Construction personnel unsafe behavior identification and early warning method and system based on improved YOLOv11

Abstract

The invention provides a construction personnel unsafe behavior identification and early warning method and system based on improved YOLOv11, and belongs to the technical field of building construction safety risk identification, and the method comprises the steps: obtaining to-be-processed construction scene video data; and processing the obtained to-be-processed construction scene video data by using a pre-trained identification model to obtain an unsafe behavior identification result of the construction personnel. The problems that traditional construction safety monitoring depends on manpower, the real-time performance is poor, the accuracy rate is low, and the small target detection capacity is weak are solved, efficient and intelligent technical support is provided for construction safety risk early warning, and a new path for construction safety management is developed.

Description

Technical Field

[0001] This invention relates to the field of construction safety risk identification technology, specifically to a method and system for identifying and warning of unsafe behaviors of construction workers based on an improved YOLOv11. Background Technology

[0002] Construction sites for buildings and transportation infrastructure are characterized by complex working environments, frequent cross-operations, high personnel mobility, and dense risk points. Unsafe behaviors of construction workers (such as not wearing safety measures, smoking, using mobile phones, close contact with machinery, and entering restricted areas) are the main causes of accidents.

[0003] The survey revealed that construction safety monitoring mainly relies on manual inspections and traditional video playback checks, which has significant shortcomings: First, it lacks real-time performance, as manual inspections cannot cover all scenarios, and the investigation can only be conducted after an accident occurs, making timely warnings impossible; second, it has low accuracy, relying on the experience and judgment of staff, which is easily affected by fatigue and subjective factors, resulting in high rates of missed and false detections, especially with weak identification capabilities for small targets (cigarettes, mobile phones); third, it lacks specificity, as existing technologies are mostly applicable to general scenarios and have weak ability to identify specific unsafe behaviors in different construction scenarios (such as close contact with machinery, intrusion into restricted areas); fourth, it has poor alarm stability, as dynamic event judgment is easily affected by detection jitter and boundary fluctuations, causing alarm flashing; and fifth, it is inefficient, requiring a large amount of manpower for large-scale construction sites, resulting in high monitoring costs.

[0004] While existing computer vision technologies have achieved some success in identifying abnormal behavior in fields such as transportation and security, construction scenarios present unique challenges, including variable lighting, severe occlusion, complex behavior patterns, and dense clusters of small targets. Furthermore, there is a lack of dedicated datasets and multi-algorithm fusion solutions for the four core risk scenarios, making it difficult to directly adapt existing technologies. Therefore, there is an urgent need for a method to identify unsafe behaviors of construction workers that is precise, efficient, and provides stable alarms, addressing the pain points of traditional monitoring systems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying and warning of unsafe behaviors of construction workers based on an improved YOLOv11, so as to solve at least one of the technical problems existing in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for identifying and issuing early warnings of unsafe behaviors of construction workers based on an improved YOLOv11, comprising:

[0008] Acquire the construction scene video data to be processed;

[0009] A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

[0010] As a further limitation of the first aspect of the present invention, an unsafe behavior dataset is constructed for different scenarios, and the identification objects and judgment rules are defined as follows: combining the typical risks of the construction site and the data feasibility, unsafe behaviors are divided into two categories: static violations and dynamic risk events; static violations refer to the violation state or behavior that can be judged by visual cues at a given moment, and four types of static violations are selected: not wearing a safety helmet, not wearing a reflective vest / safety vest, smoking, and playing with a mobile phone;

[0011] Let It be the input video frame at time t; the detector outputs a set of detection results:

[0012] ;

[0013] in Represents the bounding box. Indicates category, This represents the confidence score. Indicates the number of targets detected;

[0014] Static violations are judged at the personnel level; for each individual's personnel boundary frame... By using spatial association rules, security-related objects are grouped together. Assigned to this person; association rules include: IoU Either the center point of the object is located within the personnel frame, or the object frame is contained within the personnel frame.

[0015] As a further limitation of the first aspect of the present invention, the realization of static target features of construction workers based on the improved YOLOv11 model includes: adopting a single-stage Backbone-Neck-Head design architecture; wherein, the Backbone is responsible for extracting multi-scale features, the Neck enhances multi-scale representation through feature fusion, and the Head outputs classification and bounding box regression results at multiple scales; wherein...

