Search

CN-121999429-A - Intelligent safety helmet system based on image analysis

CN121999429ACN 121999429 ACN121999429 ACN 121999429ACN-121999429-A

Abstract

The invention relates to the technical field of image analysis, in particular to an intelligent safety helmet system based on image analysis; the intelligent safety helmet comprises a scene analysis module, an image analysis module and a safety decision module, wherein the scene analysis module carries out overall static risk assessment on a construction site to generate an electronic map marked with high, medium and low risk level blocks, the image analysis module dynamically matches different image acquisition frequencies according to the risk levels of the blocks where the safety helmet is located, carries out deep learning model identification and risk quantification on acquired operation images to calculate a dynamic operation risk value, and the safety decision module receives the static scene risk value and the dynamic operation risk value, carries out fusion decision through a preset two-dimensional risk matrix, outputs comprehensive risk levels and triggers corresponding grading early warning to realize the unification of safety and efficiency.

Inventors

  • WANG YONGGANG

Assignees

  • 北京新锐翔通科技有限公司

Dates

Publication Date
20260508
Application Date
20251211

Claims (8)

  1. 1. Intelligent safety helmet system based on image analysis, characterized by comprising: The scene analysis module is used for calculating scene risk values of all grid points based on an electronic scene graph and danger source information of the construction site, dividing the construction site into light, medium and high risk blocks through cluster analysis, and generating a risk block electronic map; The image analysis module is used for activating the camera to acquire an operation image at a matched frequency according to a risk block where the intelligent safety helmet is positioned in real time, carrying out dangerous element identification and quantitative analysis on the image, and calculating a dynamic operation risk value; and the safety decision module is used for receiving the scene risk value and the operation risk value, determining the comprehensive risk level through a risk matrix fusion algorithm, and executing a corresponding hierarchical safety early warning decision.
  2. 2. The intelligent helmet system based on image analysis of claim 1, wherein the scene parsing module performs the steps of: s1, acquiring an electronic scene graph marked with a dangerous source and attributes, wherein the attributes at least comprise a dangerous type, dangerous intensity and diffusion attenuation coefficient; S2, distributing uniform grid points in the electronic scene graph, and calculating a scene risk value of each grid point after being affected by all dangerous sources and overlapped; And S3, dividing the scene risk values of all grid points into three categories by adopting a clustering algorithm, and correspondingly generating mild, moderate and high risk blocks.
  3. 3. The intelligent helmet system based on image analysis according to claim 2, wherein a scene risk value after each grid point is superimposed by all dangerous sources is calculated: S1-2, uniformly distributing grid points at positions except for dangerous sources in an electronic scene graph, identifying and extracting a sitting mark of each grid point as Pi (xi, yi, zi), wherein Pi represents any grid point, i is a grid point index, and the same coordinates of the identified and extracted dangerous sources are marked as Qj (xj, yj, zj), wherein Qj represents any dangerous source, j is a dangerous source index, and a dangerous intensity value and a diffusion attenuation coefficient corresponding to the dangerous source are marked as Wj and rj respectively; s1-3, calculating a scene risk value of each grid point, wherein the specific calculation process is as follows: for each grid point, the Euclidean distance dij from each hazard source to the grid point is calculated by the following formula: For risk propagation of each risk source, a risk attenuation model is adopted, namely the risk is reduced along with the distance, and the risk contribution of the risk source Qj to the grid point Pi is calculated by using the risk attenuation model as follows: The risk contribution value Fij of each hazard source to the grid point Pi can be obtained, and then the scene risk value of the grid point Pi is obtained by summing and calculating the risk contribution value Fij, so that the scene risk value of any grid point in the electronic scene graph can be obtained.
  4. 4. The intelligent helmet system based on image analysis according to claim 3, wherein the scene risk values of all grid points are classified into three categories by using a clustering algorithm: S2-1, arranging all grid points Pi and corresponding scene risk values Hi in an electronic scene graph into a data set, and constructing a risk data matrix, wherein each data sample comprises space coordinates (xi, yi, zi) and continuous scene risk values Hi calculated by the space coordinates; S2-2, determining the cluster number K, K=3 by adopting a K-means clustering algorithm, wherein the cluster number K, K=3 respectively corresponds to three blocks of light risk, medium risk and high risk; S2-3, performing a cluster analysis process: Randomly selecting risk values of three grid points as three initial cluster centers, wherein the risk values represent preliminary mild, moderate and high risk centers respectively; Traversing each grid point Pi, calculating the absolute distance between the scene risk value Hi and the current three cluster center risk values, and distributing the grid point to the cluster represented by the cluster center closest to the scene risk value Hi; thus, all grid points are primarily divided into three clusters, and for three initial clusters formed after distribution, a new center of each initial cluster is recalculated, wherein the new center is an arithmetic average value of scene risk values of all grid points in the cluster, and the new average value is an updated cluster center; Repeating the assigning step to form a new cluster center; when any one of the following conditions is satisfied, namely the judgment algorithm converges, the iteration stops: a) The position of the cluster center does not change significantly any more, i.e. the moving distance is smaller than a preset minimum distance threshold; b) The cluster to which the grid point belongs no longer changes; c) Reaching a preset maximum iteration number; S2-4, outputting final results after algorithm convergence, namely three final cluster centers Clow, cmed and Chigh and grid points included in each cluster, comparing the scene risk values of the three final cluster centers, namely Clow < Cmed < Chigh, defining a space continuous area covered by all grid points included in the cluster with the minimum cluster center Clow as a light risk block, defining a space area corresponding to the cluster with the cluster center Cmed as a medium risk block, and defining a space area corresponding to the cluster with the maximum cluster center Chigh as a high risk block.
