Construction site safety monitoring method and system based on large model and RAG

Abstract

The invention discloses a construction site safety monitoring method and system based on a large model and RAG, and the method comprises the steps: carrying out the intelligent analysis of real-time video data of a construction region based on a visual large model, and recognizing the behavior safety, violation operation, dangerous facilities and environment risks of construction personnel; adding a time label to the analyzed security event text to form a construction site security event log; reasoning and analyzing the security event log, determining whether security early warning is generated or not, and pushing early warning information; carrying out on-site intelligent voice broadcasting according to the early warning information; through a multi-RAG architecture, a construction site safety log database, a public construction safety management database and an enterprise internal safety management database are constructed, and construction safety information is stored by using a vector database and a graph database. By adopting the technical scheme of the invention, the problems of insufficient real-time monitoring capability, limited intelligent analysis capability and lack of effective knowledge base support in the existing construction site safety supervision are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction site intelligent monitoring, and in particular relates to a construction site safety monitoring method and system based on a large model and RAG. Background Art

[0002] In recent years, with the advancement of artificial intelligence, big data, and computer vision technologies, intelligent monitoring technology for construction sites has rapidly developed. Traditional video surveillance systems primarily collect and transmit on-site video data for remote viewing and oversight by managers. However, this approach still relies on manual judgment, limited by managers' experience and energy, making it difficult to achieve real-time, efficient, and intelligent safety supervision. Particularly in complex construction environments, traditional monitoring systems struggle to effectively identify dangerous behaviors, illegal operations, and safety hazards, leading to blind spots in supervision and increasing construction safety risks.

[0003] Construction site safety management involves numerous national regulations, industry standards, and internal company rules and regulations. Managers must monitor construction activities in accordance with these regulations to ensure that the construction process meets safety requirements. However, the complex and ever-changing construction site environment makes manual monitoring difficult to ensure comprehensive, accurate, and consistent results. Furthermore, the construction process generates a large amount of safety incident information that needs to be recorded, classified, and analyzed for subsequent review and optimization of management measures. Traditional manual management methods are not only time-consuming and labor-intensive, but also hinder the full utilization of the company's information management tools to improve the efficiency and quality of construction safety management.

[0004] While existing intelligent monitoring technologies can leverage computer vision to identify the status of construction personnel and equipment, they are primarily limited to static analysis of image content, making them incapable of in-depth safety risk reasoning and dynamic early warning. Furthermore, current safety management information systems are often limited to data storage and simple queries, lacking knowledge-based intelligent analysis capabilities and unable to provide precise management recommendations for specific construction scenarios. Therefore, leveraging multi-task large models (MTLM) and multimodal retrieval-augmented generation (RAG) technologies to achieve intelligent perception, reasoning analysis, automatic early warning, and intelligent interaction of construction site safety has become a key technical challenge in the field of construction safety management. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a construction site safety monitoring method and system based on a large model and RAG, so as to solve the problems existing in the existing construction site safety supervision, such as insufficient real-time monitoring capability, limited intelligent analysis capability, and lack of effective knowledge base support.

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

[0007] A construction site safety monitoring method based on a large model and RAG includes the following steps:

[0008] Step S1: Deploy multiple cameras at the construction site to collect real-time video data of the construction area;

[0009] Step S2: Based on the visual big model, the real-time video data of the construction area is intelligently parsed to automatically identify the construction workers' behavioral safety, illegal operations, hazardous facilities and environmental risks, and generate structured data; at the same time, time tags are added to the parsed safety event text to form a construction site safety event log;

[0010] Step S3: Perform intelligent analysis on the security event log, extract key regulatory information, and assess the security risk level to obtain reasoning analysis results; at the same time, determine whether to generate a security warning based on the reasoning analysis results, and push the warning information;

[0011] Step S4: Perform on-site intelligent voice broadcast according to the warning information;

[0012] Step S5: construct a construction site safety log database, a public construction safety management database, and an enterprise internal safety management database, and use a vector database and a graph database to store construction safety information;

[0013] Step S6: Use natural language to query construction site safety incidents, regulatory recommendations, and accident handling plans. The RAG retrieval system combines with the knowledge base to achieve cross-database and cross-modal joint retrieval and generate intelligent responses.

[0014] Preferably, it also includes: automatically generating a construction safety log and storing it in a database in combination with safety event images or short videos.

[0015] Preferably, in step S5, a construction site safety log database, a public construction safety management database and an enterprise internal safety management database are constructed through a multi-RAG architecture.

