Image recognition-based real-time monitoring method for safety state of workers

Abstract

The present application belongs to the technical field of video recognition, and particularly relates to a real-time monitoring method for safety state of workers based on image recognition, comprising: obtaining images of a work site and preprocessing, and extracting a sequence of personnel center coordinate points; constructing a motion fluctuation value model based on direction continuity, and quantifying behavior stability by using displacement vector and direction angle deviation; introducing a hyperbolic tangent function with saturation characteristics, and nonlinearly coupling the motion fluctuation value and the distance of personnel from a dangerous source, and calculating a safety risk index; and dynamically adjusting the monitoring frequency through index mapping according to the risk index, and realizing graded early warning. The present application solves the technical problems of poor robustness and uneven resource allocation of traditional monitoring methods in complex environments, improves the accuracy of safety supervision through deep coupling of motion stability and spatial constraints, and realizes fine configuration of computing resources.

Description

Technical Field

[0001] This invention relates to the field of video recognition technology. More specifically, this invention relates to a method for real-time monitoring of worker safety status based on image recognition. Background Technology

[0002] With the acceleration of industrial modernization and the widespread application of intelligent manufacturing concepts, the safety of industrial production environments has received high attention from all sectors of society. In order to ensure the personal safety of workers on the production site, video surveillance technology based on computer vision (CV) has become an important means of safety supervision.

[0003] Safety monitoring at industrial work sites mainly relies on manual inspections or simple moving target detection. However, in practical applications, due to environmental noise interference such as complex lighting and dust, the accuracy of digital image processing is often difficult to guarantee, which can easily lead to calculation errors when the system acquires the positional characteristics of workers. Existing behavior analysis technologies mostly focus on the classification of specific static actions and lack effective physical quantitative evaluation indicators for the dynamic posture instability of workers in complex work paths, making it difficult for the system to accurately distinguish between normal work movements and sudden abnormal posture fluctuations.

[0004] Furthermore, existing risk assessment logic often overlooks the spatial geometric constraints between workers and hazards, failing to fully consider the nonlinear amplification effect of environmental factors on individual movement risks. This results in early warning systems lacking sufficient risk prediction windows when facing abnormal behavior in nearby hazardous areas, easily leading to false alarms or missed alarms. If monitoring systems maintain a constant high frequency of image capture and edge computing for extended periods, it generates massive data redundancy and causes severe hardware heat dissipation and network bandwidth pressure, significantly reducing system lifespan and real-time feedback efficiency. Therefore, how to improve the sensitivity of operational status monitoring while optimizing the allocation of computing resources has become an urgent issue to be addressed in the current industrial security field. Summary of the Invention

[0005] To address the technical problems of low accuracy in identifying abnormal worker movements, lack of environmental constraints in risk assessment, and unreasonable allocation of monitoring resources, this invention provides a real-time monitoring method for worker safety status based on image recognition. The method includes: acquiring and preprocessing an original image sequence from the work site to obtain a denoised image sequence; extracting the center coordinate position data of the worker from this sequence and performing median filtering to obtain a sequence of worker center coordinate points; obtaining the displacement and angle of the motion vector based on the worker center coordinate point sequence at adjacent sampling times; obtaining motion fluctuation values ​​using the displacement, angle, worker movement speed, and deflection weighting coefficient; obtaining the shortest distance from the worker center coordinate point sequence to a preset hazard source boundary; using the shortest distance as a penalty term to weight the motion fluctuation values ​​to obtain a safety risk index; obtaining the monitoring frequency based on the safety risk index; and adjusting the on-site warning intensity based on the safety risk index and the monitoring frequency.

[0006] This invention denoises the original image sequence of the work site and obtains the sequence of center coordinates of the workers. It combines the displacement and angle of the motion vector to obtain the motion fluctuation value and uses the shortest distance of the worker relative to the boundary of the preset hazard source as a penalty to obtain the safety risk index. In this way, the monitoring frequency and the intensity of on-site warnings are dynamically adjusted according to the risk level, thereby improving the dynamic response capability of the safety monitoring system in the industrial production environment to abnormal personnel movement and spatial risks.

