruction, it meets the judgment rules of the first case and no abnormal rollback is required. Therefore, the preliminary compliance state (pending confirmation state) of the current frame is directly determined as the target compliance state of the current frame, completing the correction and generation of the target compliance state.

[0122] The embodiments of the present invention achieve deep integration of state correction parameters with timing information and state transition rules. It can adaptively complete the calibration of the initial compliance state according to the current field interference conditions and detection reliability, effectively eliminating the impact of instantaneous interference and semantic conflicts, and solving the problems of lack of specificity in state correction and susceptibility to interference leading to false alarms and missed alarms in the prior art.

[0123] Furthermore, the edge computing-based industrial field wearable compliance testing system provided by the present invention will be described below. The edge computing-based industrial field wearable compliance testing system described below can be referred to in correspondence with the edge computing-based industrial field wearable compliance testing method described above.

[0124] Optional, refer to Figure 2 , Figure 2 This is a schematic diagram of the edge computing-based industrial field wearable compliance testing system provided by the present invention. The edge computing-based industrial field wearable compliance testing system includes: The adaptive region extraction module 210 is used to determine an adaptive search region containing the target's clothing parts based on the pose geometry information of the target person in the current video frame, and to extract the local region image based on the adaptive search region in the current video frame. The single-frame dual-index generation module 220 is used to perform feature reasoning based on local region images to obtain texture confidence index and spatial location index, and to generate instantaneous compliance evidence value and feature semantic matching index of the current frame based on texture confidence index and spatial location index. The time lag filtering module 230 is used to perform time lag filtering based on instantaneous compliance evidence value and compliance status information of historical frames to obtain the preliminary compliance status of the current frame. Based on the mapping relationship between the temporal stability characteristics of the preliminary compliance status and the feature semantic matching index within a preset time window, the status correction parameters are determined. The semantic joint correction module 240 is used to correct the initial compliance status based on the status correction parameters to obtain the target compliance status, and generate a structured compliance detection result based on the target compliance status.

[0125] The embodiments of the present invention solve the problem that the existing technology is difficult to balance anti-interference capability and real-time response speed in complex dynamic interference scenarios. At the same time, it reduces the false alarm rate and false alarm rate caused by the inability to distinguish between real violations and false signals, and realizes low latency response and high accuracy detection of wearable compliance detection in complex and highly interference-prone industrial operations.

[0126] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an electronic device 300 provided in this embodiment of the invention includes a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 40.

[0127] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 40.

[0128] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the edge computing-based industrial field wearable compliance testing method provided by the above methods, which includes steps 10 to 40.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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.

Claims

1. An edge computing-based industrial field wearable compliance testing method, characterized in that, include: Based on the pose geometry information of the target person in the current video frame, an adaptive search region containing the target's clothing parts is determined, and the current video frame is extracted based on the adaptive search region to obtain a local region image; Based on the local region image, feature reasoning is performed to obtain texture confidence index and spatial location index. Based on the texture confidence index and spatial location index, the instantaneous compliance evidence value and feature semantic matching index of the current frame are generated. Based on the instantaneous compliance evidence value and the compliance status information of historical frames, a time-series lag filtering process is performed to obtain the preliminary compliance status of the current frame. Based on the temporal stability characteristics of the preliminary compliance status within a preset time window and the mapping relationship with the feature semantic matching index, the status correction parameters are determined. The initial compliance status is corrected based on the status correction parameters to obtain the target compliance status, and a structured compliance detection result is generated based on the target compliance status.

2. The industrial field wearable compliance testing method based on edge computing according to claim 1, characterized in that, Generate feature semantic matching metrics for the current frame, including: Based on the relative vector direction between the centroid coordinates of the wearable outline and the center point coordinates of the head in the spatial position index, combined with the preset human head anatomical structure parameters, the boundary of the spatial tolerance area is determined, and a geometric projection interval representing the expected distribution range of the wearable is obtained. Based on the geometric projection interval, the local area image is divided to obtain a semantic attention sub-region corresponding to the physiological wearing position of the wearable. Based on the pixel coverage of the semantic attention sub-region in the local region image, the texture feature response map corresponding to the texture confidence index is spatially masked to obtain an effective texture feature set that is only retained in the semantic attention sub-region, and the connected domain topology that characterizes the continuity of the surface material of the wearable is extracted based on the effective texture feature set. Based on the geometric circumscribed rectangle boundary of the connected domain topology, the overlapping area of ​​the geometric outer frame boundary of the wearable derived from the centroid coordinates of the wearable outline in the spatial position index is compared to obtain the geometric overlap rate, which characterizes the degree of overlap between texture distribution and spatial positioning. Based on the geometric overlap rate, it is determined whether the effective texture feature set is spatially contained within the geometric outline of the wearable, and the feature semantic matching index representing the consistency between texture and spatial logic is generated.

