A real-time high-precision detection method and system for construction safety based on environmental characteristics

Abstract

The present invention relates to the field of video recognition technology, specifically a real-time, high-precision detection method and system for construction safety based on environmental characteristics. A variety of sensors are arranged in the area to be detected to collect on-site environmental data in real time; the illumination changes in the area to be detected are analyzed using a light and shadow analysis model, the light-safety misjudgment index is calculated, and potential light misjudgment areas are identified; worker videos are analyzed using a visual processing model to identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors; the local breeze and vibration detection model is used to detect the local wind speed and turbulence conditions, vibration, and resonance phenomena in the area to be detected in real time, and evaluate the local environmental stability index; the light-safety misjudgment index, the personnel behavior safety index, and the local environmental stability index are used to calculate the safety level of the area to be detected in real time, thereby achieving a multi-dimensional, real-time construction safety risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of video recognition technology, and in particular to a real-time high-precision detection method and system for construction safety based on environmental characteristics. Background Art

[0002] Working at heights is a high-risk activity in construction, and construction workers face numerous safety threats from environmental factors, equipment operation, and personal behavior. In recent years, the application of image processing technology in construction safety monitoring has gained widespread attention, particularly in real-time monitoring and dynamic safety detection. By installing cameras or sensor devices to capture real-time images and data from the construction site, image processing technology can quickly identify potential safety issues within the scene. For example, computer vision algorithms can automatically identify whether construction workers are wearing protective equipment or monitor hazardous sources in the construction environment, such as falling objects and unreinforced structures. Traditional image processing technology can identify safety hazards at construction sites to a certain extent by analyzing images.

[0003] However, the application of existing image processing technology in construction safety monitoring still has certain limitations. Traditional image processing techniques for monitoring construction safety rarely consider the impact of light and shadow. Light variations can easily lead to visual misjudgments or depth perception errors among workers (for example, misidentifying missing steps or guardrails). Furthermore, existing height work safety monitoring focuses solely on overall wind speed and fails to consider the impact of localized sudden wind forces on workers' physical stability.

[0004] Therefore, a real-time and high-precision detection method and system for construction safety based on environmental characteristics is proposed. Summary of the Invention

[0005] The present invention aims to provide a real-time, high-precision construction safety detection method and system based on environmental characteristics, enabling high-precision detection of construction safety at heights. The method involves deploying multiple sensors in the area to be detected to collect real-time on-site environmental data; analyzing illumination changes in the area to be detected using a light and shadow analysis model, calculating a light-safety misjudgment index, and identifying areas with potential light misjudgment; utilizing a visual processing model to identify key points of the human skeleton in real time, obtain a human behavior safety index, and detect potentially dangerous behaviors; utilizing a local breeze and vibration detection model to detect local wind speed, turbulence, vibration, and resonance in the area to be detected in real time, and assessing the local environmental stability index; and utilizing the light-safety misjudgment index, the human behavior safety index, and the local environmental stability index to calculate the safety level of the area to be detected in real time, generate a safety detection report, and trigger a safety warning mechanism.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A real-time, high-precision construction safety detection method based on environmental characteristics, comprising:

[0008] Deploy multiple sensors in the area to be inspected to collect real-time on-site environmental data, including lighting data, worker videos, wind speed data, humidity data, air flow data, and vibration data;

[0009] Analyzing lighting changes in the area to be inspected using a light and shadow analysis model based on the lighting data and the worker video, calculating a light-safety misjudgment index, and identifying potential light misjudgment areas;

[0010] Analyze the worker video using a visual processing model to identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors;

[0011] The local wind speed and turbulence conditions, vibration and resonance phenomena of the area to be detected are detected in real time by using a local breeze and vibration detection model to evaluate the local environmental stability index;

[0012] The light-safety misjudgment index, the personnel behavior safety index and the local environment stability index are used to calculate the safety level of the area to be inspected in real time, generate a safety inspection report and trigger a safety early warning mechanism.

[0013] Preferably, the multiple sensors include: a light sensor, a high-definition camera, a temperature and humidity sensor, a wind speed sensor, and an acceleration sensor;

[0014] The lighting data includes natural light intensity, artificial lighting brightness, lighting direction, shadow distribution and lighting uniformity parameters;

[0015] The worker video includes a high-definition video sequence of the worker working in the area to be inspected;

[0016] The wind speed data includes the local wind speed vector, the maximum instantaneous wind speed and the wind direction change trend of the area to be detected;

[0017] The humidity data includes relative humidity of air and ground humidity of the area to be detected;

[0018] The air flow data includes local airflow direction, small-scale turbulence intensity and disturbance frequency;

[0019] The vibration data includes the vibration amplitude, vibration frequency, vibration acceleration and structural resonance frequency of the area to be detected.

[0020] Preferably, the light and shadow analysis model includes: a light extraction unit, a shadow recognition unit, a dynamic light change tracking unit and a light misjudgment assessment unit;

[0021] The illumination extraction unit extracts a pixel brightness histogram and an illumination gradient distribution in the preprocessed image frame of the worker video;

[0022] The shadow recognition unit separates the dynamic lighting occlusion shadows and the static structure occlusion shadows in the area to be detected, and generates a shadow mask map;

[0023] The dynamic illumination change tracking unit analyzes illumination mutation trends of consecutive image frames according to the pixel brightness histogram, illumination gradient distribution and the shadow mask map;

[0024] The light misjudgment assessment unit integrates the shadow mask map with the construction path and the work edge area, assesses the risk probability of visual illusion caused by illumination, outputs the light-safety misjudgment index of each area in the area to be detected, and marks the area where the light-safety misjudgment index is greater than a preset misjudgment threshold as the potential light misjudgment area.