[0016] Preferably, cameras are distributed in high-altitude work areas, equipment operation areas, and areas with dense personnel.

[0017] The present invention also provides a construction site safety monitoring system based on a large model and RAG, comprising:

[0018] A video acquisition device is used to collect real-time video data of the construction area by deploying multiple cameras at the construction site;

[0019] The visual large-scale model parsing system is used to intelligently analyze real-time video data from construction areas, automatically identifying construction workers' behavioral safety violations, illegal operations, hazardous facilities, and environmental risks, and generating structured data. It also adds time tags to the parsed safety event text to form a construction site safety event log.

[0020] The reasoning analysis module is used to intelligently analyze security event logs, extract key regulatory information, and assess security risk levels to obtain reasoning analysis results. At the same time, based on the reasoning analysis results, it determines whether to generate a security warning and pushes the warning information;

[0021] Voice broadcast module, used for on-site intelligent voice broadcast based on warning information;

[0022] RAG retrieval module, used to build a construction site safety log database, a public construction safety management database, and an enterprise internal safety management database, and uses vector databases and graph databases to store construction safety information;

[0023] The language interaction module is used to query construction site safety incidents, regulatory recommendations, and accident handling plans through natural language. The RAG retrieval system is combined with the knowledge base to achieve cross-database and cross-modal joint retrieval and generate intelligent responses.

[0024] Preferably, it also includes: a construction history log text generation and storage module, which is used to automatically generate a construction safety log and store it in a database in combination with safety event images or short videos.

[0025] Preferably, the cameras of the video acquisition device are distributed in high-altitude working areas, equipment operation areas, and areas with dense populations.

[0026] The present invention converts video data collected on-site into key single-frame images, and uses a large visual model to analyze the image content to identify the safety behavior of construction workers, dangerous facilities, and environmental risks. Subsequently, the analysis results are converted into text data and time tags are attached to form a safety event record at the construction site. Furthermore, the text data is deeply analyzed using an inference model to extract key safety supervision information, and the information is pushed to the management center's large screen or mobile APP to achieve real-time early warning. A voice model is introduced to convert the warning information into a voice broadcast for real-time intelligent broadcasting at the construction site or management center to ensure that safety alerts can be quickly conveyed to relevant personnel. In addition, the present invention adopts a multi-RAG architecture to build a construction site safety event database, a public safety management specification database, and an internal enterprise safety management database, and intelligently interacts with management personnel through a language model to provide safety event query, management, and summary functions. The innovation of this system lies in the combination of computer vision, large model reasoning, multimodal retrieval enhancement generation and other technologies to achieve full-process intelligent supervision of construction site safety, improve supervision efficiency, and reduce safety risks.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] 1. Empowering multi-task large models to improve intelligent safety supervision capabilities: This invention uses a large visual model to monitor the construction site and combines it with an inference model for in-depth analysis. It can not only identify the safety behavior of construction workers, but also conduct inference analysis of safety hazards, thereby improving intelligent monitoring capabilities.

[0029] 2. Multi-RAG architecture to achieve efficient and intelligent retrieval and safety management: A construction site safety incident database, a public construction safety management database, and an internal enterprise construction safety management database have been established. Multimodal retrieval enhancement generation technology is used to achieve rapid retrieval, association analysis, and safety management optimization.

[0030] 3. Intelligent early warning push + voice broadcast to improve response efficiency: Not only can early warning information be pushed to the control center and management personnel, but it can also be quickly transmitted to the construction site through voice broadcast, improving the accessibility of safety supervision and ensuring that front-line personnel can receive alarm information in a timely manner.

[0031] 4. Natural language interaction to optimize the safety management experience: Managers can use language models to query safety incidents at construction sites, obtain national regulations and corporate safety management recommendations, thereby improving the intelligence level of construction safety management.

[0032] In summary, the present invention constructs an intelligent construction site safety monitoring system through a multi-task large model + multi-RAG architecture, which can realize real-time monitoring, intelligent reasoning, real-time broadcasting, knowledge retrieval and management decision support for construction safety; effectively improves the efficiency of construction safety management, reduces safety risks, and provides technical support for the intelligent development of the construction industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0034] Figure 1 It is a flow chart of the construction site safety monitoring method based on the large model and RAG of the present invention;

[0035] Figure 2 It is a schematic diagram of the visual large model analysis;

[0036] Figure 3 It is a schematic diagram of multi-RAG retrieval enhancement generation;

[0037] Figure 4 This is a schematic diagram of the construction safety database architecture. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Example 1:

