CN115660297A - Automatic AI early warning system and method for construction site safety - Google Patents
Automatic AI early warning system and method for construction site safety Download PDFInfo
- Publication number
- CN115660297A CN115660297A CN202110770221.7A CN202110770221A CN115660297A CN 115660297 A CN115660297 A CN 115660297A CN 202110770221 A CN202110770221 A CN 202110770221A CN 115660297 A CN115660297 A CN 115660297A
- Authority
- CN
- China
- Prior art keywords
- early warning
- analysis
- safety
- module
- analysis module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The application discloses automatic AI early warning system of building site safety and method, system include streaming media server, AI safety analysis platform, security incident early warning engine, intelligent visual situation perception center and spread over construction site's basic station, supervisory equipment, streaming media server is with video transcoding for rtsp video stream, AI safety analysis platform snatchs newly-increased or take place environmental change's rtsp video stream in the monitoring area, and the algorithm of calling simultaneously carries out analysis and processing, security incident early warning engine carries out logic analysis to the analysis result, final definite early warning information, intelligent visual situation perception center carries out the early warning to user terminal to regularly carry out risk assessment, visual show and statistical analysis. The invention realizes real-time analysis of monitoring videos, capturing of problem pictures and early warning through AI loading capacity, realizes automatic detection and alarm of construction site safety problems, and has more comprehensive safety supervision coverage and higher inspection efficiency.
Description
Technical Field
The invention belongs to the technical field of construction site safety early warning, and particularly relates to a construction site safety automatic AI early warning system and a method.
Background
The application of artificial intelligence in various social fields is gradually increased, and the AI algorithm is gradually developed into daily application of people in various key technologies such as continuous iterative updating, deep learning, computer vision and the like, such as face recognition. However, the general application of artificial intelligence is still only in common scenes, such as daily life, express delivery, transportation and the like, and the application is not common in the production fields of many special professions, such as construction sites, mines, oil plants and the like, and in the actual environments of the professions, the automatic instant early warning of safety problems is almost difficult to realize. Information technology integration application capabilities of BIM, big data, intellectualization, mobile communication, cloud computing, internet of things and the like are put forward and enhanced in a '2016-2020 building industry informatization development outline' issued by the ministry of construction.
At present, the safety problem supervision of a construction site is mainly realized by a mode of manual monitoring and manual inspection, and the method comprises two methods: firstly, the problem is discovered by a mode that a person inspects inspection problems or adds workers to supervise each other, and secondly, a video monitoring center and a supervisor are set up, and the worker observes the monitored video to discover the problem and reports the problem.
The method comprises the steps that personnel inspection or mutual supervision of workers is carried out, the personnel inspection is periodic and periodical, potential safety hazards and problems of a construction site cannot be inspected in real time, and the problems that an inspection area is difficult to completely cover, inspection personnel are loosened and the like exist in the inspection; the mutual supervision of workers mainly examines the self-consciousness of the workers and is difficult to ensure the objectivity and the enthusiasm of the mutual supervision of the workers. Therefore, no matter the mode of the personnel inspection or the mode of the mutual supervision of the personnel inspection auxiliary workers, the problems of incomplete safety problem coverage and untimely report exist.
And a monitoring center and monitoring personnel are set, so that the requirement on the monitoring personnel is high, the number of the required personnel is large, the energy consumption is high when the user watches the video all the time, the situations that the safety problem is not found and the like are difficult to avoid, and the problems that the safety problem coverage is incomplete, the report is not in time and the like exist.
In addition, the two methods are both manually checked, the requirement on inspectors is high, the cost is high in the aspect of personnel investment, all-weather safety supervision cannot be realized, and the method is not beneficial to all-around safety production of a construction site.
Disclosure of Invention
Based on the above, the invention aims to provide the automatic AI early warning system for the construction site safety, which has the advantages of simple structure, reasonable design and convenient use, aiming at the defects and shortcomings of the prior art, and the automatic AI safety instant early warning system for the construction site is constructed on the basis of deep learning and intelligent visual analysis on the premise of fully integrating the complex environment of the construction site and artificial intelligence.
The application discloses an automatic AI early warning system of building site safety, including streaming media server, AI safety analysis platform, security incident early warning engine, intelligent visual situation perception center and all over the basic station, the supervisory equipment of job site, streaming media server transcodes the video into rtsp video stream, then distributes to AI safety analysis platform, AI safety analysis platform snatchs the rtsp video stream that the video content changes in the monitoring area, calls the algorithm simultaneously and carries out analysis and processing to transmit the analysis result to security incident early warning engine, security incident early warning engine carries out logic analysis to the analysis result, finally confirms early warning information, and transmits early warning information for intelligent visual situation perception center, intelligent visual situation perception center carries out the early warning to user terminal to regularly carry out risk assessment, visual show and statistical analysis.
