Method for detecting wearing condition of safety helmets in hoisting operation area

Abstract

The invention relates to a method for detecting the wearing condition of safety helmets in a hoisting operation area, and belongs to the crossing field of safety monitoring and artificial intelligence target detection. The method comprises the following steps: determining a video monitoring point location and angle and a hoisting area; positive and negative samples are collected, a labeling tool is used for labeling the collected pictures, and then data enhancement is carried out; training by adopting a YOLO-nas algorithm to obtain a model for detecting a lifting hook, a safety helmet and a person; the model is used for analyzing the real-time video, a detection result is obtained, and the working state of the hoisting area is judged; counting the number of the safety helmets worn in the hoisting area in work; and safety rules are preset according to the safety management specifications, the safety state of each hoisting area is judged according to the preset safety rules, and a multi-state prompting mode is adopted to give out a prompt. The method is high in detection precision, high in stability, high in scene adaptability, more comprehensive in supervision dimension and high in practicability and operability.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of safety monitoring and artificial intelligence target detection, specifically relating to a method for detecting whether safety helmets are worn in hoisting operation areas. Background Technology

[0002] Under the safety operation standard management system, to ensure the personal safety of hoisting personnel, all personnel entering the hoisting operation area must wear safety helmets as required. Currently, the management methods adopted in the industry are mainly divided into the following three categories: on-site inspections by supervisory personnel, manual inspections using video surveillance, and automatic detection of helmet wearing status by personnel within the monitored area using artificial intelligence technology.

[0003] Among them, the two methods of management relying on personnel (on-site inspection and manual video inspection) have obvious shortcomings. On the one hand, due to the limitations of human energy and attention span, it is impossible to achieve full-process, high-accuracy supervision of the operation, and safety hazards are easily overlooked due to fatigue, negligence and other factors. On the other hand, for most of the time, the content of the monitoring screen does not change significantly, and the effective information obtained by manual monitoring accounts for a very low percentage, resulting in a serious waste of human resources.

[0004] Current AI-based automated detection solutions also suffer from insufficient adaptability to the specific safety requirements of hoisting operations. Hoisting operation safety regulations clearly stipulate that only personnel within a designated area below the hoisting operation must wear safety helmets; personnel outside this area are not required to do so. Furthermore, helmet use is only mandatory after the hoisting operation has officially begun. However, existing automated helmet detection algorithms can only determine whether personnel are wearing helmets, failing to differentiate between the work area and whether the operation is in progress. This leads to a disconnect between the detection results and actual safety regulations, failing to fully meet the safety supervision needs of hoisting operation scenarios. Summary of the Invention

[0005] (a) Technical problems to be solved The technical problem to be solved by this invention is how to provide a method for detecting the wearing status of safety helmets in hoisting operation areas, so as to solve the problem that the existing automatic safety helmet wearing detection algorithm can only determine whether personnel are wearing safety helmets, but cannot distinguish the scope of the operation area and whether the operation is in progress, resulting in the detection results being out of touch with actual safety specifications and failing to fully meet the safety supervision needs of hoisting operation scenarios.

[0006] (II) Technical Solution To address the aforementioned technical problems, this invention proposes a method for detecting the wearing status of safety helmets in hoisting operation areas, the method comprising the following steps: S1. Determine the video surveillance locations and angles, as well as the mounting area. S2. Establish a hoisting operation dataset: Record videos of actual hoisting operation scenarios and extract images containing hooks, safety helmets of different colors, and workers; at the same time, randomly extract images from videos of non-hoisting operation scenarios as negative samples; use annotation tools to annotate the collected images and then perform data augmentation. S3. The YOLO-NAS algorithm is used to train models for detecting hooks, safety helmets, and people; S4. Use the model to analyze the real-time video, obtain the detection results, and determine the working status of the hoisting area; S5. Statistics on the number of safety helmets worn in the hoisting area during the work; S6. Pre-set safety rules according to safety management standards, determine the safety status of each hoisting area based on the pre-set safety rules, and provide prompts using a multi-status prompt mode.