[0025] Preferably, the visual processing model includes: a human skeleton key point recognition unit, a posture evaluation unit and an abnormal behavior detection unit;

[0026] The human skeleton key point recognition unit processes the image frames of the worker video through a posture estimation model based on deep learning, extracts the coordinates of the human body key points and generates a human skeleton structure diagram;

[0027] The posture evaluation unit calculates the angle change, center of gravity offset and motion trajectory curve between the skeleton points according to the human skeleton structure diagram to identify the posture information of the person;

[0028] The abnormal behavior detection unit compares the personnel posture information with a dangerous behavior database, detects potential dangerous behaviors in combination with the humidity data, and outputs a personnel behavior safety index.

[0029] Preferably, the local breeze and vibration detection model includes: a local wind speed and turbulence analysis unit, a vibration and resonance phenomenon analysis unit and a local environment stability assessment unit;

[0030] The local wind speed and turbulence condition analysis unit identifies local sudden wind events and small-scale turbulence characteristics based on the wind speed data and the air flow data;

[0031] The vibration and resonance phenomenon analysis unit compares the vibration data with the natural frequency of the construction structure to identify vibration and resonance phenomena;

[0032] The local environment stability assessment unit combines the local sudden wind event, small-scale turbulence characteristics, and the vibration and resonance phenomenon to output the local environment stability index.

[0033] Preferably, the calculation process of the security level includes:

[0034] Inputting the light-safety misjudgment index, the human behavior safety index, and the local environment stability index into a risk assessment model to generate a quantitative risk level score, including low risk, medium risk, high risk, and extremely high risk, to obtain the safety level of the area to be inspected;

[0035] The risk assessment model dynamically adjusts the weight of each index based on construction type, operation time, wind force level and historical safety accident data.

[0036] Preferably, a real-time high-precision construction safety detection system based on environmental characteristics comprises:

[0037] The data acquisition module is used to deploy multiple sensors in the area to be inspected to collect on-site environmental data, including lighting, worker videos, wind speed, humidity, air flow, and vibration;

[0038] a light and shadow analysis module, configured to analyze the lighting changes in the area to be detected using a light and shadow analysis model according to the lighting, calculate a light-safety misjudgment index, and identify potential visual illusion areas;

[0039] A visual recognition module is used to analyze the worker video using a visual processing model, identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors;

[0040] A local breeze and vibration detection module is used to detect the local wind speed and turbulence conditions of the to-be-detected area and the vibration and resonance phenomena of the to-be-detected platform in real time through a local breeze and vibration detection model, and evaluate the local environmental stability index;

[0041] The safety monitoring and early warning module is used to use the light-safety misjudgment index, the personnel behavior safety index and the local environment stability index to calculate the safety level of the area to be detected in real time, generate a safety detection report and trigger a safety early warning mechanism.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. This invention introduces the concept of a "light-safety misjudgment index" into construction safety monitoring. Using a light and shadow analysis model, it jointly models lighting data and worker videos to identify areas of visual illusion caused by uneven lighting, shadow obstruction, and strong light reflections. This model effectively addresses the problem of traditional safety detection systems neglecting "optical hazards." This model can effectively reduce accidents such as stepping on air and misjudgment of edges caused by light misjudgment in scenarios such as bridge construction and high-altitude operations, thereby improving safety detection capabilities in light and shadow scenarios.

[0044] 2. This invention utilizes human posture recognition and abnormal behavior detection technology based on a visual processing model, eliminating the traditional reliance on wearable devices (such as accelerometers, gyroscopes, etc.) and / or RFID tags. Using only worker video data, it accurately extracts key skeletal points and analyzes behavioral characteristics. This model significantly expands the system's adaptability, making it suitable for work environments such as high-altitude and confined spaces where wearing equipment is inconvenient. Furthermore, combined with a deep learning behavioral recognition model, it can detect dangerous movements such as slips, excessive bending, and imbalances in real time, effectively mitigating the risks associated with operator violations and improving the accuracy of construction safety monitoring during high-altitude operations.

[0045] 3. The present invention establishes a joint modeling mechanism of local micro-airflow and platform vibration at the aerial construction site through the local breeze and vibration detection model based on a high-precision wind speed sensor array and a ground structure vibration sensor. This model can not only monitor micro-scale disturbances such as turbulence, sudden winds, and wind direction deflection in real time, but also provide early warning judgment on whether the aerial construction platform is in a critical resonance state. Compared with the traditional "average wind speed" judgment mode, this method has higher accuracy and more timely response. It is particularly suitable for early detection of falling risk hazards in sensitive aerial work scenarios such as high-altitude steel structure construction and bridge support construction, effectively improving the detection capability of construction safety issues caused by local micro-airflow and platform vibration in aerial work scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flowchart of a method for real-time, high-precision detection of construction safety based on environmental characteristics provided by an embodiment of the present invention;

[0047] Figure 2 A schematic structural diagram of a real-time, high-precision construction safety detection system based on environmental characteristics provided by an embodiment of the present invention;

[0048] Figure 3 A diagram illustrating the working principle of the light and shadow analysis model provided by an embodiment of the present invention;

[0049] Figure 4 A schematic diagram of the structure of a visual processing model provided by an embodiment of the present invention;

[0050] Figure 5 This is a diagram of the working principle of the local breeze and vibration detection model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Working at heights is a high-risk task in construction, and construction workers face multiple safety threats from environmental factors, equipment operation, and personal behavior. Traditional image processing technologies used to monitor construction safety rarely consider the impact of light and shadow. Light fluctuations can easily lead to visual misjudgments and depth perception errors (for example, misidentifying missing steps or guardrails). Furthermore, existing safety monitoring for work at heights focuses solely on overall wind speed, failing to consider the impact of localized sudden wind forces on workers' physical stability.