Further, the AI safety analysis platform comprises six analysis modules, specifically a flame analysis module, a safety helmet analysis module, a reflective garment analysis module, a smoke analysis module, a mask analysis module and a perimeter analysis module; each analysis module is provided with an independent AI algorithm and is stored in a database, and the flame analysis module is used for monitoring whether micro flames exist in each area in the construction site or not; the safety helmet analysis module, the reflective garment analysis module and the mask analysis module are used for monitoring whether workers correctly wear safety helmets, reflective garments and masks when entering a construction area; the smoke analysis module is used for monitoring whether smoke occurs or not by capturing images of static characteristics and dynamic characteristics of the smoke; the perimeter analysis module is used for monitoring the critical area of the dangerous perimeter in the construction site and whether field personnel have illegal out-of-range behaviors.
Furthermore, six functional modules are arranged in the intelligent visual situation perception center and comprise a security event management module, a mobile phone early warning module, a security situation module, a risk evaluation module, a statistical analysis module and an emergency treatment module, wherein the security event management module is used for managing user permission levels, the mobile phone early warning module is used for receiving early warning information and positioning employees, the risk evaluation module is used for carrying out danger rating on the early warning information, the statistical analysis module is used for counting the times of the early warning information, the danger rating and the type of the early warning information, and the emergency treatment module is used for matching the received early warning information and codes of a monitoring area with managers and acousto-optic equipment IP preset in a database and carrying out reminding and early warning through acousto-optic equipment.
Furthermore, the security event management module is used for classifying the perfect user permission levels, distributing the permissions of the managers according to the positions of the managers and the differences of the responsible fields of the managers, and opening other corresponding functional modules to the users by identifying the permission of the current user.
Further, the security situation module carries out multi-dimensional classification processing on the information on the security events counted by the statistic analysis module, and carries out trend analysis and display.
On the other hand, the disclosed construction site safety automatic AI early warning method comprises the following steps:
s100, capturing a video of a monitoring area by a streaming media server, transcoding the video into an rtsp video stream, and distributing the rtsp video stream to an AI (Artificial intelligence) security analysis platform;
s200, the AI safety analysis platform obtains video frames according to rtsp video streams newly added or subjected to environmental change in a monitored area pulled from the streaming media server, marks identification information on the video frames, and then matches an analysis module for analysis according to the identification information to generate a response detection data packet, wherein the analysis module comprises a flame analysis module, a safety helmet analysis module, a reflective coat analysis module, a smoke analysis module, a mask analysis module and a perimeter analysis module; the identification information is divided into six types, namely flame, safety helmet, reflective clothes, smog, mask and personnel leaving boundary;
s300, the security event early warning engine receives the response detection data packet, performs logic analysis on an analysis result, and finally determines early warning information;
s400, the intelligent visual situation perception center receives the early warning information, carries out early warning on the user terminal, and carries out risk assessment, visual display and statistical analysis regularly.
Further, step S200 specifically includes:
s201, starting an AI analysis service of an AI safety analysis platform 2, and pulling an rtsp video stream with changed video content in a monitoring area from the streaming media server 1;
specifically, a frame difference method is used for extracting a target object, and the target object is distinguished from the background; the frame difference method is characterized in that an image sequence between adjacent frames is used for operation, an absolute value of a gray value of a pixel point of an image is obtained, the absolute value is compared with a threshold value, so that target object motion information is obtained, the target object is extracted, and the image difference value of the adjacent frames is calculated as follows:
D(x，y，i)＝|I(x，y，i+1)-I(x，y，i)|
d (x, y, I) represents a differential value of an image, I (x, y, I + 1) represents a gray value of an I +1 th frame, I (x, y, I) represents a gray value of an I-th frame, and when a pixel value of a certain point of the differential value of the image is greater than a preset threshold value, the point is a target object, otherwise, the point belongs to a background;
s202, analyzing an rtsp video stream to obtain a video frame; marking one or more corresponding identification information on the video frame, and correspondingly presetting an alarm threshold q of the identification information;
s203, simultaneously sending an analysis request to one or more corresponding analysis modules according to the identification information, and calling the analysis modules to analyze the video stream; obtaining a detection score of the video stream;
the specific method for obtaining the detection score comprises the following steps: extracting a frame picture of a video stream image;
dividing the whole frame into a series of small lattices;
selecting a changed object target as a feature extraction target on the basis of the divided small grids, and analyzing the feature extraction target according to the analysis module corresponding to the identification information to obtain a detection score;
s204, returning the detection score and the identification information of the video stream;
s205, comparing the detection score with an alarm threshold q, and only keeping the detection score which is more than or equal to the alarm threshold q;
and S206, generating a response detection data packet and completing video stream analysis.
Further, the step S300 specifically includes:
s301, the safety event early warning engine analyzes a response detection data packet generated by the AI safety analysis platform to generate a snapshot picture, acquire identification information and a detection score;
s302, judging an early warning type, inputting a plurality of continuous snap pictures with the same type, calling a corresponding logic model, judging whether the logic model rule is met, if not, marking misjudgment, returning to a model training library, and if so, outputting the snap pictures; the logic model refers to a fusion algorithm of target detection and image classification;
s303, overlaying DOS information on the snapshot picture, and adding a problem box and an alarm type icon;
and S304, generating an alarm picture and early warning information interpreted correspondingly.