[0007] (III) Beneficial Effects This invention proposes a method for detecting helmet wearing status in hoisting operation areas. Compared with existing technologies, the method proposed in this invention has the following significant advantages: High detection accuracy and strong stability: Utilizing the YOLO-NAS algorithm, which employs Neural Architecture Search (NAS) technology to select the optimal architecture configuration across the entire search space, achieving higher average precision (mAP). Furthermore, its quantization sensing block and selective quantization technology exhibit a significantly smaller accuracy drop when converted to the INT8 quantization version compared to other similar models, ensuring high detection accuracy even in resource-constrained hardware environments and meeting the demands of real-time analysis of multiple surveillance videos.

[0008] Highly adaptable to various scenarios: It determines whether a lifting operation has started by identifying the position of the hook (when the hook is not in operation, it is in its original position and not in the preset hook area), and accurately divides the lifting area and hook area. Only personnel in the lifting area during operation are checked for wearing safety helmets. This fully meets the safety regulations for lifting operation scenarios and avoids the drawbacks of the "one-size-fits-all" detection of existing algorithms.

[0009] More comprehensive regulatory dimensions: It can not only detect whether personnel are wearing safety helmets, but also identify the color of the safety helmets (red, yellow, blue, white). Combined with preset safety management rules (such as at least technical management personnel wearing red safety helmets and implementation personnel wearing yellow safety helmets), it can determine whether the configuration of the operation personnel is compliant. Compared with the existing methods that only identify whether safety helmets are worn, it can provide richer and more valuable information for actual operation supervision, helping managers to discover potential personnel configuration hazards in a timely manner.

[0010] Highly practical and operable: During the dataset construction phase, it supports strategies such as image rotation, flipping, and mosaic enhancement to compensate for insufficient data volume; during the model training and deployment phase, it provides clear hyperparameter settings (such as an 8:2 ratio of training to test sets, 16 images per batch, SGD optimizer, 150 training rounds, etc.) and video preprocessing workflow (H264 encoded video is decoded frame by frame into RGB images and scaled to 640*640 pixels), facilitating practical operation and application by technical personnel. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the method for detecting the wearing of safety helmets in the hoisting operation area according to the present invention. Detailed Implementation

[0012] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0013] This invention belongs to the interdisciplinary field of safety monitoring and artificial intelligence target detection, specifically involving a method for detecting the wearing status of safety helmets in hoisting operation areas. This method uses the YOLO-NAS target detection algorithm to identify the position of the hook and the wearing status of the safety helmets of the workers. Combined with preset safety management specifications, it can achieve accurate judgment and supervision of the working status of the hoisting operation area.

[0014] The purpose of this invention is to propose a method for detecting helmet wearing in hoisting operation areas, overcoming the shortcomings of existing technologies. Specifically, this method addresses the following issues: First, it accurately identifies whether hoisting operations are in progress, avoiding unnecessary helmet checks on personnel in non-operational areas; second, it precisely delineates hoisting operation areas, monitoring helmet wearing only within those areas, conforming to scenario-based safety regulations; third, it not only determines whether personnel are wearing helmets but also identifies helmet colors, combining this with preset rules to determine whether personnel configuration meets requirements, providing more comprehensive and accurate support for hoisting operation supervision and meeting actual safety management needs.

[0015] The present invention proposes a method for detecting the wearing status of safety helmets in hoisting operation areas, comprising the following steps: S1. Determine the video surveillance locations and angles, as well as the mounting area. S2. Establish a hoisting operation dataset: Record videos of actual hoisting operation scenarios and extract images containing hooks, safety helmets of different colors, and workers; at the same time, randomly extract images from videos of non-hoisting operation scenarios as negative samples; use annotation tools to annotate the collected images and then perform data augmentation. S3. The YOLO-NAS algorithm is used to train models for detecting hooks, safety helmets, and people; S4. Use the model to analyze the real-time video, obtain the detection results, and determine the working status of the hoisting area; S5. Statistics on the number of safety helmets worn in the hoisting area during the work; S6. Pre-set safety rules according to safety management standards, determine the safety status of each hoisting area based on the pre-set safety rules, and provide prompts using a multi-status prompt mode.

