Campus dual-prevention risk identification system and identification method

Abstract

The invention discloses a campus dual-prevention risk identification system and identification method, relates to the technical field of campus risk identification, and solves the technical problem that the accuracy and comprehensiveness of evaluation are affected by negligence and omission inevitably caused by manual inspection for a large number of risk areas of repeated types in a campus. The method comprises the following steps: acquiring a multi-azimuth full-color image and a scene depth-of-field image at a potential risk position through a campus dual-prevention risk knowledge base, fusing to obtain an RGB-D image of the potential risk, and carrying out image quality inspection; inputting the RGB-D image of the potential risk into a corresponding risk identification model to identify a risk feature, analyzing a risk coefficient of the potential risk, further evaluating the severity of the potential risk, and processing the risk; according to the invention, the risk checking efficiency is improved, and negligence caused by manual checking is also avoided.

Description

Technical Field

[0001] The present invention belongs to the field of campus risk identification, and specifically relates to a campus dual prevention risk identification system and identification method. Background Art

[0002] Campus safety is not only directly related to the life safety of teachers and students, but also has a profound impact on the healthy growth of students.

[0003] The existing technology uses a double prevention mechanism to manage campus safety. However, the core of the existing double prevention mechanism lies in the professional judgment and rich experience of registered safety engineers. They need to accurately identify various potential risks in various risk areas on campus, including classrooms, dormitories, laboratories, sports fields, etc., and scientifically classify them to determine whether there are safety hazards. However, this traditional manual inspection model is highly dependent on human resources, and the professional knowledge and practical experience of registered safety engineers are difficult to replicate and promote on a large scale. In addition, in the face of a large number of repetitive risk areas on campus, manual inspections will inevitably lead to negligence and omissions, which will affect the accuracy and comprehensiveness of the assessment.

[0004] Therefore, the present invention provides a campus dual prevention risk identification system and identification method. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a campus double prevention risk identification system and identification method, which are used to solve the technical problem that in the face of a large number of repeated types of risk areas on campus, negligence and omissions will inevitably occur through manual inspection, thereby affecting the accuracy and comprehensiveness of the assessment.

[0006] To achieve the above-mentioned object, the first aspect of the present invention provides a campus dual prevention risk identification system, comprising an image acquisition and processing module, and a risk knowledge base module and a risk assessment module connected thereto;

[0007] Risk knowledge base module: used to define campus risk information and the relationship between risk information, and to build a campus dual prevention risk knowledge base; the risk information includes the campus risk units, risk points, potential risks corresponding to the risk points, and inspection standards corresponding to the potential risks;

[0008] Image acquisition and processing module: used to collect multi-directional full-color images and scene depth images of potential risk locations, fuse them to obtain RGB-D images of potential risks, and perform image quality inspection; among them, image quality inspection includes quality inspection of image data format, quality inspection of image data quality, and quality inspection of image data integrity;

[0009] Risk assessment module: used to input the RGB-D image of potential risk into the corresponding risk identification model to identify risk characteristics and analyze the risk coefficient of potential risk; among which, the risk identification model is built based on the CNN neural network model.

[0010] Preferably, the definition of campus risk information and the relationship between risk information includes:

[0011] Define risk units: Encode the physical spaces on campus in the knowledge base;

[0012] Define risk points: Risk points are subordinate units of risk units. Defining risk points includes defining and coding in the knowledge base and selecting the risk units to which they belong;

[0013] Define the potential risks of risk points and define the inspection criteria for risk points.

[0014] The present invention realizes the refinement and concretization of safety management by breaking down the physical space of the campus into risk units and risk points. This "point-to-point" management method makes safety management no longer vague and general, but can be accurate to every corner of the campus, effectively improving management efficiency and pertinence; and clarifies the potential risks and inspection standards of risk points. Based on the dual prevention mechanism and the knowledge and experience of the professional team, it can predict and effectively control safety risks in advance.