[0016] To reduce redundant computations in the backbone, a lightweight module called C3k2-Faster is proposed by integrating partial convolutions: given input features The passage is divided into Each convolutional channel and There are 1 identity mapping channel, among which The output formula is:

[0017] ;

[0018] in This indicates channel concatenation; an optional 1×1 convolution is used for feature fusion / alignment.

[0019] As a further limitation of the first aspect of the present invention, the behavior recognition achieved by integrating YOLO object detection and ByteTrack tracking technologies includes:

[0020] ByteTrack is used for multi-target tracking within a Kalman prediction and Hungarian matching framework. Low-confidence detection results are further utilized to improve trajectory continuity in occluded or crowded scenes. Target trajectories are predicted using Kalman filtering, and a two-stage association matching process addresses target loss due to occlusion, generating continuous and stable movement trajectories for construction workers.

[0021] Dynamic events rely on temporally continuous trajectory information; the trajectory of worker k is represented as:

[0022] ;

[0023] in It is the bounding box tracked at time t; using reference points. To define event constraints.

[0024] The Adam optimizer was selected as the strategy for adjusting model parameters, and cross-entropy was selected as the loss function to judge the model's performance on the samples.

[0025] As a further limitation of the first aspect of the present invention, the system integrates static target features, attitude features, and trajectory features, stabilizes the alarm through time smoothing and hysteresis mechanisms, and outputs unsafe behavior identification results and classification information, including:

[0026] The early warning mechanisms for the two types of dynamic risk events are as follows:

[0027] Machine proximity risk: This risk characterizes the unsafe exposure state of a worker when too close to a machine / vehicle; for the worker k and the machine / vehicle m, let their reference points at time t be respectively and Its distance is defined as:

[0028] ;

[0029] If distance Duration reaches at least If the frame is cleared, a proximity event alarm is triggered; if the distance is... Duration reaches at least If the frame is cleared, the alarm is deactivated; to maintain robustness across different viewpoints and scales, the threshold is configured with scale-aware normalization.

[0030] As a further definition of the first aspect of this invention, restricted area intrusion: The dangerous / restricted area is defined as a polygon. If the reference point Duration reaches at least If a frame is detected, an intrusion event alarm will be triggered; if Duration reaches at least If the frame is cleared, the alarm is deactivated;

[0031] The time smoothing and hysteresis strategy used for alarm stabilization detects jitter, occlusion, and boundary lingering that can cause frequent risk state transitions, resulting in alarm flashing. First, time smoothing is applied to the instantaneous risk decision, followed by a hysteresis update rule to determine the alarm state. :

[0032] ;

[0033] in, and These are the minimum durations for triggering and clearing alarms, respectively, and the introduction of time "forbidden zones" helps suppress flickering caused by short-term fluctuations.

[0034] Secondly, this invention provides a construction worker unsafe behavior identification and early warning system based on an improved YOLOv11, comprising:

[0035] The acquisition module is used to acquire video data of the construction scene to be processed;

[0036] The processing module utilizes a pre-trained recognition model to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the recognition model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers under different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; fusing static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the unsafe behavior identification results and classification information.

[0037] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11 as described in the first aspect.

[0038] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11 as described in the first aspect.

[0039] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the construction worker unsafe behavior identification and early warning method based on the improved YOLOv11 as described in the first aspect.

[0040] The beneficial effects of this invention are: it solves the problems of traditional construction safety monitoring, such as reliance on manual labor, poor real-time performance, low accuracy, and weak small target detection capabilities, and provides efficient and intelligent technical support for construction safety risk early warning, thus opening up a new path for construction safety management.

[0041] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating the identification process for unsafe behaviors of construction workers as described in an embodiment of the present invention.

[0044] Figure 2 This is a flowchart illustrating the video image data preprocessing process described in an embodiment of the present invention.

[0045] Figure 3 This is a diagram of the improved YOLOv11 algorithm architecture described in an embodiment of the present invention.

[0046] Figure 4 This is a diagram of the ByteTrack algorithm framework as described in an embodiment of the present invention.