[0016] Example 1: This invention proposes a method for detecting the wearing of safety helmets in hoisting operation areas. Through a process of "area division - dataset construction - model training - real-time detection - status judgment," it achieves precise monitoring of the safety status of hoisting operation areas. The specific implementation steps are as follows: 1. Determine the location and angle of video surveillance points and the mounting area. (1) Selection of monitoring points and angles: The height of the monitoring points needs to be measured on-site to ensure that the angle between the monitoring equipment and the common position of the hook during the lifting operation area (such as the key position during the lifting, translation and lowering process) is greater than 15°, so as to avoid the inability to accurately judge the actual position of the hook in the two-dimensional monitoring video due to the angle being too small, which would affect the accuracy of the judgment of the operation status.

[0017] (2) Area coordinate division: Using rectangular boxes, each hoisting operation area and its corresponding hook area above it are delineated, and the coordinates of the upper left and lower right vertices of each rectangular area are recorded (based on the pixel coordinates of the monitoring video screen). Since the hook activity area is located at a certain height above the ground, it does not overlap with the hoisting area on the ground in the two-dimensional monitoring screen. Therefore, it is necessary to clearly distinguish the two areas and establish a one-to-one correspondence (i.e., one hook area corresponds to one hoisting operation area).

[0018] 2. Establish a hoisting operation dataset. (1) Data collection: By recording videos of actual hoisting operations, images containing hooks, safety helmets of different colors (red, yellow, blue, and white) and workers are extracted from them; at the same time, images are randomly extracted from videos of non-hoisting operations as negative samples to improve the model's ability to resist interference from non-target scenarios.

[0019] (2) Data annotation: Use professional annotation tools (such as LabelImg) to annotate the collected images. The annotation categories include hooks, red safety helmets, yellow safety helmets, blue safety helmets, white safety helmets, and people. Ensure that the annotation box of each target accurately surrounds the target area and that the annotation information is complete and accurate.

[0020] (3) Data augmentation: If the amount of raw data collected is insufficient, data augmentation strategies such as image rotation (90°, 180°, 270°), horizontal flipping, vertical flipping, and mosaic enhancement (spoofing local areas of multiple images into a new image) are adopted to expand the scale of the dataset and improve the generalization ability of the model.

[0021] 3. A model for detecting hooks, safety helmets, and people was trained using the YOLO-NAS algorithm. (1) Data partitioning: The completed dataset is divided into training set and test set in a ratio of 8:2. The training set is used for model parameter learning, and the test set is used to evaluate the detection performance of the model.

[0022] (2) Hyperparameter settings: The training batch size was set to 16 images / batch, the stochastic gradient descent (SGD) optimizer was used, and the total number of training rounds was 150. The specific values ​​of the key hyperparameters are as follows: network depth parameter is 0.80, network width parameter is 0.95, initial learning rate is 0.01, stochastic gradient descent momentum parameter is 0.99, and weight decay coefficient is 0.00025, in order to balance the model training speed and detection accuracy.

[0023] (3) Model training and validation: Use the training set to train the YOLO-nas model. After a certain number of training rounds (e.g., 10 rounds), use the test set to evaluate the model's average precision (mAP), recall, and other metrics. Adjust the hyperparameters based on the evaluation results (e.g., increase the weight decay coefficient appropriately when the model is overfitting) until the model performance reaches the preset standard (e.g., mAP ≥ 90%).

[0024] 4. Use the model to analyze the real-time video, obtain the detection results, and determine the working status of the hoisting area. (1) Video preprocessing: decode the real-time monitored H264 encoded video frame by frame into RGB format images; perform preprocessing operations on each RGB image, first scale the image size to 640*640 pixels (consistent with the model training input size), then divide the RGB value of each pixel in the image by 255 and normalize it to the 0-1 range to reduce the impact of data volume on the model inference speed.