[0041] like Figure 1 As shown, an embodiment of the present invention provides a construction site safety monitoring method based on a large model and RAG, comprising the following steps:

[0042] Step S1: Deploy multiple cameras at the construction site to collect real-time video data of the construction area;

[0043] Step S2: Based on the visual big model, the real-time video data of the construction area is intelligently parsed to automatically identify the construction workers' behavioral safety, illegal operations, hazardous facilities and environmental risks, and generate structured data; at the same time, time tags are added to the parsed safety event text to form a construction site safety event log;

[0044] Step S3: Perform intelligent analysis on the security event log, extract key regulatory information, and assess the security risk level to obtain reasoning analysis results; at the same time, determine whether to generate a security warning based on the reasoning analysis results, and push the warning information;

[0045] Step S4: Perform on-site intelligent voice broadcast according to the warning information;

[0046] Step S5: Build a construction site safety log database, a public construction safety management database, and an enterprise internal safety management database through a multi-RAG architecture;

[0047] Step S6: Use natural language to query construction site safety incidents, regulatory recommendations, and accident handling plans. The RAG retrieval system combines with the knowledge base to achieve cross-database and cross-modal joint retrieval and generate intelligent responses.

[0048] As an implementation method of the embodiment of the present invention, Figure 2 As shown, step S2 includes:

[0049] (1) Use large visual models (Qwen-VL, miniCPM, SAM, YOLO, etc.) to detect the helmet wearing, illegal behavior, equipment status, etc. of construction site personnel;

[0050] (2) Combined with OCR recognition technology, it automatically reads the content of construction safety signs and notice boards and determines whether they comply with safety management regulations;

[0051] (3) Generate structured text data, including event type, time of occurrence, location, people involved, violation description, equipment status, risk level (low, medium, high), etc.

[0052] As an implementation of the embodiment of the present invention, step S3 includes:

[0053] (1) Using reasoning analysis models (such as GPT-o1, Claude, Gemini 2.0 Flash Thinking, DeepSeek-r1, etc.), combined with construction safety rules, to perform logical reasoning on event texts and automatically determine the severity of violations;

[0054] (2) Combined with the construction site historical data vector database, predict future risks and provide rectification suggestions;

[0055] (3) Compare historical similar events and automatically recommend the best accident handling plan, including handling methods from history or standard procedures.

[0056] As an implementation of an embodiment of the present invention, step S4 includes:

[0057] (1) Adopting a hierarchical early warning mechanism:

[0058] Low risk: Push reminder notifications to managers;

[0059] Medium risk: Send an APP warning notification + on-site voice broadcast;

[0060] High risk: triggering emergency alarms and construction suspension notices;

[0061] (2) Intelligent voice broadcast supports multiple languages and dialects to ensure that all construction workers can understand the alarm content;

[0062] (3) Combined with smart wearable devices (such as smart phones, tablets, helmets, and walkie-talkies), early warning signals are sent to construction workers in real time.

[0063] As an implementation method of the embodiment of the present invention, Figure 3 As shown, step S5 includes:

[0064] (1) Use vector databases (such as FAISS and Milvus) to store construction event logs to support semantic-level retrieval;

[0065] (2) Use a graph database (such as Neo4j) to store the relationships between construction safety events and support knowledge graph analysis and multi-hop queries;

[0066] (3) Combined with the multi-RAG architecture, cross-database retrieval of construction safety incidents, regulatory information, and enterprise management specifications is performed, and intelligent safety management recommendations are provided.

[0067] As an implementation method of the embodiment of the present invention, Figure 4 As shown, step S6 includes:

[0068] (1) Construction safety query. Users can enter "Are there any cases of people not wearing safety helmets at the construction site?" The RAG system searches the database and returns the violation records for the past 7 days, along with screenshots or video evidence;

[0069] (2) Regulation matching. Combined with the construction regulation database, it provides compliance analysis, such as: "This behavior violates Article X of the Construction Safety Code."

[0070] (3) Safety incident summary: Automatically generate a construction safety report and export it to PDF / Excel format for management review.

[0071] As an implementation method of an embodiment of the present invention, it further includes: automatically generating a construction safety log and storing it in a database in combination with safety event images or short videos, specifically including:

[0072] (1) Automatic log generation. Based on the visual big model + language model, the construction log is automatically generated with risk level, violation description, and recommended corrective measures;

[0073] (2) Image / video storage. A screenshot or short video (e.g., 5-10 seconds) will be saved for each security incident and stored together with the log text for subsequent retrospective analysis.