[0047] Figure 5 This is a flowchart illustrating the implementation of the early warning mechanism described in an embodiment of the present invention. Detailed Implementation

[0048] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0049] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0050] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0051] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0052] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0053] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0054] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0055] This invention relates to the field of construction safety management technology, and in particular to a method for identifying unsafe behaviors of construction workers that integrates advanced model algorithms such as deep learning. It is applicable to real-time safety monitoring and risk warning for behaviors involving small target detection in different construction scenarios, such as not wearing a safety helmet, smoking, and dynamic risk scenarios such as approaching large machinery.

[0056] This invention provides a method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11, relating to the fields of construction safety management and computer vision technology. The method includes the following steps: multi-dimensional preprocessing of video images of construction scenes to optimize image quality and small target recognition in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm to identify static violations (not wearing a safety helmet, not wearing a reflective vest, smoking, using a mobile phone); fusing improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract the movement trajectory of construction workers and identify dynamic risk events (close contact with machinery, entering restricted areas); and outputting unsafe behavior identification results and classification information by weighted fusion of static target features, posture features, and trajectory features, and introducing time smoothing and hysteresis mechanisms to stabilize alarms. This invention solves the problems of traditional construction safety monitoring, such as reliance on manual labor, poor real-time performance, low accuracy, and weak small target detection capabilities. It provides efficient and intelligent technical support for construction safety risk early warning and opens up a new path for construction safety management.

[0057] Example 1

[0058] In this embodiment 1, a construction worker unsafe behavior identification and early warning system based on improved YOLOv11 is first provided, including: an acquisition module for acquiring construction scene video data to be processed; and a processing module for processing the acquired construction scene video data using a pre-trained identification model to obtain the identification results of construction worker unsafe behaviors. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of construction worker unsafe behaviors under different construction scenarios, defining identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting unsafe behavior identification results and classification information.

[0059] In this embodiment, using the aforementioned system, a method for identifying and issuing early warnings of unsafe behaviors of construction workers based on the improved YOLOv11 is implemented, including:

[0060] Acquire the construction scene video data to be processed;

[0061] A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

[0062] like Figure 1As shown, specifically, the steps include: Step 1: Preprocessing video images of construction scenes to optimize image quality in complex environments; Step 2: Constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; Step 3: Extracting features of construction workers and safety-related targets (safety helmets, reflective vests, cigarettes, mobile phones) based on the improved YOLOv11 algorithm; Step 4: Integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; Step 5: Integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting unsafe behavior identification results and classification information.

[0063] like Figure 2 As shown, step 1 specifically involves employing multi-dimensional image preprocessing technology to address the characteristics of construction scenarios, such as variable lighting, dust interference, equipment and building material obstruction, and cluttered backgrounds.

[0064] 1) Image denoising: Mean filtering, median filtering and Gaussian filtering algorithms are applied to remove high-frequency noise caused by construction dust and mechanical vibration, and smooth image texture; data augmentation parameter settings are shown in Table 1.

[0065] 2) Image enhancement: Improve the recognition of human body contours and motion details through contrast enhancement and adaptive histogram equalization algorithms, and optimize the feature extraction effect in complex backgrounds;

[0066] 3) Occlusion Repair: An occlusion area repair algorithm is adopted, which utilizes the continuity of surrounding pixel texture and pose to complete the joints and motion features of construction workers that are occluded by scaffolding, building materials or other people.

[0067] 4) Illumination correction: Adjust image brightness through grayscale stretching and color normalization to eliminate uneven illumination caused by direct strong light and shadow occlusion, and ensure the consistency of feature extraction under different lighting conditions;

[0068] 5) Data augmentation: Perform random rotations [-10°, 10°], ±0.1 horizontal / vertical offsets, horizontal flips, and scaling operations of 0.8-1.2 times on the images to expand the diversity of training data and reduce the risk of model overfitting.

[0069] Table 1 Data Augmentation Parameter Settings

[0070]

[0071] Specifically, step 2 involves constructing datasets of unsafe behaviors in different scenarios and defining the objects to be identified and the judgment rules:

[0072] Based on typical risks at construction sites and data availability, unsafe behaviors are divided into two main categories: static violations and dynamic risk events.

[0073] Static violations refer to violations that can be determined at a given moment through visual cues, characterized by being "judgmentable in a single frame and continuously fusionable." Four types of static violations were selected: not wearing a safety helmet, not wearing reflective clothing / safety vest, smoking, and using a mobile phone.

