Intelligent standard dressing identification method based on target detection and human body key points

Abstract

The invention relates to the technical field of artificial intelligence application, and discloses a standard dressing intelligent identification method based on target detection and human body key points, and the method comprises the steps: carrying out the frame skipping human body detection of a monitoring video stream, extracting the human body key points, and carrying out the screening according to a preset confidence coefficient threshold value, determining a head area, an upper body area and a lower body area based on the screened high-confidence-coefficient key points; when the head area and the upper body area exist at the same time, if the safety helmet and the work clothes are detected at the same time, and when and only when all the following conditions are met at the same time, it is judged that the user is normatively worn: the ratio of the width of the minimum head external rectangle to the width of a safety helmet recognition frame is within a first preset threshold range, and the distance between the center points of the minimum head external rectangle and the width of the safety helmet recognition frame is smaller than a second preset value; the ratio of the height of the minimum upper body external rectangle to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-parallel ratio of the two is larger than a fourth preset value. According to the invention, the condition that the person is partially shielded can be more effectively handled, and each main part of the human body can be accurately positioned.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence application technology, and in particular to a method for intelligently identifying standardized clothing based on target detection and key points of the human body. Background Art

[0002] In many industrial production, construction, energy extraction, and specialized operations scenarios, workers are often required to wear appropriate workwear, such as hard hats, overalls, protective pants, and masks, to ensure their personal safety. Despite these clear regulations, some workers still fail to wear these uniforms due to carelessness, complacency, or a desire for convenience. This undoubtedly increases the risk of accidents and poses a potential threat to human life and corporate property.

[0003] To effectively monitor the compliance of workwear, the traditional reliance on manual inspections is not only inefficient and costly, but also difficult to achieve all-weather, full-coverage, real-time monitoring. Therefore, the use of computer vision technology for automated workwear identification has become a research hotspot and industry demand. Existing workwear detection methods based on image recognition typically use target detection algorithms to directly detect specific workwear, such as helmets and work clothes, in images. However, in actual work scenarios, the environment is often complex, for example: 1. Crowded and obstructed environments: When multiple workers are close together or obstruct each other, simple target detection algorithms struggle to accurately and correctly associate detected workwear (such as a hard hat) with a specific worker. This can easily lead to worker A's hard hat being misidentified as worker B, or the workwear or person not being fully detected due to obstruction.

[0004] 2. Insufficient recognition accuracy: Under conditions of severe occlusion, changing lighting, and changing postures of people, the recognition accuracy of existing algorithms will drop significantly, making it difficult to meet the high-precision requirements of practical applications.

[0005] 3. Interference from close proximity: Even if the work clothes and the human body can be detected, if there is no effective association mechanism, it is easy to be interfered with by the work clothes worn by the adjacent personnel, resulting in misjudgment.

[0006] Therefore, how to improve the accuracy and robustness of workwear recognition in complex environments such as crowded and obstructed environments, and achieve precise matching of workwear with corresponding personnel, is an urgent problem to be solved in the current technical field. Summary of the Invention

[0007] In order to solve the problems existing in the above-mentioned prior art, the technical solutions provided by the present invention include: The intelligent recognition method of standard clothing based on target detection and human key points includes the following steps: S1. Perform frame-skipping human body detection on the surveillance video stream, extract human key points and filter them according to a preset confidence threshold, and determine the head area, upper body area, and lower body area based on the filtered high-confidence key points; S2. When both the head area and the upper body area are present, perform wear detection. S21. If both a helmet and work clothes are detected, the wearer is deemed to be wearing the work clothes properly if and only if all of the following conditions are met simultaneously; otherwise, the wearer is deemed to be wearing the work clothes improperly: S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet identification frame is within a first preset threshold range, and the distance between the center points of the two is less than a second preset value; S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-and-union ratio of the two is greater than a fourth preset value.

[0008] Preferably, step S2 further includes: stopping the process when either the head area or the upper body area is missing.

[0009] Preferably, step S2 further includes: S22. If a helmet, work clothes, and work pants are detected simultaneously, the wear is determined to be standard if and only if all of the following conditions are met simultaneously; otherwise, the wear is determined to be irregular: S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet identification frame is within a first preset threshold range, and the distance between the center points of the two is less than a second preset value; S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-over-union ratio of the two is greater than a fourth preset value; S223. The ratio of the height of the minimum lower body circumscribed rectangle determined by the key points of the lower body area to the height of the work pants identification frame is within the fifth preset threshold range, and the intersection-and-union ratio of the two is greater than the sixth preset value.