[0015] Preferably, the construction of a campus dual prevention risk knowledge base includes:

[0016] Abstract the physical space of the school into the database, and mark the risk units, risk points, potential risks corresponding to the risk points, potential risk points and inspection standards of the campus as entities of the knowledge graph;

[0017] Extract the risk units of the campus, the risk points corresponding to the risk units, the potential risks corresponding to the risk points, and the inspection standards corresponding to the potential risk points from the expert knowledge base; use the one-to-one correspondence between the risk units, risk points, potential risks and inspection standards as the connection relationship between the knowledge graph entities;

[0018] By constructing the entity-relationship-entity triple data group of the knowledge graph, the campus dual prevention risk knowledge base is obtained.

[0019] The method of the present invention can comprehensively cover all physical spaces on campus, and systematically present the complex campus safety management system in the database. In the form of a knowledge graph, it not only displays the hierarchical structure between risk units and risk points, but also deeply reveals the intrinsic connection between potential risks and inspection standards, and constructs a complete and systematic campus safety risk management system; based on the extraction and construction of the expert knowledge base, the automatic generation and updating of the knowledge graph is realized, which is helpful to quickly locate problems, formulate targeted measures, and improve the level of safety management.

[0020] Preferably, the collecting of multi-directional full-color images and scene depth images of potential risk locations and fusing them to obtain RGB-D images of potential risks include:

[0021] The full-color image and the scene depth-of-field image are collected by the risk identification inspection device; wherein the risk identification inspection device includes a full-color camera and a TOF camera, which respectively collect the scene color image and the scene depth-of-field image;

[0022] The displacement deviation between the full-color image and the scene depth image is corrected through an image alignment algorithm, so that the two images correspond one to one, the image information is merged, and an RGB-D image of potential risk is obtained; wherein the image alignment algorithm includes a feature-based method or a pixel-based method.

[0023] The TOF camera of the present invention is insensitive to lighting conditions and can work stably in various environments, while the full-color camera can capture rich color details. The combination of the two makes the risk identification system more adaptable to the complex and changeable campus environment, so that no matter whether it is lighting changes, interference from obstructions or other complex scenes, RGB-D images can provide a stable and rich source of information; making the identification of potential risks more comprehensive and accurate, color information helps to identify the appearance characteristics of objects, and depth information can reveal the spatial position and shape of objects. The combination of the two can better adapt to the complex and changeable campus environment, laying the foundation for the subsequent identification of abnormal features from images.

[0024] Preferably, image quality inspection is performed on the RGB-D image with potential risks, including:

[0025] Step S1: Verify whether the image shooting time is consistent with the inspection task time, and verify whether the image GPS coordinates are consistent with the risk unit location; if yes, proceed to step 2; if no, reshoot;

[0026] Step S2: Perform quality inspection on the collected images;

[0027] Quality inspection of image data format: converting images into a unified image format;

[0028] The quality inspection of image data includes image size and resolution, image color mode, image clarity and contrast;

[0029] Image size and resolution: resize the images to the same size and resolution;

[0030] Image color mode: Check whether the color image is in RGB color mode; if yes, do not adjust the image color; if no, adjust the image to RGB color mode;

[0031] Image clarity and contrast: The edge detection algorithm is used to analyze the edge strength of the image and compare it with the clarity threshold to evaluate the image clarity;

[0032] Analyze the contrast of the image through the image histogram;

[0033] Quality inspection of image data integrity: Perform integrity checks on potential risk items at each risk point, and confirm that each object or equipment facility has images from multiple angles.

[0034] The present invention ensures the timeliness and accuracy of the data by verifying the consistency between the image shooting time and the inspection task time, as well as the matching of the image GPS coordinates and the risk unit location, avoiding misjudgment due to time or space dislocation, and providing a reliable data basis for risk analysis; secondly, the comprehensive quality inspection of image data further improves the availability of the data, allowing image data to be seamlessly connected between different systems and platforms, facilitating subsequent analysis and processing, and ensuring visual accuracy in the risk identification process.