Further, the step S400 specifically includes:
s401, the intelligent visual situation perception center receives early warning information and extracts early warning information categories, a risk evaluation module built in the intelligent visual situation perception center carries out risk rating according to detection scores, and the early warning information categories are divided into three categories: smoke and flames, safe dressing, perimeter loitering; the danger rating is classified into emergency and reminding;
s402, judging the occurrence region and the type of the early warning information, and calling codes of the early warning type and the monitoring region to be matched with an acousto-optic device IP preset in a database; the acousto-optic equipment warns or reminds, and the IP of the acousto-optic equipment is bound in the fixed monitoring area; simultaneously, a plurality of base stations in an early warning information generation area transmit broadcasts to determine the positions of nearest employees;
s403, the mobile phone early warning module sends the early warning message to corresponding managers; and when the early warning information types are smoke and flame, sending the early warning information to the staff nearest to the early warning information generation area, and reminding the nearest staff of going to detection.
Further, a specific method for determining the position of the nearest employee is as follows: a plurality of base stations in the early warning information generation area transmit broadcast, and the spatial position point of each base station is D (x) n ，y n ) Obtaining the spatial position point P (x) of all the employees in the area m ，y m ) And calculating the relative spatial position P (X) of each employee and each base station m ，Y m ) By calculating min (X) m +Y m ) Thereby identifying the nearest employee, and calculating the relative spatial position formula for each employee as follows:
has the beneficial effects that: 1. the automatic detection of the safety problem of the construction site is realized: real-time analysis of monitoring videos, problem picture capture and early warning are achieved through AI loading capacity, automatic detection and alarming of construction site safety problems are achieved, safety supervision coverage is more comprehensive, and inspection efficiency is higher;
2. the multi-monitoring capability of one camera is realized: the AI analysis capability is arranged at the platform side in a rear mode, a plurality of AI identification capabilities can be applied to a plurality of monitoring videos according to requirements, the AI identification capability of a monitoring area can be adjusted at any time, and a new AI algorithm can be continuously updated or superposed to realize the maximum utilization of resources;
3. safety precaution is more timely, the terminal coverage is more comprehensive: the multi-type terminals act on different areas and personnel through the computer terminal, the mobile phone terminal and the voice control equipment terminal, so that the safety early warning is more timely, and the reminding and treatment effects are better;
4. more be fit for the building site scene: the safety problem of adaptation building site environment and building site can be matchd more, and recognition efficiency and success rate are higher.
Drawings
The embodiments described below with reference to the drawings are illustrative and intended to explain and illustrate the present invention and should not be interpreted as limiting the scope of the invention.
FIG. 1 is a block diagram of the modules of the present invention;
FIG. 2 is a flow chart of the main steps of the present invention;
FIG. 3 is an analysis block diagram of the AI security analysis platform of the present invention;
FIG. 4 is a logic decision block diagram of a security event early warning engine according to the present invention;
FIG. 5 is a flow chart of the training of the logical model of the present invention;
fig. 6 is a diagram of an early warning release of an intelligent visual situation awareness center according to the present invention.
Description of the reference numerals:
1. a streaming media server; 2. an AI security analysis platform; 3. a security event early warning engine; 4. an intelligent visual situation perception center; 2-1, a flame analysis module; 2-2, a safety helmet analysis module; 2-3, a reflective garment analysis module; 2-4, a smoke analysis module; 2-5, a mask analysis module; 2-6, a perimeter analysis module; 4-1, a security event management module; 4-2, a mobile phone early warning module; 4-3, a security situation module; 4-4, a risk assessment module; 4-5, a statistical analysis module; 4-6 and an emergency handling module.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in more detail with reference to fig. 1 to 6.
The first embodiment of the application discloses a building site safety automatic AI early warning system, as shown in fig. 1, including a streaming media server 1, an AI safety analysis platform 2, a safety event early warning engine 3, an intelligent visual situation awareness center 4, and base stations and monitoring devices distributed on a construction site, the streaming media server 1 transcodes videos into rtsp video streams and distributes the rtsp video streams to the AI safety analysis platform 2, the AI safety analysis platform 2 captures the rtsp video streams with video contents changing in a monitoring area, simultaneously calls an algorithm to analyze and process, and transmits an analysis result to the safety event early warning engine 3, the safety event early warning engine 3 performs logic analysis on the analysis result, finally determines early warning information, and transmits the early warning information to the intelligent visual situation awareness center 4, and the intelligent visual situation awareness center 4 performs early warning on a user terminal, and performs risk assessment, visual display and statistical analysis regularly.
Further, the AI safety analysis platform 2 comprises six analysis modules, specifically a flame analysis module 2-1, a safety helmet analysis module 2-2, a reflective garment analysis module 2-3, a smoke analysis module 2-4, a mask analysis module 2-5 and a perimeter analysis module 2-6; each analysis module is provided with an independent AI algorithm and is stored in a database, and the flame analysis module 2-1 is used for monitoring whether micro flames exist in each area in a construction site; the safety helmet analysis module 2-2, the reflective garment analysis module 2-3 and the mask analysis module 2-5 are used for monitoring whether workers correctly wear safety helmets, regularly wear reflective garments and correctly wear masks when entering a construction area; the smoke analysis module 2-4 is used for monitoring whether smoke occurs or not by capturing images of static characteristics and dynamic characteristics of the smoke; the perimeter analysis module 2-6 is used for monitoring dangerous perimeter borderline areas in the construction site and whether field personnel violate the behaviors of going out of bounds or not.