[0010] Preferably, step S1 includes: S11 performs frame skipping on the surveillance video stream to obtain an image frame, and performs human body detection on the image frame to obtain a human body detection frame; S12. For each human body within the human body detection frame, extract multiple human key points and the confidence level corresponding to each human key point; S13. Filtering the human body key points according to a preset confidence threshold to retain the target human body key points with high confidence; S14. Determine the head region, upper body region, and lower body region of the human body based on the target human body key points.

[0011] Preferably, the head area is determined by four key points: the left eye, the right eye, and the left ear, the right ear; The upper body area is determined by four key points: the left shoulder, the right shoulder, and the left hip, the right hip; The lower body area is determined by four key points: the left and right hips and the left and right ankles.

[0012] Preferably, if the number of key points in a determined area is insufficient, it is determined that the area does not exist.

[0013] Preferably, the first preset threshold and the third preset threshold include 0.8-1.2; the second preset value includes half of the width of the helmet identification frame.

[0014] Preferably, the fourth preset value includes 0.7.

[0015] Preferably, the sixth preset value includes 0.7; the fifth preset threshold includes 0.9-1.1.

[0016] Beneficial effects 1. Improved the accuracy of workwear recognition in complex scenarios: The present invention determines the head area, upper body area, and lower body area based on high-confidence key points, which can more effectively deal with situations where people are partially obscured. Even in complex backgrounds or in the presence of interference objects, it can relatively accurately locate the main parts of the human body, providing a reliable spatial reference for subsequent workwear matching.

[0017] 2. Precisely matching tooling with personnel is achieved, effectively eliminating interference from adjacent personnel: The minimum bounding rectangle of each body region, determined by high-confidence key points, is rigorously geometrically verified against the tooling identified by the object detection algorithm. This ensures that the detected tooling truly belongs to the person being analyzed, not to nearby personnel. This significantly reduces misjudgments caused by crowded or close-knit groups of people.

[0018] 3. Enhanced algorithm adaptability and judgment integrity for different wear combinations: This invention not only handles the basic "hardhat + work clothes" wear scenario, but also further handles more complete wear scenarios including "hardhat + work clothes + work pants." By setting judgment conditions for different workwear combinations, the algorithm can flexibly adapt to the specific wear requirements in different working environments and provide a more comprehensive assessment of compliance.

[0019] 4. Taking into account recognition efficiency: By skipping frames in the surveillance video stream, the number of image frames that need to be processed is reduced while ensuring recognition accuracy, thereby reducing the overall computing load, helping to improve the system's real-time processing capabilities and meet actual monitoring needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flow chart of a method for intelligently identifying standard clothing based on target detection and key points of the human body provided in a preferred embodiment of the present invention; Figure 2 A schematic diagram of wear detection results is provided in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings. In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or components referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0022] Example 1 like Figure 1 As shown, this embodiment provides a method for intelligently identifying standard clothing based on target detection and key points of the human body, including the following steps: S1. Perform frame-skipping human body detection on the surveillance video stream, extract human key points and filter them according to the preset confidence threshold, and determine the head area, upper body area and lower body area based on the filtered high-confidence key points.

[0023] Surveillance video streams typically contain a large number of consecutive image frames, with high inter-frame information redundancy. Performing complex human body detection and subsequent analysis on every frame consumes enormous computing resources and makes it difficult to meet real-time requirements. Therefore, the present invention considers employing an optimized frame-skipping strategy. Instead of processing every frame in the video stream, a subset of frames are selected for processing at intervals (for example, sampling every N frames, or dynamically adjusting the sampling frequency based on scene changes).

[0024] Human detection refers to the use of deep learning models (such as the YOLO series, Faster R-CNN, SSD, etc.) or traditional computer vision methods to identify and locate all human targets in the selected image frames, and output the position and range of each detected human body in the form of a rectangular bounding box.

[0025] Human key points typically refer to the locations of human joints (such as the top of the head, neck, shoulders, elbows, wrists, hips, knees, ankles, etc.) and some significant feature points (such as the eyes, ears, and nose). This is usually achieved through specialized human pose estimation algorithms (such as OpenPose, HRNet, AlphaPose, etc.). These algorithms not only provide the position coordinates of key points, but also output a confidence score for each key point, which indicates the algorithm's confidence in the accuracy of the key point's prediction. Due to the complexity of real-world scenes (such as occlusion, blur, lighting changes, and changing postures), the detection results of some key points may be unreliable. Therefore, the present invention considers screening according to a preset confidence threshold, that is, only retaining key points with a confidence level above a preset threshold, and eliminating low-confidence and potentially erroneous detection results to improve the accuracy of subsequent analysis.