[0053] This invention proposes a real-time, high-precision construction safety detection method and system based on environmental characteristics. This method is suitable for high-risk work scenarios such as height work, bridge construction, and steel structure erection. It can comprehensively perceive lighting conditions, human behavior, and environmental disturbances to achieve multi-dimensional, real-time construction safety risk assessment and early warning. To illustrate the effectiveness of the method for real-time, high-precision detection of height work construction safety, the following two examples will illustrate the effectiveness of the invention.

[0054] Example 1

[0055] In the embodiment of the present application, the method proposed by the present invention is used to describe in detail the real-time high-precision detection process of the safety of high-altitude construction work by combining the influence of light and shadow and local sudden wind force. The embodiment of the present application is aimed at the real-time high-precision detection of the safety of high-altitude construction work in a certain construction site A. Figure 1 The content details the real-time and high-precision detection process for the safety of high-altitude construction work; among them, Figure 1 The specific flow chart of the method proposed in the present invention includes: arranging multiple sensors in the area to be detected to collect on-site environmental data in real time; analyzing the illumination changes in the area to be detected through the light and shadow analysis model, calculating the light-safety misjudgment index, and identifying potential light misjudgment areas; using the visual processing model to identify key points of the human skeleton in real time, obtain the personnel behavior safety index and detect potential dangerous behaviors; using the local breeze and vibration detection model to detect the local wind speed and turbulence conditions and vibration and resonance phenomena in the area to be detected in real time, and evaluate the local environmental stability index; using the light-safety misjudgment index, the personnel behavior safety index and the local environmental stability index to calculate the safety level of the area to be detected in real time, generate a safety detection report and trigger a safety warning mechanism. Combined with Figure 1 and Figure 2 The following describes the contents:

[0056] A real-time, high-precision construction safety detection method based on environmental characteristics, comprising:

[0057] Deploy multiple sensors in the area to be inspected to collect real-time on-site environmental data, including lighting data, worker videos, wind speed data, humidity data, air flow data, and vibration data;

[0058] The multiple sensors include: light sensor, high-definition camera, temperature and humidity sensor, wind speed sensor and acceleration sensor;

[0059] The lighting data includes natural light intensity, artificial lighting brightness, lighting direction, shadow distribution and lighting uniformity parameters;

[0060] The worker video includes a high-definition video sequence of the worker working in the area to be inspected;

[0061] The wind speed data includes the local wind speed vector, the maximum instantaneous wind speed and the wind direction change trend of the area to be detected;

[0062] The humidity data includes relative humidity of air and ground humidity of the area to be detected;

[0063] The air flow data includes local airflow direction, small-scale turbulence intensity and disturbance frequency;

[0064] The vibration data includes the vibration amplitude, vibration frequency, vibration acceleration and structural resonance frequency of the area to be detected.

[0065] Specifically, six high-resolution industrial cameras with a resolution of at least 1920×1080 are deployed on the work platform (including edge areas and entrances and exits) to continuously capture worker videos. Each camera uses a structured light depth camera to capture RGB images and depth information. The cameras are connected to the local edge computing server via a wired gigabit network.

[0066] Four ultrasonic anemometers with a frequency of 10 Hz are installed at the four corners of the platform to collect wind speed and air flow data.

[0067] Two multi-channel illumination sensing modules are set on the top of the platform to collect illumination data;

[0068] Three-axis vibration accelerometers are installed at the center of the platform and the support nodes, with a sampling frequency of 1000 Hz, to collect vibration data;

[0069] Two temperature and humidity sensors are arranged in the upper, middle and lower areas of the platform to collect humidity data.

[0070] Through the simultaneous deployment and collection of multi-type, multi-dimensional sensors (video, wind speed, light, humidity, vibration, etc.), comprehensive modeling and real-time perception of the construction site environment can be achieved. Compared with traditional monitoring systems that focus on a single perception dimension, this approach significantly improves the timeliness and spatial coverage of data acquisition. It provides a high-resolution, multi-source, heterogeneous data foundation for subsequent comprehensive perception of lighting conditions, human behavior, and environmental disturbances, enabling multi-dimensional, real-time construction safety risk assessment and early warning. This, in turn, improves the accuracy and response speed of overall construction safety detection and early warning.