[0074] Let It be the input video frame at time t. The detector outputs a set of detection results:

[0075]

[0076] in Represents the bounding box. Indicate the category (e.g., person, safety helmet, reflective vest, cell phone, cigarette). This represents the confidence score. This indicates the number of targets to be detected.

[0077] Static violations are determined at the person level. For each person's personnel boundary... By using spatial association rules, security-related objects are grouped together. Assigned to this person. Association rules include: IoU Either the object's center point is located within the person's box (center-in-box), or the object's box is contained within the person's box (as an auxiliary condition).

[0078] Dynamic risk events emphasize the risk exposure of personnel over time and the changing spatial relationship with hazards, characterized by the need for "trajectory and duration constraints." Two typical types of dynamic risk events are defined as follows:

[0079] 1) Proximity risk: When personnel and machinery / vehicles are too close to each other in space for a certain period of time, it poses a potential risk of collision or turning radius loss.

[0080] 2) Entering a dangerous / restricted area: If a person's trajectory enters a predefined dangerous or restricted area (ROI) and remains there for a certain period of time, it constitutes a risk of entering a restricted area.

[0081] The summary of unsafe behaviors and the rules for their determination are shown in Table 2 below.

[0082] Table 2 Summary of Unsafe Behaviors and Judgment Rules

[0083]

[0084] like Figure 3 , Figure 4As shown, step 3 specifically involves realizing the static target features of construction workers based on the improved YOLOv11 model:

[0085] A single-stage Backbone-Neck-Head architecture is adopted. The Backbone extracts multi-scale features, the Neck enhances multi-scale representation through feature fusion, and the Head outputs classification and bounding box regression results at multiple scales. To address specific challenges in construction sites (such as limited computational resources, small safety targets, and strong background interference), targeted improvements are made to the Backbone and Neck without altering the single-stage detection paradigm.

[0086] 1) Backbone Improvement: To reduce redundant computations in the backbone, a lightweight module called C3k2-Faster is proposed by integrating partial convolutions (PConv). Given input features... The passage is divided into Each convolutional channel and There are 1 identity mapping channel, among which The output formula is:

[0087]

[0088] in This indicates channel cascading. An optional 1×1 convolution is used for feature fusion / alignment. In practice, this design reduces the effective convolution cost to approximately the order of r² while maintaining feature expressiveness.

[0089] 2) Neck Improvement: A Detail-Aware Attention Neck Network (DAA-Neck) is introduced to enhance fine-grained representations. It includes a Coordinate Attention (CA) branch and a Detail Reconstruction Unit (DRU) branch. The former aggregates features along the horizontal and vertical directions, generating position-sensitive weights to reweight the fused features; the latter uses pixel rearrangement (sub-pixel rearrangement) to recover local high-frequency details weakened by downsampling / fusion. Enhanced features are output through channel concatenation and 1×1 convolution, strengthening feature extraction for small targets (cigarettes, mobile phones) and occluded scenes; outputting bounding box coordinates, categories, and confidence scores for construction workers, safety helmets, reflective vests, cigarettes, and mobile phones.

[0090] The training parameters are set as shown in Table 3 below.

[0091] Table 3 Training Parameter Settings

[0092]

[0093] Specifically, step 4 involves integrating YOLO object detection and ByteTrack tracking technologies to achieve behavior recognition.

[0094] ByteTrack is used for multi-target tracking within a Kalman prediction and Hungarian matching framework, and further leverages low-confidence detection results to improve trajectory continuity in occluded or crowded scenes. Target trajectories are predicted using Kalman filtering, and a two-stage association matching process (high-confidence detection boxes matched with the trajectory IOU, and low-confidence detection boxes matched with the remaining trajectory) is employed to address target loss due to occlusion, generating continuous and stable movement trajectories for construction workers.

[0095] Dynamic events rely on temporally continuous trajectory information; the trajectory of worker k is represented as:

[0096]

[0097] in This is the bounding box tracked at time t. Using a reference point... Use the bottom center or geometric center of the bounding box to define event constraints.