[0035] Preferably, the risk identification model is constructed based on a CNN neural network model, including:

[0036] Extracting several potential risk images from historical data, and marking each abnormal feature in the potential risk images, and marking different abnormal features with different data numbers;

[0037] The marked potential risk images are used as standard input data for the CNN neural network model;

[0038] The data number of the abnormal features is used as the standard output data of the CNN neural network model;

[0039] The CNN neural network model is trained using standard input data and standard output data to obtain a risk identification model.

[0040] The present invention extracts images of potential risks from historical data and marks the abnormal features in each image in detail. This method ensures the authenticity and representativeness of the training data. These marks not only point out the specific locations of risks in the images, but also distinguish different types of abnormal features through different data numbers, providing rich information input for the model. Secondly, the risk identification method based on deep learning not only improves the accuracy and efficiency of identification, but also has strong generalization ability, and can cope with complex and changeable risk scenarios.

[0041] Preferably, the risk coefficient of analyzing potential risks includes:

[0042] Count the number of abnormal features detected from the potential image, and calculate the proportion of each abnormal feature, that is, the ratio of the number of each abnormal feature to the total number of abnormal features;

[0043] The risk coefficient FXS of potential risk is calculated by the formula FXS = ∑ (αi × YBi); where αi is the weight coefficient, αi>0, ∑αi=1; YBi is the proportion of abnormal features; i is the type of abnormal feature; ∑ is the sum of i;

[0044] Compare the risk factor with the preset threshold to assess the degree of danger of potential risks.

[0045] Preferably, the risk reporting module is used to output an inspection report after completing the inspection of all risk points.

[0046] Preferably, a method for identifying dual prevention risks on campus comprises the following steps:

[0047] Step Q1: First select the risk unit to be checked, and then select the subordinate risk points;

[0048] Step Q2: After selecting the risk point, the risk identification and inspection device reads the potential risk list corresponding to the corresponding risk point from the campus dual prevention risk knowledge base, as well as the inspection standard corresponding to each potential risk, and performs a safety inspection on the risk point according to the inspection standard; wherein the risk identification and inspection device includes a full-color camera and a TOF camera, which respectively collect scene color images and scene depth images;

[0049] Step Q3: Use risk identification inspection equipment to collect data according to the inspection standards, usually by collecting full-color images and scene depth images of items or equipment facilities with potential risks from all directions; when collecting data, full-color images and depth images are collected simultaneously to obtain RGB-D images of potential risks;

[0050] Step Q4: Perform image quality inspection on the RGB-D image with potential risks; wherein the image quality inspection includes quality inspection of image data format, quality inspection of image data quality and quality inspection of image data integrity;

[0051] Step Q5: Identify the risk features of the RGB-D image of potential risk through a risk identification model; wherein the risk identification model is constructed based on a CNN neural network model;

[0052] Step Q6: Analyze the risk factor of potential risks and assess the risk level of potential risks.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention proposes a campus dual prevention risk identification device and system based on knowledge base and image recognition. First, the campus entities and risk inspection related facilities are abstracted, and various risk inspections are standardized within the limited scenarios of the campus. The inspection workflow is streamlined in combination with the risk identification inspection equipment, which solves the dependence of the registered safety engineer on the campus for dual prevention inspections, and realizes the replication of the knowledge and experience of registered safety engineers in campus safety inspection scenarios. Secondly, artificial intelligence is used to analyze the data collected during the safety inspection, and it is found in practice that safety inspections involve small equipment and facilities, and the accuracy of risk analysis achieved by color image recognition alone is low. Therefore, a TOF camera is added to the acquisition device to synthesize RGB-D images to improve the accuracy of risk analysis. Image recognition replaces traditional manual risk analysis, which improves the efficiency of risk inspections and avoids negligence in manual inspections.