[0071] Preferably, Figure 3 As shown, based on the illumination data and the worker video, the illumination changes in the area to be inspected are analyzed by a light and shadow analysis model, the light-safety misjudgment index is calculated, and potential light misjudgment areas are identified;

[0072] The light and shadow analysis model includes: a light extraction unit, a shadow recognition unit, a dynamic light change tracking unit and a light misjudgment assessment unit;

[0073] The illumination extraction unit extracts a pixel brightness histogram and an illumination gradient distribution in the preprocessed image frame of the worker video;

[0074] The shadow recognition unit separates the dynamic lighting occlusion shadows and the static structure occlusion shadows in the area to be detected, and generates a shadow mask map;

[0075] The dynamic illumination change tracking unit analyzes illumination mutation trends of consecutive image frames according to the pixel brightness histogram, illumination gradient distribution and the shadow mask map;

[0076] The light misjudgment assessment unit integrates the shadow mask map with the construction path and the work edge area, assesses the risk probability of visual illusion caused by illumination, outputs the light-safety misjudgment index of each area in the area to be detected, and marks the area where the light-safety misjudgment index is greater than a preset misjudgment threshold as the potential light misjudgment area.

[0077] Specifically, the shadow recognition unit is based on a trained deep convolutional neural network model (such as ShadowNet or U-Net), which takes as input the video image frame and the pixel brightness histogram and illumination gradient distribution obtained by the illumination extraction unit. It performs pixel-level shadow area segmentation.

[0078] Combined with the time series analysis method, it is determined whether the shadow area moves over time or remains unchanged, thereby separating the dynamic lighting occlusion shadow and the static structure occlusion shadow;

[0079] Output a shadow mask map, in which the shadow areas blocked by dynamic lighting and the shadow areas blocked by static structures are marked separately for use by the subsequent lighting change tracking module.

[0080] The dynamic illumination change tracking unit compares the video image frames frame by frame, compares the brightness histogram and the illumination gradient map in the continuous multiple frames, and calculates the illumination change amplitude in the time dimension;

[0081] Perform inter-frame difference on the shadow mask image to analyze the changing trend of shadow area, position and shape over time;

[0082] Introducing optical flow estimation technology to determine whether the light source (such as the sun or construction lights) has relative displacement or brightness mutation;

[0083] Based on the above information, a set of time series feature vectors describing the degree of regional illumination change is output, reflecting whether there are potential visually misleading changes in the area.

[0084] The light misjudgment assessment unit fuses and compares the shadow mask map with the predefined construction path map and the work area boundary map to determine whether shadows or strong reflections cover the critical work path or the edge of the safety guardrail. If so, there is a risk of visual illusion in the area.

[0085] With reference to the human eye vision model (including contrast sensitivity, edge blur recognition ability, etc.), the probability of construction workers making misjudgments is evaluated, and the light-safety misjudgment index of each area to be inspected is output; if the light-safety misjudgment index of a certain area exceeds the preset misjudgment threshold (such as 0.8), it will be marked as a potential light misjudgment area for subsequent risk synthesis assessment and early warning system triggering.

[0086] Table 1 shows the efficiency of illumination misjudgment area recognition.

[0087] Table 1 Illumination misjudgment area recognition efficiency

[0088] ;

[0089] This embodiment proposes a light and shadow analysis model based on image brightness, shadow recognition, and time series analysis. It can intelligently detect visual risks arising from changing lighting conditions, such as misidentification of steps and holes due to shadow occlusion. This model addresses the shortcomings of traditional monitoring systems in complex lighting scenarios, enabling a quantitative assessment of light-induced safety misjudgment risks, and enhancing its practicality and intelligence in real-world high-altitude construction conditions.

[0090] Preferably, Figure 4 As shown, the worker video is analyzed using a visual processing model to identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors;

[0091] The visual processing model includes: a human skeleton key point recognition unit, a posture evaluation unit and an abnormal behavior detection unit;

[0092] The human skeleton key point recognition unit processes the image frames of the worker video through a posture estimation model based on deep learning, extracts the coordinates of the human body key points and generates a human skeleton structure diagram;

[0093] The posture evaluation unit calculates the angle change, center of gravity offset and motion trajectory curve between the skeleton points according to the human skeleton structure diagram to identify the posture information of the person;

[0094] The abnormal behavior detection unit compares the personnel posture information with a dangerous behavior database, detects potential dangerous behaviors in combination with the humidity data, and outputs a personnel behavior safety index.

[0095] Specifically, the human skeleton key point recognition unit includes: a person target detection layer, a skeleton key point detection layer, a key point coordinate post-processing layer and a perspective fusion recognition layer;

[0096] The human object detection layer detects the worker subjects in each frame using a human object detection model, preferably using a lightweight or high-precision detection network such as YOLOv7 or Faster R-CNN, to identify the worker subjects in the image and obtain their bounding box coordinates. The detected bounding box area is cropped to obtain an independent image region for each human subject.

[0097] The skeletal keypoint detection layer uses a deep learning pose estimation model to identify skeletal keypoints in the extracted human image regions. This deep learning pose estimation model, which can utilize advanced network structures such as HRNet, OpenPose, ViTPose, and DETR-Pose, predicts the coordinates of each worker's skeletal keypoints (head, shoulders, elbows, wrists, hips, knees, ankles, and feet, a total of 25 points) and outputs a confidence score for each keypoint. If the confidence score of a keypoint falls below a set confidence threshold (e.g., 0.7), the system deems it unusable data and compensates for the missing data or ignores it in post-processing.