[0098] The Adam optimizer is chosen as the strategy for model parameter tuning, and cross-entropy is selected as the loss function to judge the model's performance on the samples. The update method of the Adam optimizer is shown in the following equation:

[0099]

[0100] In the formula, For learning rate, First-order momentum, It is a second-order momentum. The coefficients that make the denominator stable.

[0101] The cross-entropy loss function plays a crucial role in classification tasks, guiding model optimization by measuring the difference between the model's predicted probability distribution and the true label distribution. Its calculation method for multi-class classification tasks is shown in the following equation:

[0102]

[0103] In the formula, Number of categories; is the sign function, which takes the value 1 when the true class of sample i is c, and 0 otherwise; Let be the probability that sample i belongs to category c.

[0104] Specifically, step 5 involves fusing static target features, attitude features, and trajectory features, stabilizing the alarm through time smoothing and hysteresis mechanisms, and outputting the unsafe behavior identification results and classification information.

[0105] like Figure 5 As shown, to suppress alarm flickering caused by detection jitter and boundary hovering, temporal smoothing and hysteresis strategies can be applied to event decision-making. Specific early warning mechanisms for two types of dynamic risk events are as follows:

[0106] 1) Proximity-to-machinery risk: This risk characterizes the unsafe exposure of workers when they are too close to machinery / vehicles. For worker k and machinery / vehicle m, let their reference points at time t be respectively... and Its distance is defined as:

[0107]

[0108] If distance Duration reaches at least If the frame is cleared, a proximity event alarm is triggered; if the distance is... Duration reaches at least If the frame is cleared, the alarm is deactivated. To maintain robustness across different viewpoints and scales, the threshold is configured with scale-aware normalization.

[0109] 2) Restricted-area intrusion: Define the hazardous / restricted area (ROI) as a polygon. If the reference point (Within the area) the duration reaches at least If a frame is detected, an intrusion event alarm will be triggered; if (Outside the area) the duration reaches at least If the frame is cleared, the alarm is deactivated.

[0110]

[0111] ROI can be obtained through lightweight manual annotation or defined according to site specifications. A duration-based triggering and releasing strategy is employed to ensure alarm stability. Temporal smoothing and hysteresis strategies for alarm stabilization detect jitter, occlusion, and boundary hovering, which can cause frequent risk state transitions, resulting in alarm flickering. Temporal smoothing (moving average or majority voting) is first applied to the instantaneous risk decision, followed by a hysteresis update rule to determine the alarm state. :

[0112]

[0113] in, and These are the minimum durations for triggering and clearing alarms, respectively, and the introduction of a time "dead zone" helps suppress flickering caused by short-term fluctuations.

[0114] Example 2

[0115] This embodiment 2 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the above-described method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11. The method includes:

[0116] Acquire the construction scene video data to be processed;

[0117] A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

[0118] Example 3

[0119] This embodiment 3 provides a computer device, including a memory and a processor. The processor and the memory communicate with each other. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the above-described method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11. The method includes:

[0120] Acquire the construction scene video data to be processed;

[0121] A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

[0122] Example 4

[0123] This embodiment 4 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions to implement the above-described method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11, the method including:

[0124] Acquire the construction scene video data to be processed;

[0125] A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

[0126] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and issuing early warnings of unsafe behaviors of construction workers based on an improved YOLOv11, characterized in that, include: Acquire the construction scene video data to be processed; A pre-trained identification model is used to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the identification model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers in different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; integrating static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the identification results and classification information of unsafe behaviors.

2. The method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11 according to claim 1, characterized in that: To construct a dataset of unsafe behaviors in different scenarios, the identification objects and judgment rules are defined as follows: combining typical risks at construction sites and data feasibility, unsafe behaviors are divided into two categories: static violations and dynamic risk events; static violations refer to violations that can be judged by visual cues at a given moment, and four types of static violations are selected: not wearing a safety helmet, not wearing a reflective vest / safety vest, smoking, and using a mobile phone. Let It be the input video frame at time t; the detector outputs a set of detection results: ； in Represents the bounding box. Indicates category, This represents the confidence score. Indicates the number of targets detected; Static violations are judged at the personnel level; for each individual's personnel boundary frame... By using spatial association rules, security-related objects are grouped together. Assigned to this person; association rules include: IoU Either the center point of the object is located within the personnel frame, or the object frame is contained within the personnel frame.