[0055] 2. The present invention collects multi-directional full-color images and scene depth images of potential risk locations through the campus dual prevention risk knowledge base, fuses them to obtain RGB-D images of potential risks, and performs image quality inspection; the RGB-D images of potential risks are input into the corresponding risk identification model to identify risk characteristics, and the risk coefficients of potential risks are analyzed, thereby evaluating the severity of potential risks and processing the risk locations; the present invention uses expert knowledge bases and artificial intelligence to achieve replication of the knowledge and experience of registered safety engineers in campus safety inspection scenarios, automation of risk identification, and automation of risk report generation, thereby greatly reducing the cost of campus dual prevention safety inspections and improving the quality of safety inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 It is a schematic diagram of the system structure of the present invention;

[0058] Figure 2 A schematic diagram of the campus risk report output of the present invention;

[0059] Figure 3 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0060] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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 creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 , the first aspect of the present invention provides a campus dual prevention risk identification system, including an image acquisition and processing module, and a risk knowledge base module and a risk assessment module connected thereto;

[0062] The risk knowledge base module defines campus risk information and the relationship between risk information, and constructs a campus dual prevention risk knowledge base; wherein, risk information includes campus risk units, risk points, potential risks corresponding to risk points, and inspection standards corresponding to potential risks;

[0063] Specifically, define risk units: define and encode the physical space on campus in the knowledge base; among them, risk units include teaching buildings, sports and arts halls, dormitory buildings, outdoor stadiums, etc.

[0064] Define risk points: Risk points are subordinate units of risk units. Defining risk points includes defining and coding in the knowledge base and selecting the risk units to which they belong;

[0065] For example, taking the teaching building of the risk unit as an example, the subordinate risk points include classrooms, offices, toilets, etc.; taking the dormitory building of the risk unit as an example, the subordinate risk points include dormitory buildings, laundry rooms, corridors, etc.

[0066] Define the potential risks of risk points: For each risk point, there are common or specific potential risks. The basis for the definition comes from the relevant laws, regulations and standards for dual prevention, as well as the knowledge and experience accumulated by the registered safety engineer team in the process of safety inspection practice;

[0067] Taking the risk-point classroom as an example, there may be problems such as ceiling fans, lamps, blackboards and other facilities not being installed securely; window limiters failing, students climbing windows and corridor railings, etc.; the blackboard being incomplete, damaged, and glaring; switches, sockets and lighting fixtures being damaged and other potential risks.

[0068] Define inspection criteria for potential risks: After defining potential risks, each potential risk has its corresponding inspection criteria; the basis for the definition is related to the basis for the definition of potential risks; the inspection criteria also define the data collection process. The risk identification inspection equipment includes a full-color camera and a TOF camera, which respectively collect scene color images and scene depth images.

[0069] Abstract the physical space of the school into the database, and mark the risk units, risk points, potential risks corresponding to the risk points, potential risk points and inspection standards of the campus as entities of the knowledge graph;

[0070] Extract the risk units of the campus, the risk points corresponding to the risk units, the potential risks corresponding to the risk points, and the inspection standards corresponding to the potential risk points from the expert knowledge base; use the one-to-one correspondence between the risk units, risk points, potential risks and inspection standards as the connection relationship between the knowledge graph entities;

[0071] By constructing the entity-relationship-entity triple data group of the knowledge graph, the campus dual prevention risk knowledge base is obtained.

[0072] The image acquisition and processing module collects multi-directional full-color images and scene depth images of potential risk locations, fuses them to obtain RGB-D images of potential risks, and performs image quality inspection;

[0073] Specifically, the full-color image and the scene depth image are collected by the risk identification and inspection equipment; wherein the risk identification and inspection equipment includes a full-color camera and a TOF camera, which respectively collect the scene color image and the scene depth image;

[0074] The displacement deviation between the full-color image and the scene depth image is corrected through an image alignment algorithm, so that the two images correspond one to one, the image information is merged, and an RGB-D image of potential risk is obtained; wherein the image alignment algorithm includes a feature-based method or a pixel-based method.

[0075] It should be noted that since the RGB image and the depth image may be captured at different times or under different conditions, there may be a certain displacement deviation between the two. The image alignment algorithm can be used to combine the color information of the RGB image with the spatial information of the depth image to generate a color image with depth information and obtain an RGB-D image.