[0098] After obtaining the preliminary recognition results, the key point coordinate post-processing layer performs post-processing on the key point coordinates; maps the two-dimensional points in the image coordinate system to three-dimensional space to obtain the spatial skeleton point coordinates in the actual high-altitude working environment; smoothes the skeleton point positions between frames through methods such as Kalman filtering and exponentially weighted sliding average; calculates the connection vector and angle relationship between adjacent key points, and constructs a complete human skeleton structure diagram for motion evaluation.

[0099] The perspective fusion recognition layer fuses multi-camera perspective image information based on image splicing, feature fusion or spatial reconstruction to enhance recognition accuracy.

[0100] The posture assessment unit constructs a bone vector between each pair of key points in the human skeleton structure diagram, and uses the cosine angle formula to calculate the angle between the bones and the angle change between the bone points;

[0101] The center of gravity offset is calculated using the coordinates of the key points of the whole body to calculate the center of mass position of the human body, and then combined with the trajectory of the center of gravity in the time series to determine whether there is a significant offset or imbalance tendency;

[0102] The trajectory of the skeleton key point coordinates in several consecutive frames is modeled, and the human motion curve is constructed using polynomial fitting or spline interpolation. Dynamic behaviors such as bending, jumping, and sudden squatting are identified to obtain the motion trajectory curve.

[0103] Based on the above angle changes, center of gravity shift and motion curve changes, a personnel behavior safety index calculation function is constructed to quantify the stability of the personnel posture; the formula for the personnel behavior safety index is:

[0104] ;

[0105] in, It is the personnel behavior safety index; Adjust the weight for the angle change term; is the total number of skeleton key points; For the The key points of the skeleton are The angle of the frame; is the inter-frame interval; The reference angle fluctuation threshold of the key point of the skeleton within the normal range; Adjust the weight for the center of gravity offset term; For the The coordinates of the human body's center of gravity in the frame; The reference center of gravity coordinates of the human body when standing normally; is the maximum acceptable center of gravity offset distance; For the The velocity sequence of key points; Adjust the weight for the degree of motion curve change; For the The standard deviation of the velocity sequence of each skeleton key point; It is the standard value of speed fluctuation for reference smooth motion;

[0106] Grading personnel behavior safety index to detect potential dangerous behaviors:

[0107] like : Stable posture, no potentially dangerous behavior;

[0108] like : There may be slight posture instability and potential dangerous behavior;

[0109] like : The posture is unstable and there are serious potential dangerous behaviors.

[0110] The abnormal behavior detection unit compares the skeletal key point angle features, center of gravity trajectory dynamic time warping (DTW) or human behavior recognition model based on graph neural network (GCN) output by the posture evaluation unit with the system's built-in dangerous behavior database, and jointly analyzes the humidity data and posture recognition results;

[0111] If a person is detected to be slipping or losing balance, and the ambient humidity is higher than 90%, or the dew point is close to the ground temperature, the system increases the confidence level of the behavior.

[0112] A dual judgment mechanism (posture matching + humidity anomaly) is established for suspected abnormal behaviors, and the behavior results are output, including the anomaly type, start timestamp and confidence score, and finally the personnel behavior safety index is output.

[0113] Table 2 shows the comparison table of personnel behavior recognition accuracy.

[0114] Table 2 Comparison of human behavior recognition accuracy

[0115] ;

[0116] The present application proposes a visual processing model that utilizes a deep learning posture recognition network and an abnormal behavior recognition module. This model can track and analyze construction workers' dynamic postures in real time without relying on wearable devices. By identifying dangerous behaviors such as "slipping," "climbing," and "improper bending," this method achieves highly efficient and low-invasive human safety risk monitoring. Compared to traditional manual inspections and wearable devices, this method is more adaptable to complex construction environments and has higher recognition accuracy and coverage.

[0117] Preferably, Figure 5 As shown, the local breeze and vibration detection model is used to detect the local wind speed and turbulence conditions and vibration and resonance phenomena in the area to be detected in real time, and the local environmental stability index is evaluated;

[0118] The local breeze and vibration detection model includes: a local wind speed and turbulence analysis unit, a vibration and resonance phenomenon analysis unit and a local environment stability assessment unit;

[0119] The local wind speed and turbulence condition analysis unit identifies local sudden wind events and small-scale turbulence characteristics based on the wind speed data and the air flow data;

[0120] The vibration and resonance phenomenon analysis unit compares the vibration data with the natural frequency of the construction structure to identify vibration and resonance phenomena;

[0121] The local environment stability assessment unit combines the local sudden wind event, small-scale turbulence characteristics, and the vibration and resonance phenomenon to output the local environment stability index.

[0122] Specifically, the local wind speed and turbulence analysis unit determines whether a local sudden wind event occurs by calculating the short-term (within 3 seconds) wind speed change amplitude and wind direction deviation angle; calculates the spectral density function of the local airflow through Fourier transform, identifies airflow disturbances with high-frequency components (such as >3Hz) in the frequency distribution, and determines them as small-scale turbulence characteristics.

[0123] The vibration and resonance phenomenon analysis unit applies fast Fourier transform (FFT) to the vibration sequence collected by the sensor, extracts the main frequency components, including the vibration frequency peak, amplitude and number of harmonics, and generates a structural vibration spectrum diagram; the natural frequency of the construction component is queried through the structural design parameter library and compared with the current main vibration frequency. If the preset conditions are met, it is determined that the structure has a resonance risk.