  5. 5. The intelligent helmet system based on image analysis of claim 4, wherein the image analysis module performs the steps of: acquiring the position of the intelligent safety helmet in real time, and matching different image acquisition frequencies according to the type of the block in the risk block electronic map, wherein the high risk block matching frequency f1, the medium risk block matching frequency f2, the light risk block matching frequency f3, and f1> f2> f3; Collecting operation images at matched frequencies, identifying dangerous elements in the images by adopting a deep learning model, and outputting confidence coefficient and preset dangerous weight for each identified dangerous element; filtering dangerous elements with confidence coefficient lower than a preset threshold value, and multiplying the confidence coefficient of the effective dangerous elements by the dangerous weight of the dangerous elements to obtain a single risk contribution value; and carrying out nonlinear superposition and normalization processing on all the single risk contribution values to obtain the operation risk values in the range of [0,1 ].
  6. 6. The intelligent helmet system based on image analysis according to claim 5, wherein the nonlinear superposition and normalization process is specifically: All the single risk contribution values Rk are aggregated in a nonlinear superposition mode to obtain a superposition value V, k represents the index of an effective dangerous element, and the nonlinear superposition formula is as follows: ; and then introducing an exponential saturation function to map the superposition value V to obtain an operation risk value, wherein the exponential saturation function is as follows: where λ is the risk sensitivity coefficient, λ >0, which controls the rate at which the job risk value increases with V.
  7. 7. The intelligent helmet system based on image analysis of claim 6, wherein the security decision module performs the steps of: Presetting a scene risk interval and a job risk interval, and respectively quantifying continuous scene risk values and job risk values into three levels of low, medium and high; The method comprises the steps of constructing a two-dimensional risk matrix taking a scene risk level as a first dimension and an operation risk level as a second dimension, defining comprehensive risk levels corresponding to different level combinations by the matrix, inquiring the risk matrix according to the level combination determined by the scene risk value and the operation risk value fed back by the intelligent safety helmet in real time to obtain the comprehensive risk level, and triggering early warning actions corresponding to the comprehensive risk level.
  8. 8. The intelligent helmet system based on image analysis of claim 7, wherein the pre-warning actions defined by the two-dimensional risk matrix comprise at least: The method comprises the steps of presetting an operation risk interval, defining a scene where an intelligent safety helmet is located as a high risk scene when a scene risk value is larger than the upper limit of the operation risk interval, defining the scene where the intelligent safety helmet is located as a medium risk scene when the scene risk value is within the operation risk interval, and defining the scene where the intelligent safety helmet is located as a low risk scene when the scene risk value is smaller than the lower limit of the operation risk interval; The method comprises the steps of presetting an operation risk interval, defining operation behaviors monitored by an intelligent safety helmet as high-risk operation when an operation risk value is larger than the upper limit of the operation risk interval, defining the operation behaviors monitored by the intelligent safety helmet as medium-risk operation when the operation risk value is within the operation risk interval, and defining the operation behaviors monitored by the intelligent safety helmet as low-risk operation when the operation risk value is smaller than the lower limit of the operation risk interval; the two-dimensional risk matrix mapping process comprises the following steps: the combination of the low-risk scene and the low-risk operation has the comprehensive risk level of I, does not need to actively alarm, and only records the state in a background log; A combination of a medium risk scenario and a low risk scenario or a low risk scenario and a medium risk scenario, the overall risk level being level II; changing the color of the safety helmet icon into yellow flashing on an electronic map of a background management platform; A high risk scenario and a low risk job or a combination of a medium risk scenario and a medium risk job or a low risk scenario and a high risk job, the overall risk being defined as class III, and sending a voice prompt to the corresponding intelligent safety helmet; the high risk scene and the medium risk operation or the combination of the medium risk scene and the high risk operation, wherein the comprehensive risk is defined as IV level, and alarm information is pushed to a background command center and a site safety terminal; and (3) combining a high-risk scene and high-risk operation, wherein the comprehensive risk is defined as V level, and triggering on-site audible and visual alarm and starting equipment linkage.