[0074] (3) Intelligent indexing. Using metadata indexing technology, indexes are established based on multiple dimensions such as time, personnel, and event type to improve data query efficiency;

[0075] (4) RAG data conversion: Natural language processing (NLP) technology is used to convert construction history logs into vectorized data, supporting semantic-level similarity and relevance ranking retrieval.

[0076] The present invention combines technologies such as multi-task large model (MTLM), multi-modal retrieval enhanced generation (Multi-RAG), voice broadcast, and construction log storage to achieve intelligent safety supervision, real-time early warning, historical event storage, and intelligent retrieval of regulations at construction sites, significantly improving the intelligence level of construction safety management.

[0077] The present invention deploys multiple video surveillance terminals at construction sites to collect real-time video data from the construction site. The video data is converted into keyframe images and fed into a large visual model for intelligent analysis to identify the safety behaviors of construction workers, the status of construction equipment, hazardous facilities, and other safety hazards. The key information identified by the large visual model is converted into text and time-stamped to construct a complete record of construction site safety events. Furthermore, the present invention utilizes a large inference model to conduct in-depth analysis and logical reasoning on the monitored text data to extract key safety supervision information from the construction site and determine whether there are safety hazards. When safety risks or illegal operations are detected, real-time warning information is pushed to management personnel via terminals such as the control center's large screen or mobile apps. Simultaneously, a voice model is introduced to convert the warning information into voice broadcasts for real-time intelligent broadcasting at the construction site or in the management center, ensuring that safety alerts are quickly conveyed to relevant personnel and improving the response speed of construction safety management. In terms of data storage and management, the present invention adopts a multi-RAG (multimodal retrieval augmented generation) architecture to establish a construction site safety event database, a public construction safety management specification database, and an internal enterprise safety management database. These databases store construction site safety incident records, national and industry-related construction safety management regulations, and internal company safety management systems. Managers can interact with the system using natural language via a large language model to query construction site safety incidents, obtain safety management solutions, and summarize incidents and optimize safety management based on historical construction site data and knowledge base information. Finally, the large model integrates the search results to provide managers with comprehensive analysis reports to support construction safety management decisions.

[0078] Example 2:

[0079] An embodiment of the present invention also provides a construction site safety monitoring system based on a large model and RAG, including: a video acquisition device, a data transmission device, a visual large model parsing system, an inference analysis module, a voice broadcast module, a RAG retrieval module and a language interaction module. Each module operates around key factors such as the behavior of construction workers in the construction site, safety facilities and dangerous areas of the construction site.

[0080] As one implementation of the present invention, a video capture device collects video data from the construction site using multiple network cameras. These cameras are distributed across the construction area to ensure that both the movement of construction personnel and the operating area of equipment are within the monitoring range. Camera Type: Third-generation intelligent network cameras are used. These cameras feature panoramic monitoring, night vision, and high frame rates to ensure real-time monitoring of the construction environment around the clock.

[0081] As an implementation method of an embodiment of the present invention, the data transmission equipment uses POE power supply and fiber transmission to ensure that high-definition video data can be transmitted safely and quickly to the server. The monitoring data is initially filtered and compressed through switches, routers, and edge computing devices to reduce data transmission delays.

[0082] As an implementation method of an embodiment of the present invention, the visual large model parsing system is used to analyze the collected construction site video data, and automatically identify the safety behaviors of construction personnel, dangerous facilities, and potential safety hazards. Main functions: 1. Construction personnel behavior identification: Detect whether there are behaviors such as not wearing a safety helmet, illegal climbing, and not wearing a safety rope. 2. Equipment operation status analysis: Identify the status of tower crane operation, excavator operation, construction vehicle driving, etc., and determine whether there are any illegal operations. 3. Environmental risk assessment: Detect environmental factors such as obstacles, exposed wires, and the stacking of dangerous goods in the construction area.

[0083] Furthermore, in visual model selection, we utilize commonly used large-scale models in the industry, such as Vision Transformer, Wenxin Yige, Qianwen Tuwen Large Model, SenseTime Ririxin SenseNova, and Zhipu CogView / CogVLM. These models are combined with computer vision algorithms such as YOLOv8, SAM (SegmentAnything Model), and CLIP to achieve high-precision recognition. By combining large models with edge computing, we improve the real-time performance of construction safety identification and avoid latency issues associated with cloud-based processing.