[0007] Preferably, the preprocessing of the original image sequence of the work site includes: denoising the original image sequence of the work site collected by the monitoring terminal using a standard Gaussian filtering algorithm.

[0008] This invention utilizes a standard Gaussian filtering algorithm to denoise the original image sequences of the work site collected by the monitoring terminal, reducing the interference of environmental noise such as complex lighting or dust on image quality, and providing a high-quality image data foundation for accurately locating the work personnel area.

[0009] Preferably, the motion fluctuation value satisfies the expression: In the formula, express Motion fluctuation value at any given moment; express The displacement vector of motion at any given moment; express The displacement vector of motion at any given moment; Indicates the speed at which the workers move; express The angle of the motion vector at any given moment; express The angle of the motion vector at any given moment; This represents the deflection weighting coefficient; Represents the natural constant; Represents the absolute value symbol.

[0010] This invention utilizes motion fluctuation values ​​containing exponential terms to analyze displacement and angle changes at adjacent sampling times, transforming abrupt changes in motion direction into numerical increases in motion fluctuation values. This reduces identification bias caused by operator posture instability or positioning drift, and improves the ability to capture abnormal posture fluctuations.

[0011] Preferably, the deflection weight coefficient is 1.2.

[0012] Preferably, obtaining the shortest distance of the worker's center coordinate point sequence relative to the preset hazard source boundary includes: using a monocular depth mapping model to map the image pixel coordinates to physical space, and calculating the geometric distance between the worker's center coordinate point sequence and the preset hazard source boundary.

[0013] This invention utilizes a monocular depth mapping model to map image pixel coordinates to physical space and calculates the geometric distance between the sequence of center coordinate points of the operator and the preset hazard source boundary. This introduces physical space constraints for safety assessment and reduces the inaccuracy of risk assessment due to a lack of environmental depth information.

[0014] Preferably, the safety risk index satisfies the expression: In the formula, express The safety risk index at any given moment; express Motion fluctuation value at any given moment; express The shortest distance to the source of danger at any given moment; Indicates distance response sensitivity; Represents the hyperbolic tangent function; This represents the basic risk bias term.

[0015] Preferably, the distance response sensitivity is 0.2.

[0016] Preferably, the monitoring frequency satisfies the expression: In the formula, express Monitoring frequency at any given moment; Indicates the basic sampling frequency; Indicates frequency adjustment gain; express The safety risk index at any given moment; This represents an exponential function with the natural constant as its base.

[0017] This invention dynamically obtains the monitoring frequency based on the safety risk index using an exponential function with a natural constant as the base. This allows the system to automatically increase the image capture density when the risk increases, reducing hardware heat loss and network bandwidth pressure when personnel are in a low-risk state.

[0018] Preferably, the frequency adjustment gain is 3.

[0019] Preferably, adjusting the intensity of the on-site warning includes: when the safety risk index exceeds a preset alarm threshold, sending a command to the on-site alarm terminal to increase the flashing frequency of the on-site warning light and increase the output decibel of the alarm sound, while simultaneously driving the image acquisition device to increase the image capture density.

[0020] The beneficial effects of this invention are as follows: This invention obtains the sequence of center coordinates of workers through background modeling and foreground segmentation techniques, and assesses the safety risk index by combining motion vector characteristics and the shortest distance to the hazard source. It nonlinearly couples individual motion state with environmental spatial constraints, reducing the chance of missed or false alarms in safety warnings in complex industrial environments.

[0021] This invention dynamically obtains the monitoring frequency based on changes in the safety risk index, enabling real-time feedback adjustment of monitoring intensity and operational risk level. While ensuring monitoring strength at risk moments, it reduces data redundancy during the operation of the monitoring system and extends the service life of the monitoring hardware.