Description

Intelligent safety helmet system based on image analysis Technical Field The invention relates to the technical field of image analysis, in particular to an intelligent safety helmet system based on image analysis. Background The intelligent safety helmet is modern personal protective equipment integrating various intelligent hardware, sensors and software systems on the basis of physical protection of the traditional safety helmet, and is not only a protective shell, but also a movable intelligent terminal and a data acquisition node; Although the intelligent safety helmet has made technical progress in the aspect of construction technical safety, in practical application, the intelligent safety helmet needs to integrate a plurality of sensors, and for long-time outdoor operation, the integrated numerous sensors can bring small power consumption, and the battery endurance of the intelligent safety helmet can influence the continuity of work, reduces work efficiency and increases work risk. Disclosure of Invention The present invention is directed to an intelligent helmet system based on image analysis, so as to solve the above-mentioned problems of the background art. The invention aims at realizing the following technical scheme that the intelligent safety helmet system based on image analysis comprises: The scene analysis module is used for calculating scene risk values of all grid points based on an electronic scene graph and danger source information of the construction site, dividing the construction site into light, medium and high risk blocks through cluster analysis, and generating a risk block electronic map; The image analysis module is used for activating the camera to acquire an operation image at a matched frequency according to a risk block where the intelligent safety helmet is positioned in real time, carrying out dangerous element identification and quantitative analysis on the image, and calculating a dynamic operation risk value; and the safety decision module is used for receiving the scene risk value and the operation risk value, determining the comprehensive risk level through a risk matrix fusion algorithm, and executing a corresponding hierarchical safety early warning decision. Preferably, the scene parsing module performs the steps of: s1, acquiring an electronic scene graph marked with a dangerous source and attributes, wherein the attributes at least comprise a dangerous type, dangerous intensity and diffusion attenuation coefficient; S2, distributing uniform grid points in the electronic scene graph, and calculating a scene risk value of each grid point after being affected by all dangerous sources and overlapped; And S3, dividing the scene risk values of all grid points into three categories by adopting a clustering algorithm, and correspondingly generating mild, moderate and high risk blocks. Preferably, a scene risk value is calculated after each grid point is overlapped by all the influence of the dangerous sources: S1-2, uniformly distributing grid points at positions except for dangerous sources in an electronic scene graph, identifying and extracting a sitting mark of each grid point as Pi (xi, yi, zi), wherein Pi represents any grid point, i is a grid point index, and the same coordinates of the identified and extracted dangerous sources are marked as Qj (xj, yj, zj), wherein Qj represents any dangerous source, j is a dangerous source index, and a dangerous intensity value and a diffusion attenuation coefficient corresponding to the dangerous source are marked as Wj and rj respectively; s1-3, calculating a scene risk value of each grid point, wherein the specific calculation process is as follows: for each grid point, the Euclidean distance dij from each hazard source to the grid point is calculated by the following formula: For risk propagation of each risk source, a risk attenuation model is adopted, namely the risk is reduced along with the distance, and the risk contribution of the risk source Qj to the grid point Pi is calculated by using the risk attenuation model as follows: The risk contribution value Fij of each hazard source to the grid point Pi can be obtained, and then the scene risk value of the grid point Pi is obtained by summing and calculating the risk contribution value Fij, so that the scene risk value of any grid point in the electronic scene graph can be obtained. Preferably, based on the scene risk values of all grid points, the scene risk values are classified into three categories by using a clustering algorithm: S2-1, arranging all grid points Pi and corresponding scene risk values Hi in an electronic scene graph into a data set, and constructing a risk data matrix, wherein each data sample comprises space coordinates (xi, yi, zi) and continuous scene risk values Hi calculated by the space coordinates; S2-2, determining the cluster number K, K=3 by adopting a K-means clustering algorithm, wherein the cluster number K, K=3 respectively correspo