[0124] The local environment stability assessment unit inputs the frequency of local sudden wind events, turbulence intensity, vibration and resonance phenomena into the local environment stability index calculation function:

[0125] ;

[0126] in, is the local environmental stability index; is the weight of the frequency item of local sudden wind events; is the frequency of local sudden wind events; is the evaluation time window; is the weight of the turbulence intensity term; is the turbulence intensity; is the weight of vibration and resonance phenomenon; It is a vibration and resonance phenomenon. There is a risk of resonance. There is no resonance risk.

[0127] Table 3 shows the correlation analysis table between platform disturbance and resonance risk.

[0128] Table 3 Correlation analysis of platform disturbance and resonance risk

[0129] ;

[0130] A local breeze and vibration detection model analyzes wind speed, turbulence, and structural resonance risks. This model can capture real-time perturbations in the local wind field at high altitudes or in marginal areas, and predicts structural stability risks through vibration spectrum and structural resonance identification mechanisms. This fine-grained wind-vibration coupled monitoring method significantly outperforms the coarse-grained monitoring of conventional integral anemometers. It can proactively detect risks of falling personnel or abnormal structural shaking caused by local environmental instability, providing technical support for high-risk work scenarios and enhancing dynamic on-site safety prevention and control capabilities.

[0131] Preferably, the light-safety misjudgment index, the personnel behavior safety index and the local environment stability index are used to calculate the safety level of the area to be inspected in real time, generate a safety inspection report and trigger a safety warning mechanism;

[0132] The calculation process of the safety level includes:

[0133] Inputting the light-safety misjudgment index, the human behavior safety index, and the local environment stability index into a risk assessment model to generate a quantitative risk level score, including low risk, medium risk, high risk, and extremely high risk, to obtain the safety level of the area to be inspected;

[0134] The risk assessment model dynamically adjusts the weight of each index based on construction type, operation time, wind force level and historical safety accident data.

[0135] Specifically, the formula for generating a quantitative risk level score by the risk assessment model is:

[0136] ;

[0137] in, Score the quantitative risk level; is the light-safety misjudgment index weight; is the light-safety misjudgment index; is the personnel behavior safety index weight; It is the personnel behavior safety index; is the weight of the local environmental stability index; is the local environmental stability index;

[0138] Risk levels are classified according to quantitative risk level scores:

[0139] : Low risk;

[0140] : Medium risk;

[0141] : High risk;

[0142] : Extremely high risk.

[0143] The risk assessment model dynamically adjusts the weights of each index based on construction type, operation time, wind force level, and historical safety accident data. The dynamic adjustment process is as follows:

[0144] Lighting-Safety Misjudgment Index Weight: Dynamically adjusted based on lighting conditions and operating time (e.g., daytime or nighttime). The weight is increased appropriately at night or in poor lighting conditions, as the risk of misjudgment increases.

[0145] Personnel Behavior Safety Index Weighting: The weighting of the Personnel Behavior Safety Index is adjusted based on the construction type and historical accident data. For example, if a certain type of operation has a high incidence of accidents, the risk assessment of personnel behavior will be given a higher weighting.

[0146] Local environment stability index weight: The stability of the local environment is affected by factors such as wind speed and turbulence. When the wind level is high or the environment is unstable, the environmental stability weight will increase.

[0147] This embodiment comprehensively considers the light-safety misjudgment index, the human behavior safety index, and the local environmental stability index, utilizing a dynamically adjusted risk assessment model to achieve real-time safety assessments for construction areas. This model outputs a comprehensive risk level based on multi-dimensional collaboration and triggers multi-channel early warning measures based on the risk level. Compared to traditional qualitative analysis or single-factor threshold judgment, this method comprehensively considers the impact of light misjudgment, human behavior, and localized sudden wind disturbances on the safety of high-altitude construction work. It offers significant advantages in response accuracy, comprehensive coverage, and intelligence, thereby improving the accuracy and real-time performance of safety detection for high-altitude construction work.

[0148] This invention integrates multiple environmental sensors and intelligent analysis models to achieve comprehensive, real-time monitoring and high-precision safety risk assessment of construction sites operating at height. First, by deploying light sensors, video cameras, wind speed sensors, humidity sensors, air flow sensors, and vibration sensors, multi-dimensional environmental data from the construction area is collected in real time. This data is processed and analyzed by a light and shadow analysis model, a visual processing model, and a local breeze and vibration detection model to accurately identify potential risk factors such as lighting changes, worker behavior, wind turbulence, and structural vibration. Specifically, the light and shadow analysis model analyzes lighting changes in the construction area using light data and video images, calculates a light-safety misjudgment index, and identifies areas of insufficient lighting or misjudgment, thereby reducing the risk of misjudgment in environments with complex lighting variations. The visual processing model identifies key points of the human skeleton in real time, calculates a human behavior safety index, and identifies potentially dangerous behaviors such as slipping, losing balance, or not wearing protective equipment, ensuring the safety of workers. The local breeze and vibration detection model uses wind speed, turbulence, and vibration data to monitor the local environmental stability of the construction area in real time, identifying potential safety hazards such as wind impact and structural vibration. After comprehensive evaluation of this data, an intelligent risk assessment model is used to calculate the safety level of the construction area. Based on the assessment results, the system generates real-time safety inspection reports and automatically triggers safety warning mechanisms when potential high-risk situations are identified, prompting construction personnel and management to take necessary safety measures, reducing the likelihood of accidents and significantly improving safety inspections at construction sites.