3. The industrial field wearable compliance testing method based on edge computing according to claim 2, characterized in that, Generate the instantaneous compliance evidence value for the current frame, including: The determination is based on the geometric inclusion relationship between the centroid coordinates of the wearable outline in the spatial location index and the spatial tolerance area. If the centroid coordinates of the wearable outline do not fall within the spatial tolerance area, it is determined to be an absolute violation of spatial location, a spatial inactivation signal is generated, and the instantaneous compliance evidence value is mapped to a static zero evidence level that characterizes the determination of the violation status. If the centroid coordinates of the wearable outline fall within the spatial tolerance area, a spatial enabling signal is generated. Based on the spatial enabling signal, the geometric overlap rate in the feature semantic matching index is judged. If the geometric overlap rate is less than a preset consistency threshold, it is determined that there is a semantic conflict between texture distribution and spatial positioning in the current frame. A semantic conflict identifier is generated, the texture confidence index is marked as a state to be verified, and the instantaneous compliance evidence value is mapped to a level of evidence to be confirmed that characterizes the uncertainty of observation. If the geometric overlap rate is greater than or equal to a preset consistency threshold, the texture and spatial height of the current frame are determined to be highly coordinated, a semantic coordination identifier is generated, the texture confidence index is determined to be a valid and trustworthy state, and the instantaneous compliance evidence value is mapped to a dynamic high evidence level that represents the determination of the compliance state.

4. The industrial field wearable compliance testing method based on edge computing according to claim 1, characterized in that, The instantaneous compliance evidence value includes a static zero evidence level, an evidence level to be confirmed, or a dynamic high evidence level. The step of performing time-lag filtering based on the instantaneous compliance evidence value and the compliance status information of historical frames to obtain the preliminary compliance status of the current frame includes: Based on the level type identifier of the instantaneous compliance evidence value and the preset time sensitivity configuration table, a minimum consecutive confirmation frame threshold matching the current evidence level is obtained, and a dynamic time consistency verification standard for the current frame is constructed based on the minimum consecutive confirmation frame threshold. Based on the dynamic temporal consistency verification standard and the state sequence of multiple historical frames in the compliance status information of historical frames, the longest consecutive run length of the target state pointed to by the instantaneous compliance evidence value of the current frame at the end of the state sequence is determined. A temporal continuity compliance indicator is obtained by comparing the longest consecutive run length with the minimum consecutive acknowledgment frame threshold. Based on the temporal continuity compliance indicator and the level type indicator of the instantaneous compliance evidence value, a state inertial unlocking determination is made, and an instruction is generated based on the determination result to determine the state transition permission of the current frame; wherein, when the instantaneous compliance evidence value is a static zero evidence level, a forced flip instruction is generated to skip the inertial constraint; when the temporal continuity compliance indicator is true, an inertial unlocking instruction is generated; when the temporal continuity compliance indicator is false, an inertial hold instruction is generated. Based on the state transition permission and the compliance status of the previous frame, and combined with the preset industrial safety state machine transition graph, reachability verification is performed. When the state transition permission allows flipping and the target state has a direct directed connection with the previous frame state in the state machine transition graph, the target state is confirmed as the preliminary compliance status of the current frame. When there is no direct directed connection or the state transition permission prohibits flipping, the compliance status of the previous frame is determined as the preliminary compliance status of the current frame.