[0084] Furthermore, the use of visual models: Using large visual models such as Qwen-VL, or large visual models based on the Transformer architecture, to support target detection, OCR recognition, and scene semantic understanding in complex scenarios, to achieve safety monitoring of construction sites. Lightweight multimodal models, including but not limited to the miniCPM large visual model, can run on edge computing devices to achieve real-time analysis of construction worker behavior and equipment status. SAM (SegmentAnything Model), used for object segmentation at the construction site, accurately locates construction workers, equipment, and dangerous facilities. YOLO (You Only Look Once), used for real-time target detection models, is suitable for efficient monitoring of construction sites.

[0085] The processing flow is as follows: 1. Video data is captured by cameras and converted into single-frame images. 2. The visual model analyzes the status of construction personnel and equipment and converts it into structured text. 3. A security event log is generated, recording the time, personnel, behavior category, and detection results.

[0086] As an implementation method of an embodiment of the present invention, the safety reasoning analysis module performs deep reasoning analysis based on the recognition results of the visual large model, determines whether the construction behavior is compliant, and generates a safety warning.

[0087] Furthermore, the core analysis logic includes: If a construction worker is not wearing a helmet, the inference model determines whether the work area is dangerous. It also determines whether the construction equipment is overloaded or outside the safe operating range. By combining this with the construction specification database, it determines whether the construction operation complies with safety standards.

[0088] Furthermore, risk-based warnings are: 1. Low risk: Minor violations (e.g., temporarily removing a helmet) – pop-up notification. 2. Medium risk: Serious violations (e.g., not following the designated route) – push notifications via the app / screen. 3. High risk: Major hazards (e.g., crane operation errors, personnel falls) – on-site voice alarms + emergency notifications.

[0089] As an implementation method of an embodiment of the present invention, the voice warning module converts safety warning information into voice through a large voice model (such as VALL-E / VALL-E, Bark, iFLYTEKZhisheng, Paraformer series, WeNet and other voice models) and intelligently broadcasts it at the construction site.

[0090] Furthermore, the system features: multi-language support (Mandarin, dialects, Cantonese, etc.) to meet the needs of different construction workers. It also features scenario-adaptive voice broadcast volume and frequency adjustment based on different risk levels. It also integrates with the management system to simultaneously push notifications to the construction control center and the smart devices worn by construction workers.

[0091] Further, for example: If a low-risk event occurs, the broadcast will be: "Please note, construction is underway in the area ahead, please wear a safety helmet." If a high-risk event occurs, the broadcast will be: Emergency alert: "Danger! Crane operating area, please evacuate immediately!"

[0092] As one implementation of this invention, the multi-RAG retrieval system stores knowledge data related to construction safety and supports intelligent retrieval. This technology combines information retrieval and text generation. When answering user queries, the system not only generates answers but also references relevant information from authoritative knowledge bases, providing accurate and reliable safety management recommendations.

[0093] Furthermore, the retrieval function allows managers to enter a query (e.g., "What should I do if someone is not wearing a hard hat on a construction site?"). RAG first searches the construction safety management database for relevant regulations, cases, and standards. It uses a vector database or graph database to store construction safety knowledge, supporting efficient, semantic-level retrieval.

[0094] Furthermore, the augmentation function: the retrieved information such as regulations and management experience will be input into large models (such as Qwen-VL, ChatGPT, DeepSeek) for understanding and integration.

[0095] As an implementation method of an embodiment of the present invention, a language interaction module (such as ChatGPT, Claude) generates clear answers that meet the needs of the construction site based on the retrieved knowledge. Safety rectification plans can also be automatically generated.

[0096] This invention uses a multi-RAG architecture to establish three major safety management databases: 1. A construction site safety incident database, used to store violation records and accident analysis; 2. A public construction safety management database, used to store national regulations and industry standards; 3. An internal enterprise safety management database, used to store enterprise safety management systems.

[0097] Workflow: 1. Construction personnel or managers can conduct safety inquiries through the language interaction module. 2. The multi-RAG system searches historical events, regulations, and corporate management methods. 3. Language models (such as ChatGPT, Qianwen, Claude, and DeepSeek) integrate information and return results in a conversational format.

[0098] The language interaction module allows construction managers to use natural language to inquire about the safety conditions of the construction site.

[0099] Query examples: You can ask, "Were there any safety hazards at the construction site yesterday?", "Have any dangerous behaviors occurred at the construction site in the past month?", "What should I do if I find a worker not wearing a safety helmet?", "What are the company's safety regulations for working at heights?"

[0100] Return content: Construction event log (yesterday's violation records); industry standards (interpretation of relevant regulations); enterprise safety regulations (company internal management measures).