Specifically, the method comprises the following steps: when the safety helmet analysis module 2-2 monitors that a person enters a construction area, if the person does not correctly wear a safety helmet and regularly wears reflective clothes, the person is pushed to a safety supervisor in the corresponding area through the intelligent visual situation perception center 4, and sound equipment closest to the area immediately sends a notice that people in the XX area are related to pay attention to safety and wear the safety helmet/reflective clothes.
A monitoring camera is installed in a critical area of a dangerous perimeter in a construction site, and when the perimeter analysis modules 2-6 monitor that the field personnel have illegal behaviors, the intelligent visual situation perception center 4 immediately informs relevant management personnel and gives out acousto-optic alarm. For example, at the near-edge opening of the operation area, if someone approaches the monitoring critical, if any part of the body exceeds the critical line, an audible and visual alarm is sent out to remind and inform the safety manager that 'someone operates in a dangerous area, please pay attention immediately'; in the material accumulation area, if a person is in a specified critical loitering area, an audible and visual alarm is immediately sent out, and a material manager is informed to check the situation as soon as possible;
when the smoke analysis module 2-4 monitors that the working face generates images similar to static and dynamic smoke characteristics, the intelligent visual situation perception center 4 immediately gives an alarm, directly sounds an alarm and informs project management personnel;
when the flame analysis module 2-1 monitors areas such as a working face, a worker dormitory, a stockpiling area and the like and micro flames exist, the intelligent visual situation perception center 4 immediately takes a snapshot and gives an alarm.
Furthermore, six functional modules are arranged in the intelligent visual situation awareness center 4, and the functional modules comprise a security event management module 4-1, a mobile phone early warning module 4-2, a security situation module 4-3, a risk assessment module 4-4, a statistical analysis module 4-5 and an emergency treatment module 4-6, wherein the security event management module 4-1 is used for managing user permission levels, the mobile phone early warning module 4-2 is used for receiving early warning information and positioning employees, the risk assessment module 4-4 is used for carrying out risk rating on the early warning information, the statistical analysis module 4-5 is used for carrying out statistics on the times, risk rating and early warning information types of the early warning information, and the emergency treatment module 4-6 is used for matching the received early warning information and codes of a monitoring area with managers and acousto-optic devices IP preset in a database, and carrying out reminding and early warning through acousto-optic devices.
Furthermore, the security event management module 4-1 is used for perfect user permission level classification, performing permission allocation on managers according to the positions of the managers and the differences of the responsible fields of the managers, and opening other corresponding functional modules to the users by identifying the permission of the current user, and meanwhile, the security event management module 4-1 gives the project responsible person management permission so that the authority of the managers can be modified and updated autonomously according to actual conditions and specific requirements.
Further, the security situation module 4-3 performs multidimensional classification processing on the information about the security events counted by the statistical analysis module 4-5, and performs trend analysis and display. Specifically, a report form is generated, and related software or a system is arranged on each employee smart phone and is used for each employee to check.
A second embodiment discloses a method for automatically early warning AI of safety of a construction site, as shown in fig. 2, including the following steps:
s100, the streaming media server 1 captures a video in a monitoring area, transcodes the video into an rtsp video stream and distributes the rtsp video stream to the AI safety analysis platform 2;
s200, the AI safety analysis platform 2 acquires video frames according to rtsp video streams newly added or subjected to environmental change in a monitored area pulled from the streaming media server 1, marks identification information on the video frames, and then matches an analysis module for analysis according to the identification information to generate a response detection data packet, wherein the analysis module comprises a flame analysis module 2-1, a safety helmet analysis module 2-2, a reflective garment analysis module 2-3, a smoke analysis module 2-4, a mask analysis module 2-5 and a perimeter analysis module 2-6; the identification information is divided into six types, namely flame, safety helmet, reflective clothes, smog, mask and personnel leaving boundary; for example: when an rtsp video stream received by the AI safety analysis platform 2 is about a person leaving boundary, marking identification information of the person leaving boundary on the rtsp video stream, then matching with a perimeter analysis module 2-6 for analysis, fuzzily confirming information that the person leaving boundary possibly occurs through a deep learning algorithm, and simultaneously performing detection scoring according to actions of related persons, wherein the scoring standard is between 0 and 100, the scoring standard can select factors such as distance from the person leaving the dangerous boundary, movement speed, movement amplitude, weather condition, video definition and the like, and each factor is provided with corresponding weight, so that evaluation is performed.
S300, the security event early warning engine 3 receives the response detection data packet, performs logic analysis on the analysis result, and finally determines early warning information;
s400, the intelligent visual situation awareness center 4 receives the early warning information, carries out early warning on the user terminal, and carries out risk assessment, visual display and statistical analysis regularly.