[0026] S2. When the head area and upper body area exist at the same time, wear detection is performed. The detection results are as follows: Figure 2 As shown. It should be understood that the wear detection refers to a series of detailed logic used to determine whether the helmet, work clothes, and / or work pants are worn as required. The requirement to simultaneously detect the head area and the upper body area is intended to avoid invalid or erroneous work clothes identification when the body information is incomplete or unreliable. For example, if a person's head area cannot be determined due to severe occlusion, then determining whether they are wearing a helmet is meaningless and may even lead to misidentification due to other objects in the scene. In some preferred embodiments, the process is stopped when either the head area or the upper body area is missing. It should be understood that only when both the head area and the upper body area are present, indicating that the person in the image is not obscured, is subsequent recognition meaningful. In other preferred embodiments, the presence or absence of the lower body area can be included in the determination of process termination, that is, the process is stopped when any of the three areas is missing. This is mainly used in application scenarios where the requirements for the lower body work pants are high. Obviously, those skilled in the art can flexibly design the process termination conditions based on actual site needs. S21. If both a helmet and work clothes are detected, the wear is determined to be standard wear only if all of the following conditions are met simultaneously; otherwise, it is determined to be irregular wear. It should be understood that the presence of the helmet, work clothes, and / or work pants is identified using conventional target detection methods in the art, and this invention does not impose any further limitations on this.

[0027] S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head area to the width of the helmet identification frame is within the first preset threshold range, and the distance between the center points of the two is less than the second preset value. Those skilled in the art will know that, under normal circumstances, the visible width of the helmet worn on the head should be roughly the same as or slightly larger than the width of the head. If the width of the helmet identification frame is much larger or much smaller than the width of the head circumscribed rectangle, it may mean that the helmet is not worn on the person's head, or the detection is incorrect. Furthermore, if the helmet is indeed worn on the person's head, the distance between the center point of the helmet identification frame and the center point of the head circumscribed rectangle should be very small (ideally close to zero). If this distance is too large, it means that the helmet deviates too much from the head position, and it is likely that it is not worn by the person, or the wearing method is extremely irregular.

[0028] S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection over union ratio of the two is greater than a fourth preset value. The visible height of a properly worn work clothes in the image should roughly match the height of the upper body of the human body. If the height of the work clothes identification frame is much greater or much less than the height of the upper body circumscribed rectangle, it may indicate that the work clothes are not worn by the person, or the detection is incorrect. The intersection over union (IoU) is a commonly used indicator to measure the degree of overlap between two bounding boxes, and its value range is between 0 and 1. Here, the intersection over union ratio of the upper body circumscribed rectangle and the work clothes identification frame is required to be large, indicating that most of the work clothes area overlaps with the upper body area. If the IoU is too small, it means that the work clothes may only be partially draped on the body, or the correspondence with the upper body area is very weak, and it cannot be considered as standard wear.

[0029] The above steps verify the matching of work clothes and human body from two dimensions: size and overlapping area, through dual geometric constraints (width ratio and center point distance, height ratio and intersection ratio). This greatly improves the accuracy and reliability of the matching between work clothes and corresponding personnel and their body parts. It can effectively eliminate interference and misjudgment caused by adjacent personnel, other similar objects in the environment, and improper wearing (such as improper wearing of safety helmets and work clothes), thereby realizing high-precision automatic recognition of standardized wearing behavior of work clothes in complex working scenarios.

[0030] Example 2 like Figure 1 As shown, this embodiment further expands the scope of workwear compliance testing based on the first embodiment, adding the assessment of the wearing condition of work pants. This enables the present invention to adapt to work scenarios that also have clear requirements for lower body protection, providing a more comprehensive workwear compliance assessment capability. Specifically, step S2 also includes: S22. If a safety helmet, work clothes and work pants are detected at the same time, it is determined to be standard wear if and only if all the following conditions are met at the same time, otherwise it is determined to be irregular wear. Based on Example 1, this step sets the premise and rules for standardization judgment of a full set of work clothes including a safety helmet, work clothes and work pants. It can meet the requirements of work scenarios that require full-body protection, such as certain special working environments such as chemical, firefighting, high-voltage electricity, etc. By detecting work pants, the deficiency of focusing only on upper body protection is compensated, which helps to more comprehensively supervise and ensure the safety of workers.