[0025] (2) Model inference and hook position judgment: Input the preprocessed image into the trained YOLO-nas model to obtain the detection results including target category, detection box coordinates and confidence; traverse all detected hook targets and calculate the center point coordinates of each hook detection box (center point x-coordinate = (detection box left boundary x-coordinate + right boundary x-coordinate) / 2, center point y-coordinate = (detection box upper boundary y-coordinate + lower boundary y-coordinate) / 2); determine whether the hook center point is within any preset hook area. If it is not within any hook area, determine that the corresponding lifting area of ​​the hook is in a non-working state and skip the subsequent checks; if it is within a hook area, mark the hook area and its corresponding lifting area as "working".

[0026] 5. Statistics on the number of safety helmets worn in the hoisting area during operations. (1) Personnel area judgment: For the hoisting area marked as "working", traverse all personnel targets detected by the model and calculate the coordinates of the center point of the bottom edge of each personnel detection box (the horizontal coordinate of the center point of the bottom edge = (the horizontal coordinate of the left boundary of the detection box + the horizontal coordinate of the right boundary) / 2, and the vertical coordinate = the vertical coordinate of the bottom boundary of the detection box); determine whether the center point is within the corresponding hoisting area, only count personnel targets within the area, and exclude interference from personnel outside the area.

[0027] (2) Safety helmet wearing statistics: For each personnel target in the hoisting area, traverse the detected safety helmet targets and determine whether the center point of the safety helmet detection frame is within the personnel detection frame (if it is, then the personnel is determined to be wearing a safety helmet); according to the safety helmet color (red, yellow, blue, white), count the number of personnel wearing different colored safety helmets in the hoisting area and the number of personnel not wearing safety helmets.

[0028] 6. Pre-set safety rules according to safety management regulations, determine the safety status of each hoisting area based on the pre-set safety rules, and provide prompts using a multi-status prompt mode. (1) Pre-set safety rules: Based on industry safety management standards and the actual needs of hoisting operations, the following judgment rules are set: Safety helmet colors correspond to personnel roles: red safety helmets represent technical management personnel, yellow safety helmets represent implementation personnel, blue safety helmets represent technical guidance personnel, and white safety helmets represent supervisors.

[0029] Non-operational status (hook not in the hook area): If there are personnel in the lifting area who are not wearing safety helmets, immediately issue a "not wearing safety helmet" warning; if all personnel are wearing safety helmets, no additional warning is required.

[0030] Operating status (hook is in the hook area): If any person is not wearing a helmet, issue a "No helmet worn" warning. If all personnel are wearing safety helmets, further assessment of personnel configuration is required: If no one is wearing a yellow safety helmet, a "lack of personnel" message will be issued. If the number of people wearing red safety helmets is 0, a "Lack of technical management personnel" message will be issued; If the number of people wearing yellow and red safety helmets is greater than 0, and the number of people wearing white or blue safety helmets is greater than 0, a "Supervisor / Technical Instructor Present" alert will be issued. If the number of people wearing both yellow and red safety helmets is greater than 0, and there are no other prompts, it is considered a normal working condition.

[0031] (2) Multi-state prompt mode: The mechanism of "one prompt is triggered when a condition is met" is adopted, which supports the simultaneous output of multiple status prompts (such as the simultaneous issuance of "not wearing a safety helmet" and "lacking technical management personnel" prompts in a certain hoisting area). This ensures that management personnel can fully grasp the safety hazards in the work area.

[0032] Compared with existing technologies, the method for detecting helmet wearing in hoisting operation areas proposed in this invention has the following significant advantages: High detection accuracy and strong stability: Utilizing the YOLO-NAS algorithm, which employs Neural Architecture Search (NAS) technology to select the optimal architecture configuration across the entire search space, achieving higher average precision (mAP). Furthermore, its quantization sensing block and selective quantization technology exhibit a significantly smaller accuracy drop when converted to the INT8 quantization version compared to other similar models, ensuring high detection accuracy even in resource-constrained hardware environments and meeting the demands of real-time analysis of multiple surveillance videos.