Further, according to the block diagram shown in fig. 3, step S200 may specifically be:
s201, starting an AI analysis service of an AI safety analysis platform 2, and pulling an rtsp video stream which is newly added or has environmental change in a monitoring area from the streaming media server 1; specifically, a frame difference method is used for extracting a target object, and the target object is distinguished from the background; the frame difference method is characterized in that an image sequence between adjacent frames is used for operation, absolute values of gray values of pixel points of an image are obtained, and then the absolute values are compared with a threshold value to obtain target object motion information, so that the target object is extracted, and the image difference value of the adjacent frames is calculated as follows:
D(x，y，i)＝|I(x，y，i+1)-I(x，y，i)|
d (x, y, I) represents a difference value of the image, I (x, y, I + 1) represents a gray value of an I +1 th frame, and I (x, y, I) represents a gray value of an I-th frame, when a pixel value of a certain point of the difference value of the image is greater than a preset threshold value, the point is a target object, otherwise, the point belongs to the background.
When the video of the monitored area is in a static state, the extraction is not carried out, so that the calculation power is saved. Meanwhile, video identification types monitored in different areas of a construction site can be set, for example, at an entrance and an exit, problems such as safety helmets, safety clothes and the like can be identified independently; or monitoring where the perimeter of the worksite only identifies personnel leaving the boundary problem.
S202, analyzing an rtsp video stream to obtain a video frame; marking one or more corresponding identification information on the video frame, specifically, by requesting a database, then comparing the video frame with the information of the database and marking the identification information, and correspondingly presetting an alarm threshold q of the identification information; the alarm threshold q may be set according to actual conditions, for example: the alarm threshold q of a fire or smoke may be set low because of its high hazard, and may be set high, for example, in the case of clothing.
S203, judging the type of the identification information, simultaneously sending an analysis request to one or more corresponding analysis modules, and calling the analysis modules to analyze the video stream; obtaining a detection score of the video stream;
the specific method for obtaining the detection score comprises the following steps: extracting the frame picture of the video stream at the time t, marked as F t ；
Dividing the whole frame picture into a series of small lattices; and selecting a certain changed object target as a feature extraction target on the basis of the divided small grids, and analyzing the feature extraction target according to the analysis module corresponding to the identification information to obtain a detection score.
For example: when the perimeter analysis module 2-6 analyzes that the person is out of bound, assuming that the target A is a particle, the pixel coordinate At of the position where the particle is located At the time t is (xt, yt), and the pixel coordinate At + τ of the position where the particle corresponding to the frame Ft + τ At the time t + τ is (xt + τ, yt + τ) At the elapsed time τ; the boundary coordinate limiting condition is x =1, y =1, whether a pixel coordinate (xt + τ, yt + τ) exceeds the boundary coordinate is judged, then each particle point of a target object in a video picture frame is processed one by one to obtain a corresponding position relation, then whether the particle point exceeds the boundary is determined according to the comparison of the pixel coordinate and the boundary coordinate, and then a detection score is obtained according to the proportion of pixel points exceeding the boundary coordinate and the total pixel points of the target object. When the flame analysis module 2-1 analyzes the flame, only the time-varying intensity of the pixel points at a certain fixed position needs to be analyzed, and then the detection score is given according to the varying intensity of the gray value.
S204, returning detection scores and identification information of the video stream;
s205, comparing the detection score with an alarm threshold q, and only keeping the detection score which is more than or equal to the alarm threshold q;
and S206, generating a response detection data packet to finish video stream analysis.
Further, according to the block diagram shown in fig. 4, the step S300 specifically includes:
s301, the safety event early warning engine 3 analyzes a response detection data packet generated by the AI safety analysis platform 2 to generate a snapshot picture, and acquires identification information and a detection score;
s302, judging an early warning type, inputting a plurality of continuous snap pictures with the same type, calling a corresponding logic model, judging whether the logic model rule is met, namely judging whether the confidence coefficient is more than 95%, if not, marking misjudgment, returning to a model training library, and if so, outputting the snap pictures;
the logic model adopts target detection algorithms such as the idea of YOLO, non-max compression, and anchor box of fast R-CNN to position the position and the type of the target object, and then adopts a classification model to finely classify the target object; and analysis result logic fault-tolerant processing-target classification comparison switch: when the classification comparison switch is turned on, early warning events with extremely high feature similarity in the same place can be filtered within a self-defined time interval, so that the model precision is improved; in order to deal with the complex and changeable environment of a construction site, the data set is marked and cleaned manually; in the initial stage, normal violation events cannot meet the data size required by the deep learning model, and the corresponding data size is increased by simulating the violation events deep into a construction site, so that the target positive and negative sample proportion is balanced, the misinformation is reduced, the precision is improved, and the robustness of the model is enhanced.
According to the block diagram shown in fig. 5, the specific training process of the logic model is as follows: and selecting an identification area of the image, matching the existing identification model according to the selected image identification area, and if the existing identification model is matched with the identification area, directly calling the selected identification model, otherwise, learning a new model.
And (3) new model learning, namely firstly acquiring and simulating target behaviors on site, then carrying out manual labeling and cleaning, secondly selecting a reference model, preparing training data and a responsive data label, and training a convolutional neural network to obtain a training image classifier. Meanwhile, each new training model is stored to obtain a training result logic model.