[0022] This invention adjusts the flashing frequency of on-site warning lights, the decibel level of alarm sounds, and the density of image capture in conjunction with the safety risk index when it exceeds a preset alarm threshold. Through a multi-dimensional early warning strategy, it enhances the on-site alerting effect on abnormal work behavior and provides industrial workers with a longer risk prediction window when they are near dangerous areas. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the real-time monitoring method for worker safety status based on image recognition in this invention; Figure 2 This is a schematic diagram illustrating the change of motion fluctuation values ​​over sampling time; Figure 3 This is a schematic diagram illustrating how the safety risk index changes with the shortest distance from the hazard source; Figure 4 This is a schematic diagram illustrating how the monitoring frequency changes with the safety risk index. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] This invention discloses a method for real-time monitoring of worker safety status based on image recognition, referring to... Figure 1 This includes steps S1 to S4: S1. Obtain and preprocess the original image sequence of the work site to obtain a denoised image sequence. Extract the center coordinate position data of the workers from the sequence and perform median filtering to obtain the center coordinate point sequence of the workers.

[0027] It should be noted that, due to the random noise and environmental interference in the original image sequence of the work site collected by the monitoring terminal, direct feature extraction will lead to calculation deviations in subsequent algorithms. This invention uses a standard Gaussian filtering algorithm to denoise the original image sequence of the work site, aiming to eliminate environmental noise and locate the area of ​​the workers, thus providing a high-quality data foundation for subsequent dynamic analysis.

[0028] Specifically, this invention acquires continuous time-series frame images, defined as the original image sequence of the work site; processes the original image sequence of the work site using a standard Gaussian filtering algorithm to obtain a denoised image sequence of the work site; and processes the denoised image sequence of the work site using background modeling and foreground segmentation techniques to obtain the sequence of center coordinate points of the workers. Furthermore, this invention uses a median filtering algorithm to perform secondary smoothing on the sequence of center coordinate points of the workers, removing granular noise generated during image processing and ensuring the numerical stability of the sequence of center coordinate points of the workers.

[0029] S2. Obtain the motion vector displacement and motion vector angle based on the sequence of center coordinate points of the workers at adjacent sampling times, and obtain the motion fluctuation value using the motion vector displacement, motion vector angle, worker movement speed and deflection weight coefficient.

[0030] It should be noted that the motion vector of a worker during normal operation usually has a smooth directionality. However, when a worker falls, shakes violently, or the sensor drifts due to obstruction, the magnitude and direction angle of the motion vector will suddenly distort. Traditional linear smoothing models cannot effectively distinguish between normal work movements and abnormal posture instability. Therefore, this invention obtains motion fluctuation values ​​to evaluate the physical stability of the worker's behavior.

[0031] Specifically, this invention calculates the motion fluctuation value of the worker based on the center coordinate displacement vector between the current moment and the previous moment, and the motion fluctuation value satisfies the expression:

[0032] In the formula, express Motion fluctuation value at any given moment; express The displacement vector of motion at any given moment; express The displacement vector of motion at any given moment; Indicates the speed at which the workers move; express The angle of the motion vector at any given moment; express The angle of the motion vector at any given moment; This represents the deflection weighting coefficient; Represents the natural constant; Represents the absolute value symbol.

[0033] In the formula, the exponential term reflects the exponential contribution of sudden changes in motion direction to the instability score; as the deviation of the motion vector angle between adjacent sampling times increases, the value of the exponential mapping term rises rapidly, causing the exponential mapping term to increase and driving the final generated motion fluctuation value. It gets bigger.

[0034] It should be further noted that the deflection weighting coefficient in this invention... The value is set to 1.2; if the deflection weight coefficient... If the setting is too small, the system will lack sufficient sensitivity to detect early signs of a worker suddenly turning or falling; if the deflection weight coefficient is too small... If the setting is too high, normal work movements will trigger a high-frequency fluctuation alarm; this invention uses 1.2 as an empirical value, which can accurately capture the abnormal movement state of the workers while ensuring a low false alarm rate.

[0035] For example, Figure 2 This is a schematic diagram illustrating the change in motion fluctuation values ​​over sampling time. The figure shows the evolution trend of the stability index of the worker under different motion states. In the initial sampling stage, the worker is in a stable walking state with small deviations in the motion vector angle, and the motion fluctuation value remains in a low and smooth range. As time goes on, the worker experiences sudden posture deflection or positioning instability. Due to the exponential amplification effect of the deflection weighting coefficient on the angle change, the motion fluctuation value shows a significant pulse-like increase, demonstrating high sensitivity in capturing abnormal behavior characteristics of personnel.