[0033] Highly adaptable to various scenarios: It determines whether a lifting operation has started by identifying the position of the hook (when the hook is not in operation, it is in its original position and not in the preset hook area), and accurately divides the lifting area and hook area. Only personnel in the lifting area during operation are checked for wearing safety helmets. This fully meets the safety regulations for lifting operation scenarios and avoids the drawbacks of the "one-size-fits-all" detection of existing algorithms.

[0034] More comprehensive regulatory dimensions: It can not only detect whether personnel are wearing safety helmets, but also identify the color of the safety helmets (red, yellow, blue, white). Combined with preset safety management rules (such as at least technical management personnel wearing red safety helmets and implementation personnel wearing yellow safety helmets), it can determine whether the configuration of the operation personnel is compliant. Compared with the existing methods that only identify whether safety helmets are worn, it can provide richer and more valuable information for actual operation supervision, helping managers to discover potential personnel configuration hazards in a timely manner.

[0035] Highly practical and operable: During the dataset construction phase, it supports strategies such as image rotation, flipping, and mosaic enhancement to compensate for insufficient data volume; during the model training and deployment phase, it provides clear hyperparameter settings (such as an 8:2 ratio of training to test sets, 16 images per batch, SGD optimizer, 150 training rounds, etc.) and video preprocessing workflow (H264 encoded video is decoded frame by frame into RGB images and scaled to 640*640 pixels), facilitating practical operation and application by technical personnel.

[0036] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting the wearing status of safety helmets in a hoisting operation area, characterized in that, The method includes the following steps: S1. Determine the video surveillance locations and angles, as well as the mounting area. S2. Establish a hoisting operation dataset: Record videos of actual hoisting operation scenarios and extract images containing hooks, safety helmets of different colors, and workers; at the same time, randomly extract images from videos of non-hoisting operation scenarios as negative samples; use annotation tools to annotate the collected images and then perform data augmentation. S3. The YOLO-NAS algorithm is used to train models for detecting hooks, safety helmets, and people; S4. Use the model to analyze the real-time video, obtain the detection results, and determine the working status of the hoisting area; S5. Statistics on the number of safety helmets worn in the hoisting area during the work; S6. Pre-set safety rules according to safety management standards, determine the safety status of each hoisting area based on the pre-set safety rules, and provide prompts using a multi-status prompt mode.

2. The method for detecting helmet wearing status in a hoisting operation area as described in claim 1, characterized in that, Determining the video monitoring points and angles includes: the height of the monitoring points needs to be calculated on-site to ensure that the angle between the monitoring equipment and the common position of the hook during the hoisting operation is greater than 15°, so as to avoid the inability to accurately determine the actual position of the hook in the two-dimensional monitoring video due to the angle being too small, which would affect the accuracy of the judgment of the operation status.

3. The method for detecting helmet wearing status in a hoisting operation area as described in claim 1, characterized in that, Determining the hoisting area includes: using rectangular boxes to mark each hoisting operation area and its corresponding hook area above it, and recording the coordinates of the upper left and lower right vertices of each rectangular area, based on the pixel coordinates of the monitoring video image; since the hook activity area is located at a certain height above the ground, it does not overlap with the hoisting area on the ground in the two-dimensional monitoring image, so it is necessary to clearly distinguish the two areas and establish a one-to-one correspondence, that is, one hook area corresponds to one hoisting operation area.

4. The method for detecting helmet wearing status in a hoisting operation area as described in claim 1, characterized in that, In step S2, the annotation of the collected images using the annotation tool includes: using the annotation tool to annotate the collected images, with annotation categories including hooks, red safety helmets, yellow safety helmets, blue safety helmets, white safety helmets, and people, ensuring that the annotation box of each target accurately surrounds the target area and that the annotation information is complete and accurate.

5. The method for detecting helmet wearing status in a hoisting operation area as described in claim 1, characterized in that, In S2, data augmentation includes: if the amount of raw data collected is insufficient, image rotation, horizontal flipping, vertical flipping, and mosaic enhancement are used to expand the dataset size.