[0101] As an implementation method of an embodiment of the present invention, it also includes: a construction history log text generation and storage module, which is used to automatically generate a construction safety log and store it in a database in combination with safety event images or short videos.

[0102] The construction history log module supports visual analysis of construction safety events, including:

[0103] (1) Timeline analysis: Displays construction history events in a timeline format, enabling quick retrieval of safety issues within a certain period of time.

[0104] (2) Heat map of violations. Use heat maps to display high-frequency areas of violations, helping managers optimize safety protection measures;

[0105] (3) Accident correlation analysis: Combined with knowledge graph technology, the correlation between construction accidents is analyzed to provide decision support for safety management.

[0106] The present invention constructs an intelligent construction site safety supervision system based on multi-task large model + multi-RAG + intelligent voice broadcast.

[0107] System Advantages: Comprehensive monitoring of construction sites, efficient identification of violations. Intelligent reasoning and risk warnings enhance safety supervision capabilities. Voice broadcast and intelligent interaction optimize the user experience. Multi-RAG knowledge base supports regulatory query and management optimization.

[0108] The present invention can be widely used in construction sites of buildings, bridges, tunnels and other projects to improve the level of construction safety management and reduce accident risks.

[0109] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A construction site safety monitoring method based on a large model and RAG, characterized in that: The following steps are involved: Step S1: Deploy multiple cameras at the construction site to collect real-time video data of the construction area; Step S2: Based on the visual big model, the real-time video data of the construction area is intelligently parsed to automatically identify the construction workers' behavioral safety, illegal operations, hazardous facilities and environmental risks, and generate structured data; at the same time, time tags are added to the parsed safety event text to form a construction site safety event log; Step S3: Perform intelligent analysis on the security event log, extract key regulatory information, and assess the security risk level to obtain reasoning analysis results; at the same time, determine whether to generate a security warning based on the reasoning analysis results, and push the warning information; Step S4: Perform on-site intelligent voice broadcast according to the warning information; Step S5: construct a construction site safety log database, a public construction safety management database, and an enterprise internal safety management database, and use a vector database and a graph database to store construction safety information; Step S6: Use natural language to query construction site safety incidents, regulatory recommendations, and accident handling plans. The RAG retrieval system combines with the knowledge base to achieve cross-database and cross-modal joint retrieval and generate intelligent responses.

2. The construction site safety monitoring method based on a large model and RAG according to claim 1, characterized in that: Also includes: Automatically generate construction safety logs and store them in the database along with safety event images or short videos.

3. The construction site safety monitoring method based on a large model and RAG according to claim 2, characterized in that: In step S5, a construction site safety log database, a public construction safety management database, and an enterprise internal safety management database are constructed through a multi-RAG architecture.

4. The construction site safety monitoring method based on a large model and RAG according to claim 3 is characterized in that: Cameras are distributed in high-altitude work areas, equipment operation areas, and areas with dense crowds of people.

5. A construction site safety monitoring system based on a large model and RAG, characterized in that: include: A video acquisition device is used to collect real-time video data of the construction area by deploying multiple cameras at the construction site; The visual large-scale model parsing system is used to intelligently analyze real-time video data from construction areas, automatically identifying construction workers' behavioral safety violations, illegal operations, hazardous facilities, and environmental risks, and generating structured data. It also adds time tags to the parsed safety event text to form a construction site safety event log. The reasoning analysis module is used to intelligently analyze security event logs, extract key regulatory information, and assess security risk levels to obtain reasoning analysis results. At the same time, based on the reasoning analysis results, it determines whether to generate a security warning and pushes the warning information; Voice broadcast module, used for on-site intelligent voice broadcast based on warning information; The RAG retrieval module is used to build a construction site safety log database, a public construction safety management database, and an internal enterprise safety management database, and uses a vector database and a graph database to store construction safety information; The language interaction module is used to query construction site safety incidents, regulatory recommendations, and accident handling plans through natural language. The RAG retrieval system is combined with the knowledge base to achieve cross-database and cross-modal joint retrieval and generate intelligent responses.

6. The construction site safety monitoring system based on large model and RAG as claimed in claim 5, characterized in that: Also includes: The construction history log text generation and storage module is used to automatically generate construction safety logs and store them in the database in combination with safety event images or short videos.

7. The construction site safety monitoring system based on a large model and RAG according to claim 6, characterized in that: The cameras of the video acquisition device are distributed in high-altitude working areas, equipment operation areas, and areas with dense crowds of people.

Images