[0149] Example 2

[0150] In Example 1, the method proposed by the present invention successfully achieved real-time, high-precision detection of construction safety at height by combining the effects of light and shadow, localized sudden wind, and other factors. To further verify the effectiveness of the present invention, the present example also conducted real-time, high-precision detection of construction safety at height at a construction site B.

[0151] A real-time high-precision construction safety detection system based on environmental characteristics, such as Figure 2 Shown, including:

[0152] The data acquisition module is used to deploy multiple sensors in the area to be inspected to collect on-site environmental data, including lighting, worker videos, wind speed, humidity, air flow, and vibration;

[0153] The multiple sensors include: light sensor, high-definition camera, temperature and humidity sensor, wind speed sensor and acceleration sensor;

[0154] The lighting data includes natural light intensity, artificial lighting brightness, lighting direction, shadow distribution and lighting uniformity parameters;

[0155] The worker video includes a high-definition video sequence of the worker working in the area to be inspected;

[0156] The wind speed data includes the local wind speed vector, the maximum instantaneous wind speed and the wind direction change trend of the area to be detected;

[0157] The humidity data includes relative humidity of air and ground humidity of the area to be detected;

[0158] The air flow data includes local airflow direction, small-scale turbulence intensity and disturbance frequency;

[0159] The vibration data includes the vibration amplitude, vibration frequency, vibration acceleration and structural resonance frequency of the area to be detected.

[0160] Preferably, a light and shadow analysis module is used to analyze the illumination changes of the area to be detected through a light and shadow analysis model according to the illumination, calculate the light-safety misjudgment index, and identify potential visual illusion areas;

[0161] The light and shadow analysis model includes: a light extraction unit, a shadow recognition unit, a dynamic light change tracking unit and a light misjudgment assessment unit;

[0162] The illumination extraction unit extracts a pixel brightness histogram and an illumination gradient distribution in the preprocessed image frame of the worker video;

[0163] The shadow recognition unit separates the dynamic lighting occlusion shadows and the static structure occlusion shadows in the area to be detected, and generates a shadow mask map;

[0164] The dynamic illumination change tracking unit analyzes illumination mutation trends of consecutive image frames according to the pixel brightness histogram, illumination gradient distribution and the shadow mask map;

[0165] The light misjudgment assessment unit integrates the shadow mask map with the construction path and the work edge area, assesses the risk probability of visual illusion caused by illumination, outputs the light-safety misjudgment index of each area in the area to be detected, and marks the area where the light-safety misjudgment index is greater than a preset misjudgment threshold as the potential light misjudgment area.

[0166] Preferably, a visual recognition module is used to analyze the worker video using a visual processing model, identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors;

[0167] The visual processing model includes: a human skeleton key point recognition unit, a posture evaluation unit and an abnormal behavior detection unit;

[0168] The human skeleton key point recognition unit processes the image frames of the worker video through a posture estimation model based on deep learning, extracts the coordinates of the human body key points and generates a human skeleton structure diagram;

[0169] The posture evaluation unit calculates the angle change, center of gravity offset and motion trajectory curve between the skeleton points according to the human skeleton structure diagram to identify the posture information of the person;

[0170] The abnormal behavior detection unit compares the personnel posture information with a dangerous behavior database, detects potential dangerous behaviors in combination with the humidity data, and outputs a personnel behavior safety index.

[0171] Preferably, the local breeze and vibration detection module is used to detect the local wind speed and turbulence conditions of the to-be-detected area and the vibration and resonance phenomena of the to-be-detected platform in real time through a local breeze and vibration detection model, and evaluate the local environmental stability index;

[0172] The local breeze and vibration detection model includes: a local wind speed and turbulence analysis unit, a vibration and resonance phenomenon analysis unit and a local environment stability assessment unit;

[0173] The local wind speed and turbulence condition analysis unit identifies local sudden wind events and small-scale turbulence characteristics based on the wind speed data and the air flow data;

[0174] The vibration and resonance phenomenon analysis unit compares the vibration data with the natural frequency of the construction structure to identify vibration and resonance phenomena;

[0175] The local environment stability assessment unit combines the local sudden wind event, small-scale turbulence characteristics, and the vibration and resonance phenomenon to output the local environment stability index.

[0176] The safety monitoring and early warning module is used to use the light-safety misjudgment index, the personnel behavior safety index and the local environment stability index to calculate the safety level of the area to be detected in real time, generate a safety detection report and trigger a safety early warning mechanism.

[0177] The calculation process of the safety level includes:

[0178] Inputting the light-safety misjudgment index, the human behavior safety index, and the local environment stability index into a risk assessment model to generate a quantitative risk level score, including low risk, medium risk, high risk, and extremely high risk, to obtain the safety level of the area to be inspected;

[0179] The risk assessment model dynamically adjusts the weight of each index based on construction type, operation time, wind force level and historical safety accident data.

[0180] Table 4 shows a comparison table of the overall performance of the system of the present invention.