[0031] S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet identification frame is within a first preset threshold range, and the distance between the center points of the two is less than a second preset value.

[0032] S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-and-union ratio of the two is greater than a fourth preset value.

[0033] S223. The ratio of the height of the minimum lower body circumscribed rectangle determined by the key points of the lower body area to the height of the work pants identification frame is within the fifth preset threshold range, and the intersection-over-union ratio of the two is greater than the sixth preset value. It should be noted that the visible height of normally worn work pants in the image should be roughly equivalent to the height of the lower body of the human body. If the height of the work pants detection frame is much greater than or much less than the height of the lower body circumscribed rectangle, it may mean that the work pants are not worn on this person, or the detection is incorrect. Furthermore, the degree of overlap between the work pants and the lower body area is verified by the intersection-over-union ratio. The intersection-over-union ratio of the lower body circumscribed rectangle and the work pants detection frame is required to be large to ensure that the main part of the work pants actually covers the lower body area. If the IoU is too small, the work pants may only be partially draped over the legs, or the correspondence with the lower body area is weak, and cannot be considered as standard wear.

[0034] Example 3 This embodiment describes in detail the specific processing flow used in the present invention to accurately and efficiently locate the human body and delineate its key body regions (head, upper body, and lower body) from a surveillance video stream. This process serves as an important prerequisite and data foundation for the subsequent determination of the conformity of workwear (as described in Examples 1 and 2). Its core lies in ensuring that the obtained human body region information is both accurate and reliable through a series of carefully designed sub-steps. It should be understood that this embodiment is provided as a preferred option based on Example 1. Those skilled in the art may employ other methods to locate the human body and delineate its key body regions based on existing techniques, and this embodiment should not be construed as the sole alternative.

[0035] Step S1 includes: S11. Frame skipping is performed on the surveillance video stream to obtain image frames, and human body detection is performed on these image frames to obtain human body detection frames. The video generated by the surveillance system is composed of continuous image frames, typically at a high frame rate. To reduce the computational load and improve processing efficiency while ensuring effective recognition, the present invention first performs frame skipping on the original video stream. This means that the system does not analyze every frame, but instead selectively samples every other frame. For example, a fixed frame skipping interval can be set (e.g., processing one frame every five frames), or a more intelligent dynamic frame skipping strategy can be adopted (e.g., adjusting the sampling frequency based on the magnitude of scene motion, sampling more frequently during intense motion and less frequently during gentle motion). Frame skipping can significantly reduce the amount of image data requiring subsequent processing. Furthermore, for the recognition of most workwear wearing states, since the wearing state changes relatively slowly, appropriate frame skipping does not lose critical information. The result is a series of discrete image frames for subsequent analysis.

[0036] S12. For each human body within the human body detection frame, extract multiple human key points and the confidence scores corresponding to each human key point. For each human image area within the frame, the system will call a human pose estimation algorithm (such as OpenPose, HRNet, AlphaPose, MediaPipe Pose, etc.) to detect and locate a series of predefined "human key points". These key points usually include the top of the head, neck, shoulders, elbows, wrists, hips, knees, ankles, as well as the eyes, ears, nose, etc. on the face. The number and specific definition of key points can be determined according to the pose estimation algorithm used and the application requirements (for example, the COCO dataset usually defines 17 key points and the MPII dataset defines 16 key points).

[0037] Along with outputting the position coordinates of each keypoint, the pose estimation algorithm also provides a confidence score for each keypoint. This score (typically between 0 and 1) indicates the algorithm's confidence in the accuracy of its predicted position for that keypoint. A high confidence score indicates a high probability that the keypoint was accurately detected, while a low confidence score may indicate that the keypoint is occluded, blurred, or the algorithm failed to accurately identify it.

[0038] S13. Filter the human key points according to the preset confidence threshold to retain the target human key points with high confidence. Only those key points with confidence greater than or equal to the preset threshold are considered "high confidence" and "reliable" and are retained for subsequent body region determination. Key points with confidence lower than the threshold are considered unreliable or misdetected and will be discarded. The setting of this threshold is a trade-off: setting it too high may result in the loss of too many useful key points when there is slight occlusion or poor posture; setting it too low may introduce more erroneous key points, affecting the accuracy of subsequent region division. Therefore, this threshold usually needs to be empirically adjusted or obtained through experimental optimization based on the actual application scenario, the performance of the posture estimation algorithm used, and the accuracy requirements. In some preferred embodiments, the confidence threshold can be set to 0.3.