6. The method for detecting helmet wearing status in a hoisting operation area as described in any one of claims 1-5, characterized in that, S3 includes: Data partitioning: The completed dataset is divided into training and test sets according to the proportion. The training set is used for model parameter learning, and the test set is used to evaluate the detection performance of the model. Hyperparameter settings: Set the training batches and total training epochs, and use the stochastic gradient descent optimizer; key hyperparameters include: network depth parameter, network width parameter, initial learning rate, stochastic gradient descent momentum parameter, and weight decay coefficient, to balance model training speed and detection accuracy; Model training and validation: The YOLO-nas model is trained using the training set. After each training cycle, the average precision and recall of the model are evaluated using the test set. Hyperparameters are adjusted based on the evaluation results until the model performance reaches the preset standard.

7. The method for detecting helmet wearing status in a hoisting operation area as described in claim 6, characterized in that, S4 includes: Video preprocessing: The real-time monitored H264 encoded video is decoded frame by frame into RGB format images; each RGB image is preprocessed by first scaling the image size to match the model training input size, and then normalizing the RGB value of each pixel in the image to the 0-1 range. Model inference and hook position determination: The preprocessed image is input into the trained YOLO-NAS model to obtain detection results including target category, detection box coordinates, and confidence score; all detected hook targets are traversed, and the center point coordinates of each hook detection box are calculated; it is determined whether the hook center point is within any preset hook area. If it is not within any hook area, the lifting area corresponding to the hook is determined to be in a non-working state, and subsequent checks are skipped; if it is within a hook area, the hook area and its corresponding lifting area are marked as "working".

8. The method for detecting helmet wearing status in a hoisting operation area as described in claim 7, characterized in that, S5 includes: Personnel area determination: For hoisting areas marked as "working", traverse all personnel targets detected by the model, calculate the coordinates of the center point of the bottom edge of each personnel detection box; determine whether the center point is within the corresponding hoisting area, only count personnel targets within the area, and exclude interference from personnel outside the area; Safety helmet wearing statistics: For each personnel target in the hoisting area, iterate through the detected safety helmet targets and determine whether the center point of the safety helmet detection frame is within the personnel detection frame. If it is, the personnel is determined to be wearing a safety helmet. According to the safety helmet color classification, count the number of personnel wearing different colored safety helmets in the hoisting area and the number of personnel not wearing safety helmets.

9. The method for detecting helmet wearing status in a hoisting operation area as described in claim 8, characterized in that, In S6, the preset safety rules are judgment rules set according to industry safety management standards and combined with the actual needs of hoisting operations, including: Safety helmet colors correspond to personnel roles: red safety helmets represent technical management personnel, yellow safety helmets represent implementation personnel, blue safety helmets represent technical guidance personnel, and white safety helmets represent supervisors. In non-operational status, i.e., when the hook is not in the hook area: If there are personnel in the lifting area who are not wearing safety helmets, immediately issue a "No safety helmet" warning; if all personnel are wearing safety helmets, no additional warning is required. Operating status, i.e., the hook is in the hook area: If any person is not wearing a helmet, issue a "No helmet" warning. If all personnel are wearing safety helmets, further assessment of personnel configuration is required: If the number of people wearing yellow safety helmets is 0, a "lack of personnel" message will be issued. If the number of people wearing red safety helmets is 0, a "lack of technical management personnel" message will be issued; If the number of people wearing yellow and red safety helmets is greater than 0, and the number of people wearing white or blue safety helmets is greater than 0, a "Supervisor / Technical Instructor is present" prompt will be issued. If the number of people wearing both yellow and red safety helmets is greater than 0 and there are no other prompts, it is considered a normal working condition.

10. The method for detecting helmet wearing status in a hoisting operation area as described in claim 9, characterized in that, The provision of prompts using a multi-state prompt mode includes: adopting a mechanism of "one prompt is triggered when a condition is met", supporting the simultaneous output of multiple state prompts to ensure that managers can fully grasp the safety hazards in the work area.

Images