5. The industrial field wearable compliance testing method based on edge computing according to claim 4, characterized in that, The determination of state correction parameters based on the mapping relationship between the temporal stability characteristics of the initial compliance state within a preset time window and the feature semantic matching index includes: Based on the historical state sequence of the initial compliance state within a preset time window, the cumulative number of state transitions of the initial compliance state within the time window is determined, and the cumulative number is compared with a preset stability grading threshold range to obtain the temporal stability level identifier of the current frame. The semantic consistency level identifier of the current frame is obtained by comparing the geometric overlap rate in the feature semantic matching index with the preset semantic consistency level threshold range. Based on the combined encoding of the temporal stability level identifier and the semantic consistency level identifier, a state correction strategy instruction that uniquely corresponds to the current combined encoding is obtained by searching in a preset two-dimensional state correction mapping table. The state correction strategy instruction is parsed to generate state correction parameters that include the time lag compensation step size, conflict suppression level identifier, and state lock flag.

6. The industrial field wearable compliance testing method based on edge computing according to claim 1, characterized in that, The state correction parameters include the time lag compensation step size, the conflict suppression level indicator, and the state lock flag. The step of correcting the preliminary compliance status based on the status correction parameters to obtain the target compliance status includes: The state freeze determination is performed based on the state lock flag bit included in the state correction parameters. When the state lock flag bit indicates forced lock, the target compliance state of the previous frame is determined as the target compliance state of the current frame. When the state lock flag indicates that it is not locked, a dynamic backtracking window is constructed with the current video frame as the endpoint and the time lag compensation step size as the length, based on the time lag compensation step size included in the state correction parameters. The time-series state slice sequence is determined based on all historical compliance state records and their corresponding stability tags in the dynamic backtracking window. Based on the equality comparison of each historical compliance status record in the time-series state slice sequence with the preliminary compliance status of the current frame, the state continuity length is obtained. Based on the relationship between the state continuity length and the time lag compensation step size, a time-series validity identifier is determined. Wherein, when the state continuity length is greater than or equal to the time lag compensation step size, the time-series validity identifier is valid; otherwise, the time-series validity identifier is invalid. Based on the timing validity identifier and the preliminary compliance status of the current frame, physical reachability is verified using a preset industrial safety state machine transition graph to obtain a state permission instruction. Specifically, when the timing validity identifier is valid and there is a legitimate directed path in the state machine transition graph from the target compliance status of the previous frame, a state permission instruction is generated as a state transition permission instruction; otherwise, a state permission instruction is generated as a state maintenance permission instruction. Based on the state permission instruction and the conflict suppression level identifier included in the state correction parameters, a state consistency verification is performed to obtain the target compliance state.

7. The industrial field wearable compliance testing method based on edge computing according to claim 6, characterized in that, Obtaining the target compliance status includes: If the state permission instruction is a state transition permission instruction, then the current preliminary compliance state is determined as the target compliance state; If the state permission instruction is a state maintenance permission instruction, then when the conflict suppression level indicator indicates strong suppression, a search is performed based on the historical state record corresponding to the dynamic backtracking window, and the most recent compliance state marked as high stability is determined as the target compliance state; when the conflict suppression level indicator indicates non-strong suppression, the target compliance state of the previous frame is determined as the target compliance state of the current frame.

8. An industrial field wearable compliance inspection system based on edge computing, characterized in that, The edge computing-based industrial field wearable compliance testing method as described in any one of claims 1 to 7 is applied; the edge computing-based industrial field wearable compliance testing system comprises: An adaptive region extraction module is used to determine an adaptive search region containing the target's clothing parts based on the pose geometry information of the target person in the current video frame, and to extract the current video frame based on the adaptive search region to obtain a local region image. The single-frame dual-index generation module is used to perform feature reasoning based on the local region image to obtain the texture confidence index and the spatial location index, and to generate the instantaneous compliance evidence value and feature semantic matching index of the current frame based on the texture confidence index and the spatial location index. The time lag filtering module is used to perform time lag filtering based on the instantaneous compliance evidence value and the compliance status information of historical frames to obtain the preliminary compliance status of the current frame. Based on the time stability characteristics of the preliminary compliance status within a preset time window and the mapping relationship with the feature semantic matching index, the status correction parameters are determined. The semantic joint correction module is used to correct the preliminary compliance state based on the state correction parameters to obtain the target compliance state, and generate a structured compliance detection result based on the target compliance state.

9. An electronic device, characterized in that, include: Memory, used to store computer software programs; A processor is configured to read and execute the computer software program, wherein when the processor executes the computer software program, it implements the edge computing-based industrial field wearable compliance testing method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements the edge computing-based industrial field wearable compliance testing method as described in any one of claims 1 to 7.

Images