S303, overlaying DOS information on the snapshot picture, and adding a problem frame and an alarm type icon;
and S304, generating an alarm picture and early warning information interpreted correspondingly.
Further, according to the block diagram shown in fig. 6, the step S400 specifically includes the steps of:
s401, the intelligent visual situation awareness center 4 receives early warning information and extracts early warning information categories, a risk evaluation module 4-4 arranged in the intelligent visual situation awareness center 4 carries out risk rating according to detection scores, and the early warning information categories are divided into three categories: smoke and flames, safe dressing, perimeter wandering; the danger rating is classified into emergency and reminding;
s402, judging the occurrence region and the type of the early warning information, and calling codes of the early warning type and the monitoring region to be matched with an acousto-optic device IP preset in a database; the acousto-optic equipment warns or reminds, and the IP of the acousto-optic equipment is bound in the fixed monitoring area; simultaneously, a plurality of base stations in an early warning information generation area transmit broadcasts to determine the positions of nearest employees;
s403, the mobile phone early warning module 4-2 sends the early warning message to corresponding managers; when the early warning information category is smoke and flame, notifying project management personnel, and simultaneously sending the early warning information to staff nearest to an early warning information generation area, for example: and (4) reminding nearest staff of going to detection by project management staff and safety supervision staff. The process flow is recorded and a security management record is generated. And generating a visual report at intervals.
Further, a specific method for determining the position of the nearest employee is as follows: a plurality of base stations in the early warning information generation area transmit broadcast, and the spatial position point of each base station is D (x) n ，y n ) Obtaining the spatial position point P (x) of all the employees in the area m ，y m ) And calculating the relative spatial position P (X) of each employee and each base station m ，Y m ) By calculating min: (X m +Y m ) Thereby identifying the nearest employee, and calculating the relative spatial position formula for each employee as follows:
in the description of the present application, it is to be understood that the terms "center", "longitudinal", "lateral", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like indicate orientations or positional relationships based on those shown in the drawings, and are used merely for convenience in describing the present application and for simplifying the description, and do not indicate or imply that the referenced device or element must have a particular orientation, be constructed in a particular orientation, and be operated, and therefore should not be construed as limiting the scope of the present application.
Although only the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and all changes are included in the scope of the present invention.
Claims (10)
1. The utility model provides an automatic AI early warning system of building site safety which characterized in that: the system comprises a streaming media server (1), an AI security analysis platform (2), a security event early warning engine (3), an intelligent visual situation perception center (4), base stations and monitoring equipment, wherein the base stations and the monitoring equipment are distributed on a construction site, videos are transcoded into rtsp video streams by the streaming media server (1) and then distributed to the AI security analysis platform (2), the RTSP video streams with changed video contents in a monitoring area are grabbed by the AI security analysis platform (2), algorithms are called to analyze and process the rtsp video streams, analysis results are transmitted to the security event early warning engine (3), the security event early warning engine (3) performs logic analysis on the analysis results, early warning information is finally determined and transmitted to the intelligent visual situation perception center (4), and the intelligent visual situation perception center (4) performs early warning on a user terminal and performs risk assessment, visual display and statistical analysis regularly.
2. The automatic AI pre-warning system for worksite safety according to claim 1, characterized in that the AI safety analysis platform (2) comprises six analysis modules, in particular a flame analysis module (2-1), a helmet analysis module (2-2), a reflective coat analysis module (2-3), a smoke analysis module (2-4), a mask analysis module (2-5), a perimeter analysis module (2-6); each analysis module is provided with an independent AI algorithm and is stored in a database, and the flame analysis module (2-1) is used for monitoring whether micro flames exist in each area in the construction site; the safety helmet analysis module (2-2), the reflective garment analysis module (2-3) and the mask analysis module (2-5) are used for monitoring whether workers correctly wear safety helmets, regularly wear reflective garments and correctly wear masks when entering a construction area; the smoke analysis module (2-4) is used for monitoring whether smoke occurs or not by grabbing images of static characteristics and dynamic characteristics of the smoke; and the perimeter analysis module (2-6) is used for monitoring whether illegal out-of-bounds behaviors occur to field personnel in dangerous perimeter border areas in a construction site.
3. The AI early warning system for the safety of the construction site according to claim 1, characterized in that six functional modules are built in the intelligent visual situation perception center (4), including a safety event management module (4-1), a mobile phone early warning module (4-2), a safety situation module (4-3), a risk assessment module (4-4), a statistical analysis module (4-5) and an emergency treatment module (4-6), wherein the safety event management module (4-1) is used for managing user authority levels, the mobile phone early warning module (4-2) is used for receiving early warning information and positioning employees, the risk assessment module (4-4) is used for performing risk rating on the early warning information, the statistical analysis module (4-5) is used for counting the number of times of the early warning information, risk rating and type of the early warning information, and the emergency treatment module (4-6) is used for matching the received early warning information and codes of a monitoring area with managers and acousto-optic devices preset in a database, and performing early warning and warning through acousto-optic devices.