[0039] S14. Determine the head area, upper body area, and lower body area of the human body based on the target human body key points. Those skilled in the art will appreciate that the human body area can be determined using a minimum bounding rectangle, but more complex shapes such as a convex hull or a polygon can also be used as needed. In some preferred embodiments, the minimum bounding rectangle is a computationally simple and practical option for comparison with a tooling detection frame (usually a rectangle). Specifically, the following are included: The head area is determined by four key points: the left eye, the right eye, and the left ear, the right ear; The upper body area is determined by four key points: the left shoulder, the right shoulder, and the left hip, the right hip; The lower body area is determined by four key points: the left and right hips and the left and right ankles.

[0040] It should be noted that if, after S13 screening, the number of keypoints used to define a specific region is insufficient (for example, defining the head requires at least three high-confidence head-related keypoints, but only one is actually detected), the region is deemed non-existent or unreliable. This enhances the robustness of the method and avoids inferences based on insufficient information.

[0041] Example 4 This embodiment aims to provide a specific set of preset threshold parameters that have been empirically verified or experimentally optimized for use in determining the geometric matching rules between the workwear (safety helmet, work clothes, work pants) and the corresponding body regions (head, upper body, lower body) described in Examples 1, 2, and 3. The selection of these parameters is crucial to ensuring the accuracy and robustness of workwear wear compliance identification. It should be noted that these values are exemplary only. In actual deployment, fine-tuning may be required based on the specific work scenario, monitoring equipment characteristics, workwear type, and tolerance for false alarm and missed alarm rates.

[0042] The first preset threshold is used to determine whether the ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet recognition frame is within a reasonable range, including 0.8-1.2. The lower limit of 0.8 is designed to take into account that the helmet may be slightly smaller than the head (although it is not common, this may be the case with some specially designed lightweight helmets), or that the visual width of the helmet may be slightly larger than the actual width of the head due to factors such as shooting angle and head posture (such as slightly lowering or raising the head, which reduces the visual width of the head). The overall range provides a certain degree of tolerance, allowing a relative difference of approximately 20% between the width of the head and the helmet to accommodate normal variations in actual scenes and slight measurement errors, while also excluding cases where the size mismatch is obvious (for example, an object much smaller or much larger than the head is mistaken for a worn helmet).

[0043] The second preset value is used to determine whether the distance between the center point of the minimum head circumscribed rectangle determined by the key points of the head region and the center point of the helmet identification frame is small enough. Setting it to include half the width of the helmet identification frame means that this is a dynamic threshold that will adaptively adjust according to the size of the currently detected helmet. For larger helmets, the allowable center point offset will be correspondingly larger; for smaller helmets, stricter alignment is required. And if the distance between the two center points is less than half the width of the helmet, it can be roughly understood that the center of the head at least falls within the coverage area of the helmet (or conversely, the center of the helmet falls within the circle with the center of the head as the center and the helmet width as the diameter, although the comparison here is the center of the rectangle). This condition ensures that the head and helmet are well aligned in both horizontal and vertical directions, that is, the helmet is worn straight, not tilted to one side or just resting on the head.

[0044] The third preset threshold is used to determine whether the ratio of the height of the minimum upper body bounding rectangle, determined by the key points of the upper body area, to the height of the work clothes identification frame is within a reasonable range. Settings include 0.8-1.2, and the logic is similar to the first preset threshold range, allowing for a relative difference of approximately 20% between the height of the work clothes identification frame and the height of the upper body bounding rectangle. This takes into account the style of the work clothes (some work clothes may be slightly shorter or longer than the actual upper body), the tightness of the wear, the body posture (such as bending over, which may cause the visual height of the upper body to change), and slight errors in the detection frame. This ensures that the work clothes roughly match the upper body in vertical dimensions, preventing clothing that is too short or too long from being mistaken for compliant work clothes.