[0181] Table 4 Overall performance comparison of the system of the present invention

[0182] ;

[0183] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time high-precision construction safety detection method based on environmental characteristics, characterized in that: include: Multiple sensors are deployed in the area to be inspected to collect real-time on-site environmental data, including lighting data, worker videos, wind speed data, humidity data, air flow data, and vibration data. Air flow data includes local airflow direction, small-scale turbulence intensity, and disturbance frequency. Based on lighting data and worker videos, a light and shadow analysis model analyzes lighting changes in the inspection area, calculates the light-safety misjudgment index, and identifies areas of potential lighting misjudgment. The light and shadow analysis model includes a shadow recognition unit that separates dynamic lighting occlusion shadows from static structure occlusion shadows in the inspection area and generates a shadow mask. The dynamic lighting change tracking unit analyzes lighting mutation trends in consecutive image frames based on pixel brightness histograms, lighting gradient distributions, and shadow masks. Analyze worker videos using visual processing models to identify key points of the human skeleton in real time, obtain a behavioral safety index, and detect potentially dangerous behaviors. The visual processing model includes a posture evaluation unit, which calculates the angle change between bone points, center of gravity offset and motion trajectory curve based on the human skeleton structure diagram to identify the person's posture information; The local breeze and vibration detection model is used to detect local wind speed and turbulence conditions, as well as vibration and resonance phenomena in the area to be detected in real time, and to evaluate the local environmental stability index. The local breeze and vibration detection model includes a local wind speed and turbulence analysis unit, which identifies local sudden wind events and small-scale turbulence characteristics based on wind speed data and air flow data; and a vibration and resonance phenomenon analysis unit, which compares vibration data with the natural frequency of the construction structure to identify vibration and resonance phenomena. Using the light-safety misjudgment index, personnel behavior safety index and local environment stability index, the safety level of the area to be inspected is calculated in real time, a safety inspection report is generated and a safety warning mechanism is triggered.

2. A method for real-time high-precision construction safety detection based on environmental characteristics according to claim 1, characterized in that: The multiple sensors include: light sensor, high-definition camera, temperature and humidity sensor, wind speed sensor and acceleration sensor; The lighting data includes natural light intensity, artificial lighting brightness, lighting direction, shadow distribution and lighting uniformity parameters; the worker video includes a high-definition video sequence of workers working in the area to be detected; the wind speed data includes the local wind speed vector, maximum instantaneous wind speed and wind direction change trend of the area to be detected; the humidity data includes the relative humidity of the air and the ground humidity of the area to be detected; the vibration data includes the vibration amplitude, vibration frequency, vibration acceleration and structural resonance frequency of the area to be detected.

3. A method for real-time high-precision detection of construction safety based on environmental characteristics according to claim 1, characterized in that: The light and shadow analysis model also includes: a light extraction unit and a light misjudgment assessment unit; The lighting extraction unit extracts the pixel brightness histogram and lighting gradient distribution in the image frame of the preprocessed worker video; the light misjudgment evaluation unit integrates the shadow mask map with the construction path and the work edge area, evaluates the risk probability of visual illusion caused by lighting, outputs the light-safety misjudgment index of each area in the area to be detected, and marks the area where the light-safety misjudgment index is greater than the preset misjudgment threshold as the potential light misjudgment area.

4. A method for real-time high-precision detection of construction safety based on environmental characteristics according to claim 1, characterized in that: The visual processing model also includes: a human skeleton key point recognition unit and an abnormal behavior detection unit; The human skeleton key point recognition unit processes the image frames of the worker video through a posture estimation model based on deep learning, extracts the coordinates of the human body key points and generates a human skeleton structure diagram; the abnormal behavior detection unit compares the personnel posture information with the dangerous behavior database, detects potential dangerous behaviors in combination with the humidity data, and outputs the personnel behavior safety index.

5. The method for real-time high-precision detection of construction safety based on environmental characteristics according to claim 1 is characterized in that: The local breeze and vibration detection model further includes: a local environment stability assessment unit; The local environment stability assessment unit combines the local sudden wind event, small-scale turbulence characteristics, and the vibration and resonance phenomenon to output the local environment stability index.

6. A method for real-time high-precision construction safety detection based on environmental characteristics according to claim 1, characterized in that: The calculation process of the safety level includes: The light-safety misjudgment index, the personnel behavior safety index and the local environment stability index are input into a risk assessment model to generate a quantitative risk level score, including low risk, medium risk, high risk and extremely high risk, to obtain the safety level of the area to be inspected; the risk assessment model dynamically adjusts the weight of each index according to the construction type, operation time, wind level and historical safety accident data.

7. A real-time high-precision construction safety detection system based on environmental characteristics, characterized in that: Executing the method for real-time, high-precision detection of construction safety based on environmental characteristics as claimed in claim 1, comprising: The data acquisition module is used to deploy multiple sensors in the area to be inspected to collect on-site environmental data, including lighting, worker videos, wind speed, humidity, air flow, and vibration; a light and shadow analysis module, configured to analyze the lighting changes in the area to be detected using a light and shadow analysis model according to the lighting, calculate a light-safety misjudgment index, and identify potential visual illusion areas; A visual recognition module is used to analyze the worker video using a visual processing model, identify key points of the human skeleton in real time, obtain a personnel behavior safety index, and detect potentially dangerous behaviors; A local breeze and vibration detection module is used to detect the local wind speed and turbulence conditions of the to-be-detected area and the vibration and resonance phenomena of the to-be-detected platform in real time through a local breeze and vibration detection model, and evaluate the local environmental stability index; The safety monitoring and early warning module is used to use the light-safety misjudgment index, the personnel behavior safety index and the local environment stability index to calculate the safety level of the area to be detected in real time, generate a safety detection report and trigger a safety early warning mechanism.

Images