[0076] The process of image quality inspection for potential risk RGB-D images is as follows:

[0077] Step S1: Verify whether the image shooting time is consistent with the inspection task time, and verify whether the image GPS coordinates are consistent with the risk unit location; if yes, proceed to step 2; if no, reshoot;

[0078] Step S2: Perform quality inspection on the collected images;

[0079] Quality inspection of image data formats: converting images to a unified image format; such as removing duplicates, blank images, etc.;

[0080] The quality inspection of image data includes image size and resolution, image color mode, image clarity and contrast;

[0081] Image size and resolution: resize the images to the same size and resolution;

[0082] Image color mode: Check whether the color image is in RGB color mode; if yes, do not adjust the image color; if not, use automated tools to adjust the image to RGB color mode;

[0083] Image clarity and contrast: The edge strength of the image is analyzed through an edge detection algorithm and compared with the clarity threshold to evaluate the image clarity; edge detection algorithms include Sobel operator and Canny edge detection.

[0084] Analyze the contrast of the image through the image histogram;

[0085] Quality inspection of image data integrity: Perform integrity checks on potential risk items at each risk point, and confirm that each object or equipment facility has images from multiple angles.

[0086] The risk assessment module inputs the RGB-D image of potential risks into the corresponding risk identification model to identify the risk coefficient of potential risks.

[0087] Among them, the risk identification model is trained based on the Mask R-CNN model;

[0088] Specifically, a number of potential risk images are extracted from historical data, and each abnormal feature in the potential risk image is marked, and different abnormal features are marked with different data numbers;

[0089] The marked potential risk images are used as standard input data for the Mask R-CNN model;

[0090] The data number of the abnormal features is used as the standard output data of the Mask R-CNN model;

[0091] The Mask R-CNN model is trained using standard input data and standard output data to obtain a risk identification model.

[0092] For example, taking "pipeline damage risk" as an example, the feature description is whether the pipeline has cracks, holes, and corrosion; the training data is annotated with the locations of cracks, holes, and external mold damage and corrosion;

[0093] Collect RGB-D image samples of several pipelines, annotate each sample image to obtain a training data set, and mark the characteristics of pipeline rupture with data numbers, such as cracks marked as 1; divide the data set into training set, validation set, and test set to ensure the effectiveness of model training and evaluation;

[0094] It should be noted that different pipeline features are marked with different data numbers, such as cracks are marked as 1, holes are marked as 2, and corrosion is marked as 3; assuming that there are N cracks in a potential image, the N locations are marked as the same natural value.

[0095] Through the training set, validation set, and test set Mask R-CNN model input, the cross entropy loss function was used to optimize the model. The loss value and evaluation indicators such as accuracy and recall during the training process were monitored to ensure that the model can accurately identify potential risks.

[0096] For risk points where risks occur, the school needs to make timely rectifications to eliminate the risks. After the risks are eliminated, the risk point data will be updated, marked as a normal data set, and put into the model upgrade training data set.

[0097] and, analyzing the risk factors of potential risks;

[0098] Specifically, the number of each abnormal feature detected from the latent image is counted, and the proportion of each abnormal feature is calculated, that is, the ratio of the number of each abnormal feature to the total number of abnormal features;

[0099] The risk coefficient FXS of potential risk is calculated by the formula FXS=∑(αi×YBi); wherein αi is the weight coefficient, αi>0, ∑αi=1; YBi is the proportion of abnormal features; i is the type of abnormal feature; ∑ is the sum of i.

[0100] Compare the risk factor with the preset threshold to assess the degree of danger of the potential risk. For example, if the risk factor is less than the preset threshold, the potential risk is low and maintenance measures should be taken in a timely manner; if the risk factor is greater than the preset threshold, the potential risk is high and needs to be handled in a timely manner to avoid causing more serious harm.