[0036] S3. Obtain the shortest distance of the worker's center coordinates relative to the preset hazard source boundary. Use the shortest distance of the hazard source as a penalty to weight the motion fluctuation value to obtain the safety risk index.

[0037] It should be noted that the impact of the worker's motion instability on the safety status is constrained by the environment in which they are located. When the worker is far away from the hazard source, slight motion fluctuations are within the normal range. However, when the worker is near dangerous areas such as live equipment or high-altitude edges, fluctuations of the same amplitude will pose an extremely high safety threat. Therefore, this invention introduces a hyperbolic tangent function with saturation characteristics to obtain the nonlinear coupling relationship between motion fluctuations and spatial distance.

[0038] Specifically, this invention acquires the shortest distance between the worker's center coordinates and the preset hazard source boundary in real time, and applies risk weighting to the motion fluctuation values, outputting the shortest distance. The security risk index at any given time, wherein the security risk index satisfies the expression:

[0039] In the formula, express The safety risk index at any given moment; express Motion fluctuation value at any given moment; express The shortest distance to the source of danger at any given moment; Indicates distance response sensitivity; Represents the hyperbolic tangent function; This represents the basic risk bias term.

[0040] In the formula, the terms within parentheses constitute the distance penalty factor; as the shortest distance to the hazard source... The decrease in the value of the hyperbolic tangent causes its output to approach zero, resulting in a non-linear increase in its reciprocal term, which in turn drives the safety risk index. It increases in size, thereby amplifying the risk to people near the danger source.

[0041] It should be further noted that the distance response sensitivity in this invention... The value is 0.2, and the unit is the reciprocal of the distance (meters); if the distance response sensitivity... If the value is too small, the risk warning range will be too large, leading to frequent false alarms; if the distance to the response sensitivity is too small... If the value is too large, a sudden increase in risk value will only occur at extremely close range, thus losing the window of opportunity for risk avoidance; this invention selects 0.2 to ensure a significant increase in risk signal within 1.5 meters of the hazard source for the worker. Basic Risk Bias Item The value typically ranges from 0.5 to 2.0. If this value is set too small, the weighting coefficient within the brackets will be too low when personnel are far from the hazard source, making it difficult for the safety risk index to trigger the basic monitoring logic and causing missed detection of sudden unstable behavior by personnel within the safe zone. If this value is set too large, it will overwhelm the system. The dynamic nature of the system leads to a decrease in its sensitivity to the risks of personnel approaching a hazard source.

[0042] For example, Figure 3 This is a schematic diagram showing how the safety risk index changes with the shortest distance from the hazard source. When workers are far from the boundary of the hazard source, the safety risk index remains at a low level and changes slowly; once workers enter the hazard distance threshold range, the safety risk index spikes dramatically as the distance decreases.

[0043] S4. Obtain the monitoring frequency based on the safety risk index, and adjust the on-site warning intensity according to the safety risk index and the monitoring frequency.

[0044] It should be noted that continuous high-frequency image analysis will bring huge hardware heat loss and bandwidth consumption to edge computing nodes; in order to achieve fine-grained resource allocation, it is necessary to adjust the monitoring intensity of the system in real time based on the current risk level of the operators.

[0045] Specifically, the present invention obtains the monitoring frequency based on the calculated safety risk index, wherein the monitoring frequency satisfies the expression:

[0046] In the formula, express Monitoring frequency at any given moment; Indicates the basic sampling frequency; Indicates frequency adjustment gain; express The safety risk index at any given moment; This represents an exponential function with the natural constant as its base.

[0047] In the formula, with the safety risk index The increase in the exponential term leads to a significant enhancement in the gain effect of the exponential term, causing the exponential term to increase and driving the monitoring frequency. It gets bigger.