[0045] The fourth preset value determines whether the intersection over union (IoU) between the minimum bounding rectangle of the upper body, defined by the key points of the upper body region, and the work garment identification box is sufficiently large. Setting it to 0.7 means that the overlap between the two rectangles accounts for 70% of their union area. This is a relatively high threshold. In the field of object detection, an IoU greater than 0.5 is generally considered a good match, and greater than 0.7 a very good match. This requires that the work garment identification box and the upper body region have a high degree of overlap, ensuring that the work garment is actually worn on the upper body and covers the majority of the upper body area, rather than merely partially touching or draping the body. This helps to prevent cases where the work garment is not properly worn (e.g., only one sleeve is worn, the garment is unzipped or unbuttoned, resulting in incomplete coverage) or that adjacent work garments are incorrectly associated.

[0046] The fifth preset threshold is used to determine whether the ratio of the height of the minimum lower body circumscribed rectangle determined by the key points of the lower body area to the height of the work pants identification box is within a reasonable range. The settings include 0.9-1.1, which is a narrower range than the 0.8-1.2 for work clothes. This is because the length of work pants is relatively fixed (such as long pants), and their coverage must match the natural length of the lower body more closely. In addition, in specific application scenarios, there are stricter requirements for the length standardization of work pants. For example, trouser legs that are too short or too long are not allowed. Therefore, a relative difference of approximately 10% between the height of the work pants identification box and the height of the lower body circumscribed rectangle is allowed. This ensures that the work pants match the lower body height in the vertical dimension, especially preventing the misjudgment of three-quarter pants or nine-quarter pants (if they are specified as long pants) as compliant, or the pants being stacked too long.

[0047] The sixth preset value is used to determine whether the intersection over union (IoU) between the minimum bounding rectangle of the lower body, determined by the key points of the lower body region, and the work pants identification frame is sufficiently large. The logic behind this is the same as the fourth preset value and will not be further elaborated here.

[0048] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent recognition method for standard clothing based on target detection and key points of the human body, characterized by: Including steps: S1. Perform frame-skipping human body detection on the surveillance video stream, extract human key points and filter them according to a preset confidence threshold, and determine the head area, upper body area, and lower body area based on the filtered high-confidence key points; S2. When both the head area and the upper body area are present, perform wear detection. S21. If both a helmet and work clothes are detected, the wearer is deemed to be wearing the work clothes properly if and only if all of the following conditions are met simultaneously; otherwise, the wearer is deemed to be wearing the work clothes improperly: S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet identification frame is within a first preset threshold range, and the distance between the center points of the two is less than a second preset value; S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-and-union ratio of the two is greater than a fourth preset value.

2. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: Step S2 further includes: stopping the process when either the head region or the upper body region is missing.

3. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: Step S2 further includes: S22. If a helmet, work clothes, and work pants are detected simultaneously, the wear is determined to be standard if and only if all of the following conditions are met simultaneously; otherwise, the wear is determined to be irregular: S211. The ratio of the width of the minimum head circumscribed rectangle determined by the key points of the head region to the width of the helmet identification frame is within a first preset threshold range, and the distance between the center points of the two is less than a second preset value; S212. The ratio of the height of the minimum upper body circumscribed rectangle determined by the key points of the upper body region to the height of the work clothes identification frame is within a third preset threshold range, and the intersection-over-union ratio of the two is greater than a fourth preset value; S223. The ratio of the height of the minimum lower body circumscribed rectangle determined by the key points of the lower body area to the height of the work pants identification frame is within the fifth preset threshold range, and the intersection-and-union ratio of the two is greater than the sixth preset value.

4. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: Step S1 includes: S11 performs frame skipping on the surveillance video stream to obtain an image frame, and performs human body detection on the image frame to obtain a human body detection frame; S12. For each human body within the human body detection frame, extract multiple human key points and the confidence level corresponding to each human key point; S13. Filtering the human body key points according to a preset confidence threshold to retain the target human body key points with high confidence; S14. Determine the head region, upper body region, and lower body region of the human body based on the target human body key points.

5. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: The head area is determined by four key points: the left eye, the right eye, and the left ear, the right ear; The upper body area is determined by four key points: the left shoulder, the right shoulder, and the left hip, the right hip; The lower body area is determined by four key points: the left and right hips and the left and right ankles.

6. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 5, characterized in that: If the number of key points in a region is insufficient, the region is deemed to be non-existent.

7. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: The first preset threshold and the third preset threshold include 0.8-1.2; the second preset value includes half of the width of the helmet identification frame.

8. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 1, characterized in that: The fourth preset value includes 0.

7.

9. The method for intelligently identifying standard clothing based on target detection and key points of the human body as claimed in claim 3, characterized in that: The sixth preset value includes 0.7; the fifth preset threshold includes 0.9-1.1.

Images