[0101] The present invention evaluates the danger level of potential risks through risk coefficients; the danger level of potential risks is mainly related to the proportion of each abnormal feature of the potential risk and the weight coefficient corresponding to each abnormal feature. If the proportion of certain abnormal features is high and the corresponding abnormal features are relatively large, it indicates that the potential risk has more abnormal features and a higher danger level, that is, the risk coefficient is large.

[0102] After completing the inspection of all risk points, the risk reporting module outputs an inspection report, such as Figure 2 According to the inspection report, by understanding the status of each risk point and its detailed information, it is possible to determine which risks are the most urgent or important and deal with them in priority.

[0103] See also Figure 3 The second aspect of the present invention proposes a method for identifying dual prevention risks on campus, comprising the following steps:

[0104] Step Q1: First select the risk unit to be checked, and then select the subordinate risk points;

[0105] Step Q2: After selecting the risk point, the risk identification and inspection device reads the potential risk list corresponding to the corresponding risk point from the campus dual prevention risk knowledge base, as well as the inspection standard corresponding to each potential risk, and performs a safety inspection on the risk point according to the inspection standard; wherein the risk identification and inspection device includes a full-color camera and a TOF camera, which respectively collect scene color images and scene depth images;

[0106] Step Q3: Use risk identification inspection equipment to collect data according to the inspection standards, usually by collecting full-color images and scene depth images of items or equipment facilities with potential risks from all directions; when collecting data, full-color images and depth images are collected simultaneously to obtain RGB-D images of potential risks;

[0107] Step Q4: Perform image quality inspection on the RGB-D image with potential risks; wherein the image quality inspection includes quality inspection of image data format, quality inspection of image data quality and quality inspection of image data integrity;

[0108] Step Q5: Identify the risk features of the RGB-D image of potential risk through a risk identification model; wherein the risk identification model is constructed based on a CNN neural network model;

[0109] Step Q6: Analyze the risk factor of potential risks and assess the risk level of potential risks.

[0110] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0111] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A campus dual prevention risk identification system, characterized in that: It includes an image acquisition and processing module, and a risk knowledge base module and a risk assessment module connected thereto; Risk knowledge base module: used to define campus risk information and the relationship between risk information, and to build a campus dual prevention risk knowledge base; the risk information includes the campus risk units, risk points, potential risks corresponding to the risk points, and inspection standards corresponding to the potential risks; Image acquisition and processing module: used to collect multi-directional full-color images and scene depth images of potential risk locations, fuse them to obtain RGB-D images of potential risks, and perform image quality inspection; among them, image quality inspection includes quality inspection of image data format, quality inspection of image data quality, and quality inspection of image data integrity; Risk assessment module: used to input the RGB-D image of potential risk into the corresponding risk identification model to identify risk characteristics and analyze the risk coefficient of potential risk; among which, the risk identification model is built based on the CNN neural network model.

2. A campus dual prevention risk identification system according to claim 1, characterized in that: The definition of campus risk information and the relationship between risk information include: Define risk units: Encode the physical spaces on campus in the knowledge base; Define risk points: Risk points are subordinate units of risk units. Defining risk points includes defining and coding in the knowledge base and selecting the risk units to which they belong; Define the potential risks of risk points and define the inspection criteria for risk points.

3. A campus dual prevention risk identification system according to claim 2, characterized in that: The construction of the campus dual prevention risk knowledge base includes: Abstract the physical space of the school into the database, and mark the risk units, risk points, potential risks corresponding to the risk points, potential risk points and inspection standards of the campus as entities of the knowledge graph; Extract the risk units of the campus, the risk points corresponding to the risk units, the potential risks corresponding to the risk points, and the inspection standards corresponding to the potential risk points from the expert knowledge base; use the one-to-one correspondence between the risk units, risk points, potential risks and inspection standards as the connection relationship between the knowledge graph entities; By constructing the entity-relationship-entity triple data group of the knowledge graph, the campus dual prevention risk knowledge base is obtained.