[0048] It should be further added that the frequency adjustment gain in this invention The value is 3, and the unit is the reciprocal of the square of time (seconds); excessively high frequency adjustment gain. This can cause the monitoring frequency to fluctuate frequently under normal operating conditions, damaging the lifespan of the camera's actuator; excessively low frequency adjustment gain This would prevent the provision of sufficient video stream density for post-incident tracing when workers encounter risks; the present invention selects 3, which effectively achieves a balance between energy consumption management and monitoring reliability.

[0049] Furthermore, when the safety risk index is detected to exceed the preset alarm threshold, the present invention sends a command to the on-site alarm terminal through the controller; increases the flashing frequency of the on-site warning lights and increases the output decibel of the alarm sound, while driving the image acquisition device to increase the image capture density.

[0050] For example, Figure 4 This diagram illustrates how monitoring frequency changes with the safety risk index. When the safety risk index is low, the monitoring frequency remains at a low base sampling frequency, minimizing the generation of redundant data and hardware wear and tear. When the safety risk index rises due to abnormal personnel behavior or dangerous locations, the monitoring frequency increases rapidly, enhancing image capture density and ensuring that more detailed on-site status data can be obtained for early warning at critical risk moments.

Claims

1. A method for real-time monitoring of worker safety status based on image recognition, characterized in that, include: The original image sequence of the work site is acquired and preprocessed to obtain a denoised image sequence. The center coordinate position data of the workers is extracted from the sequence and then processed by median filtering to obtain the center coordinate point sequence of the workers. The motion vector displacement and motion vector angle are obtained based on the sequence of center coordinate points of the workers at adjacent sampling times. The motion fluctuation value is obtained using the motion vector displacement, motion vector angle, worker movement speed and deflection weight coefficient. Obtain the shortest distance of the worker's center coordinates relative to the preset hazard source boundary, and use the shortest distance of the hazard source as a penalty to weight the motion fluctuation value to obtain the safety risk index; The monitoring frequency is determined based on the safety risk index, and the intensity of on-site warnings is adjusted according to the safety risk index and the monitoring frequency.

2. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The preprocessed work site original image sequence includes: The standard Gaussian filtering algorithm is used to denoise the original image sequence of the work site collected by the monitoring terminal.

3. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The motion fluctuation value satisfies the expression: ； In the formula, express Motion fluctuation value at any given moment; express The displacement vector of motion at any given moment; express The displacement vector of motion at any given moment; Indicates the speed at which the workers move; express The angle of the motion vector at any given moment; express The angle of the motion vector at any given moment; This represents the deflection weighting coefficient; Represents the natural constant; Represents the absolute value symbol.

4. The method for real-time monitoring of worker safety status based on image recognition according to claim 3, characterized in that, The deflection weight coefficient is set to 1.

2.

5. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The process of obtaining the sequence of worker center coordinates relative to the shortest distance to the preset hazard source boundary includes: The image pixel coordinates are mapped to physical space using a monocular depth mapping model, and the geometric distance between the sequence of center coordinate points of the workers and the preset hazard source boundary is calculated.

6. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The security risk index satisfies the expression: ； In the formula, express The safety risk index at any given moment; express Motion fluctuation value at any given moment; express The shortest distance to the source of danger at any given moment; Indicates distance response sensitivity; Represents the hyperbolic tangent function; This represents the basic risk bias term.

7. The method for real-time monitoring of worker safety status based on image recognition according to claim 6, characterized in that, The distance response sensitivity is set to 0.

2.

8. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The monitoring frequency satisfies the expression: ； In the formula, express Monitoring frequency at any given moment; Indicates the basic sampling frequency; Indicates frequency adjustment gain; express The safety risk index at any given moment; This represents an exponential function with the natural constant as its base.

9. The method for real-time monitoring of worker safety status based on image recognition according to claim 8, characterized in that, The frequency adjustment gain is set to 3.

10. The method for real-time monitoring of worker safety status based on image recognition according to claim 1, characterized in that, The adjustment of the on-site warning intensity includes: When the safety risk index exceeds the preset alarm threshold, a command is sent to the on-site alarm terminal to increase the flashing frequency of the on-site warning lights and increase the output decibel of the alarm sound, while simultaneously driving the image acquisition equipment to increase the image capture density.

Images