3. The method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11 according to claim 1, characterized in that, The static target features of construction workers implemented based on the improved YOLOv11 model include: adopting a single-stage Backbone-Neck-Head design architecture; where the Backbone is responsible for extracting multi-scale features, the Neck enhances multi-scale representation through feature fusion, and the Head outputs classification and bounding box regression results at multiple scales; To reduce redundant computations in the backbone, a lightweight module called C3k2-Faster is proposed by integrating partial convolutions: given input features The channel is divided into Each convolutional channel and There are 1 identity mapping channel, among which The output formula is: ； in This indicates channel concatenation; an optional 1×1 convolution is used for feature fusion / alignment.

4. The method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11 according to claim 1, characterized in that, The integration of YOLO object detection and ByteTrack tracking technologies to achieve behavior recognition includes: ByteTrack is used for multi-target tracking within a Kalman prediction and Hungarian matching framework. Low-confidence detection results are further utilized to improve trajectory continuity in occluded or crowded scenes. Target trajectories are predicted using Kalman filtering, and a two-stage association matching process addresses target loss due to occlusion, generating continuous and stable movement trajectories for construction workers. Dynamic events rely on temporally continuous trajectory information; the trajectory of worker k is represented as: ； in It is the bounding box tracked at time t; using reference points. To define event constraints; The Adam optimizer was selected as the strategy for adjusting model parameters, and cross-entropy was selected as the loss function to judge the model's performance on the samples.

5. The method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11 according to claim 1, characterized in that, By integrating static target features, attitude features, and trajectory features, and stabilizing alarms through time smoothing and hysteresis mechanisms, the system outputs unsafe behavior identification results and classification information, including: The early warning mechanisms for the two types of dynamic risk events are as follows: Machine proximity risk: This risk characterizes the unsafe exposure state of a worker when too close to a machine / vehicle; for the worker k and the machine / vehicle m, let their reference points at time t be respectively and Its distance is defined as: ； If distance Duration reaches at least If the frame is cleared, a proximity event alarm is triggered; if the distance is... Duration reaches at least If the frame is cleared, the alarm is deactivated; to maintain robustness across different viewpoints and scales, the threshold is configured with scale-aware normalization.

6. The method for identifying and warning of unsafe behaviors of construction workers based on improved YOLOv11 according to claim 5, characterized in that, Restricted Area Intrusion: Define the danger / restricted area as a polygon. If the reference point Duration reaches at least If the frame is not cleared, an intrusion event alarm will be triggered; if Duration reaches at least If the frame is cleared, the alarm is deactivated; The time smoothing and hysteresis strategy used for alarm stabilization detects jitter, occlusion, and boundary lingering that can cause frequent risk state transitions, resulting in alarm flashing. First, time smoothing is applied to the instantaneous risk decision, followed by a hysteresis update rule to determine the alarm state. : ； in, and These are the minimum durations for triggering and clearing alarms, respectively, and the introduction of time "forbidden zones" helps suppress flickering caused by short-term fluctuations.

7. A construction worker unsafe behavior identification and early warning system based on an improved YOLOv11, characterized in that, include: The acquisition module is used to acquire video data of the construction scene to be processed; The processing module utilizes a pre-trained recognition model to process the acquired construction scene video data to obtain the identification results of unsafe behaviors of construction workers. The training of the recognition model includes: preprocessing construction scene video images to optimize image quality in complex environments; constructing datasets and feature sets of unsafe behaviors of construction workers under different construction scenarios, defining the identification objects and judgment rules; extracting features of construction workers and safety-related targets based on the improved YOLOv11 algorithm; integrating improved YOLOv11 target detection and ByteTrack multi-target tracking technology to extract trajectory features of construction workers and identify static violations and dynamic risk events; fusing static target features, posture features, and trajectory features, stabilizing alarms through time smoothing and hysteresis mechanisms, and outputting the unsafe behavior identification results and classification information.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11 as described in any one of claims 1-6.

9. A computer device, characterized in that, The method includes a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions that can be executed by the processor, and the processor calling the program instructions to execute the method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11 as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions that implement the method for identifying and warning of unsafe behaviors of construction workers based on the improved YOLOv11 as described in any one of claims 1-6.

Images