4. A campus dual prevention risk identification system according to claim 1, characterized in that: The method of collecting multi-directional full-color images and scene depth images of potential risk locations and fusing them to obtain RGB-D images of potential risks includes: The full-color image and the scene depth-of-field image are collected by the risk identification inspection device; wherein the risk identification inspection device includes a full-color camera and a TOF camera, which respectively collect the scene color image and the scene depth-of-field image; The displacement deviation between the full-color image and the scene depth image is corrected through an image alignment algorithm, so that the two images correspond one to one, the image information is merged, and an RGB-D image of potential risk is obtained; wherein the image alignment algorithm includes a feature-based method or a pixel-based method.

5. A campus dual prevention risk identification system according to claim 4, characterized in that: Perform image quality inspection on RGB-D images with potential risks, including: Step S1: Verify whether the image shooting time is consistent with the inspection task time, and verify whether the image GPS coordinates are consistent with the risk unit location; if yes, proceed to step 2; if no, reshoot; Step S2: Perform quality inspection on the collected images; Quality inspection of image data format: converting images into a unified image format; The quality inspection of image data includes image size and resolution, image color mode, image clarity and contrast; Image size and resolution: resize the images to the same size and resolution; Image color mode: Check whether the color image is in RGB color mode; if yes, do not adjust the image color; if no, adjust the image to RGB color mode; Image clarity and contrast: The edge detection algorithm is used to analyze the edge strength of the image and compare it with the clarity threshold to evaluate the image clarity; Analyze the contrast of the image through the image histogram; Quality inspection of image data integrity: Perform integrity checks on potential risk items at each risk point, and confirm that each object or equipment facility has images from multiple angles.

6. A campus dual prevention risk identification system according to claim 5, characterized in that: The risk identification model is constructed based on the CNN neural network model, including: Extracting several potential risk images from historical data, and marking each abnormal feature in the potential risk images, and marking different abnormal features with different data numbers; The marked potential risk images are used as standard input data for the CNN neural network model; The data number of the abnormal features is used as the standard output data of the CNN neural network model; The CNN neural network model is trained using standard input data and standard output data to obtain a risk identification model.

7. A campus dual prevention risk identification system according to claim 6, characterized in that: The risk factor of analyzing potential risks includes: Count the number of abnormal features detected from the potential image, and calculate the proportion of each abnormal feature, that is, the ratio of the number of each abnormal feature to the total number of abnormal features; The risk coefficient FXS of potential risk is calculated by the formula FXS=∑(αi×YBi); wherein αi is the weight coefficient, αi>0, ∑αi=1; YBi is the proportion of abnormal features; i is the type of abnormal feature; ∑ is the sum of i.

8. A campus dual prevention risk identification system according to claim 1, characterized in that: The risk report module is used to output the inspection report after completing the inspection of all risk points.

9. A method for identifying dual prevention risks on campus, characterized in that: The following steps are involved: Step Q1: First select the risk unit to be checked, and then select the subordinate risk points; Step Q2: After selecting the risk point, the risk identification and inspection device reads the potential risk list corresponding to the corresponding risk point from the campus dual prevention risk knowledge base, as well as the inspection standard corresponding to each potential risk, and performs a safety inspection on the risk point according to the inspection standard; wherein the risk identification and inspection device includes a full-color camera and a TOF camera, which respectively collect scene color images and scene depth images; Step Q3: Use risk identification inspection equipment to collect data according to the inspection standards, usually by collecting full-color images and scene depth images of items or equipment facilities with potential risks from all directions; when collecting data, full-color images and depth images are collected simultaneously to obtain RGB-D images of potential risks; Step Q4: Perform image quality inspection on the RGB-D image with potential risks; wherein the image quality inspection includes quality inspection of image data format, quality inspection of image data quality and quality inspection of image data integrity; Step Q5: Identify the risk features of the RGB-D image of potential risk through a risk identification model; wherein the risk identification model is constructed based on a CNN neural network model; Step Q6: Analyze the risk factor of potential risks and assess the risk level of potential risks.

Images