4. The automatic AI pre-warning system for worksite safety according to claim 3, characterized in that the safety event management module (4-1) is configured to perform a complete user authority level classification, assign authorities to managers according to differences between their positions and their areas of responsibility, open corresponding other function modules to users by identifying authorities of current users, and meanwhile, the safety event management module (4-1) gives management authorities to project managers so that they can modify and update authorities of managers autonomously according to actual conditions and specific needs.
5. The worksite safety automatic AI pre-warning system according to claim 4, characterized in that the safety situation module (4-3) performs a multi-dimensional classification process on the information on the safety events counted by the statistical analysis module (4-5), and performs a trend analysis and presentation.
6. An automatic AI early warning method for construction site safety is characterized by comprising the following steps:
s100, a streaming media server (1) captures a video in a monitoring area, transcodes the video into an rtsp video stream and distributes the rtsp video stream to an AI (AI) security analysis platform (2);
s200, the AI safety analysis platform (2) acquires video frames according to an rtsp video stream which is drawn from the streaming media server (1) and changes in video content in a monitored area, marks identification information on the video frames, and then matches an analysis module to analyze according to the identification information to generate a response detection data packet, wherein the analysis module comprises a flame analysis module (2-1), a safety helmet analysis module (2-2), a reflective garment analysis module (2-3), a smoke analysis module (2-4), a mask analysis module (2-5) and a perimeter analysis module (2-6); the identification information is divided into six types, namely flame, safety helmet, reflective clothes, smog, mask and personnel leaving boundary;
s300, receiving the response detection data packet by the security event early warning engine (3), carrying out logic analysis on an analysis result, and finally determining early warning information;
s400, the intelligent visual situation perception center (4) receives the early warning information, early warns the user terminal, and carries out risk assessment, visual display and statistical analysis regularly.
7. The automatic AI early warning method for construction site safety according to claim 6, wherein the step S200 is specifically:
s201, starting an AI analysis service of an AI safety analysis platform (2), and pulling an rtsp video stream with changed video content in a monitoring area from the streaming media server (1);
specifically, a frame difference method is used for extracting a target object, and the target object is distinguished from the background; the frame difference method is characterized in that an image sequence between adjacent frames is used for operation, an absolute value of a gray value of a pixel point of an image is obtained, the absolute value is compared with a threshold value, so that target object motion information is obtained, the target object is extracted, and the image difference value of the adjacent frames is calculated as follows:
D(x，y，i)＝|I(x，y，i+1)-I(x，y，i)|
d (x, y, I) represents a differential value of an image, I (x, y, I + 1) represents a gray value of an I +1 th frame, I (x, y, I) represents a gray value of an I-th frame, and when a pixel value of a certain point of the differential value of the image is greater than a preset threshold value, the point is a target object, otherwise, the point belongs to a background;
s202, analyzing an rtsp video stream to obtain a video frame; marking one or more corresponding identification information on the video frame, and correspondingly presetting an alarm threshold q of the identification information;
s203, simultaneously sending an analysis request to one or more corresponding analysis modules according to the identification information, and calling the analysis modules to analyze the video stream; obtaining a detection score of the video stream;
the specific method for obtaining the detection score comprises the following steps: extracting a frame picture of a video stream image;
dividing the whole frame into a series of small lattices;
selecting a certain changed object target as a feature extraction target on the basis of the divided small grids, and analyzing the feature extraction target according to the analysis module corresponding to the identification information to obtain a detection score;
s204, returning the detection score and the identification information of the video stream;
s205, comparing the detection score with an alarm threshold q, and only keeping the detection score which is more than or equal to the alarm threshold q;
and S206, generating a response detection data packet to finish video stream analysis.
8. The automatic AI pre-warning method for worksite safety according to claim 6, wherein the step S300 specifically includes:
s301, the safety event early warning engine (3) analyzes a response detection data packet generated by the AI safety analysis platform (2) to generate a snapshot picture, acquire identification information and a detection score;
s302, judging an early warning type, inputting a plurality of continuous snap pictures with the same type, calling a corresponding logic model, judging whether the rules of the logic model are met, if not, marking misjudgment, returning to a model training library, and if so, outputting the snap pictures; the logical model refers to a fusion algorithm of target detection and image classification;
s303, overlaying DOS information on the snapshot picture, and adding a problem frame and an alarm type icon;
and S304, generating an alarm picture and early warning information interpreted correspondingly.
9. The automatic AI early warning method for construction site safety according to claim 6, wherein the step S400 includes the following steps:
s401, the intelligent visual situation awareness center (4) receives early warning information and extracts early warning information categories, a risk assessment module (4-4) arranged in the intelligent visual situation awareness center (4) carries out risk rating according to detection scores, and the early warning information categories are divided into three categories: smoke and flames, safe dressing, perimeter loitering; the danger rating is classified into emergency and reminding;
s402, judging the occurrence region and the type of the early warning information, and calling codes of the early warning type and the monitoring region to be matched with an acousto-optic device IP preset in a database; the acousto-optic equipment warns or reminds, and the IP of the acousto-optic equipment is bound in the fixed monitoring area; meanwhile, a plurality of base stations in an early warning information generation area transmit broadcasts to determine the positions of nearest staff;
s403, the mobile phone early warning module (4-2) sends the early warning message to corresponding managers; and when the early warning information types are smoke and flame, sending the early warning information to the staff nearest to the early warning information generation area, and reminding the nearest staff of going to detection.
10. The automatic AI pre-warning method for worksite safety according to claim 9, characterized in that the specific method for determining the location of the nearest staff is: a plurality of base stations in the early warning information generation area transmit broadcast, and the spatial position point of each base station is D (x) n ，y n ) Obtaining the spatial position point P (x) of all the employees in the area m ，y m ) And calculating the relative spatial position P (X) of each employee and each base station m ，Y m ) By calculating min (X) m +Y m ) Thereby identifying the nearest employee, and calculating the relative spatial position formula for each employee as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110770221.7A CN115660297A (en) | 2021-07-07 | 2021-07-07 | Automatic AI early warning system and method for construction site safety |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110770221.7A CN115660297A (en) | 2021-07-07 | 2021-07-07 | Automatic AI early warning system and method for construction site safety |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115660297A true CN115660297A (en) | 2023-01-31 |
Family
ID=85015001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110770221.7A Pending CN115660297A (en) | 2021-07-07 | 2021-07-07 | Automatic AI early warning system and method for construction site safety |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115660297A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433036A (en) * | 2023-06-14 | 2023-07-14 | 北京万赋互联网科技集团有限公司 | Cloud-based labor supervision system and supervision method thereof |
| CN116862244A (en) * | 2023-09-04 | 2023-10-10 | 广东鉴面智能科技有限公司 | Industrial field vision AI analysis and safety pre-warning system and method |
| CN117035378A (en) * | 2023-10-10 | 2023-11-10 | 广州海晟科技有限公司 | Intelligent building site management method and system based on Internet of things |
-
2021
- 2021-07-07 CN CN202110770221.7A patent/CN115660297A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116433036A (en) * | 2023-06-14 | 2023-07-14 | 北京万赋互联网科技集团有限公司 | Cloud-based labor supervision system and supervision method thereof |
| CN116433036B (en) * | 2023-06-14 | 2023-10-20 | 北京万赋互联网科技集团有限公司 | Cloud-based labor supervision system and supervision method thereof |
| CN116862244A (en) * | 2023-09-04 | 2023-10-10 | 广东鉴面智能科技有限公司 | Industrial field vision AI analysis and safety pre-warning system and method |
| CN116862244B (en) * | 2023-09-04 | 2024-03-22 | 广东鉴面智能科技有限公司 | Industrial field vision AI analysis and safety pre-warning system and method |
| CN117035378A (en) * | 2023-10-10 | 2023-11-10 | 广州海晟科技有限公司 | Intelligent building site management method and system based on Internet of things |
| CN117035378B (en) * | 2023-10-10 | 2024-02-02 | 广州海晟科技有限公司 | Intelligent building site management method and system based on Internet of things |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115660297A (en) | Automatic AI early warning system and method for construction site safety | |
| CN111552216A (en) | Construction site intelligent supervision system based on 5G | |
| CN110795989A (en) | Intelligent safety monitoring system for electric power operation and monitoring method thereof | |
| CN113298444A (en) | Private cloud-based intelligent power safety control platform | |
| CN111488799B (en) | Falling object identification method and system based on image identification | |
| CN111080260A (en) | Wisdom building site management platform | |
| CN109283897A (en) | A kind of wisdom power plant comprehensively monitoring center intelligent monitoring implementation method | |
| CN113706355A (en) | Method for building intelligent emergency system of chemical industry park | |
| CN211877114U (en) | Construction site supervisory system based on 5G | |
| CN110867046A (en) | Intelligent car washer video monitoring and early warning system based on cloud computing | |
| CN112112629A (en) | Safety business management system and method in drilling operation process | |
| CN112257494A (en) | Behavior recognition method based on intelligent video analysis technology and application | |
| CN111257507A (en) | Gas concentration detection and accident early warning system based on unmanned aerial vehicle | |
| CN115410354B (en) | Safety early warning method and device and safety early warning system for industrial factory | |
| CN113538825A (en) | Campus wall-turning event alarm method and system | |
| CN110543866A (en) | Safety management system and method for capital construction engineering constructors | |
| CN211184122U (en) | Intelligent video analysis system for linkage of railway operation safety prevention and control and large passenger flow early warning | |
| CN115567690A (en) | Intelligent monitoring system capable of automatically identifying dangerous points of field operation | |
| CN112261384A (en) | Artificial intelligence violation behavior recognition system | |
| CN115496640A (en) | Intelligent safety system of thermal power plant | |
| CN113392706A (en) | Device and method for detecting smoking and using mobile phone behaviors | |
| CN113506416A (en) | Engineering abnormity early warning method and system based on intelligent visual analysis | |
| CN112532927A (en) | Intelligent safety management and control system for construction site | |
| CN110533889B (en) | Sensitive area electronic equipment monitoring and positioning device and method | |
| CN111723725A (en) | Multi-dimensional analysis system based on video AI |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |