Road maintenance site AI video perception vehicle intrusion prevention and hidden danger self-inspection early warning system

Abstract

The invention discloses a road maintenance site AI video sensing vehicle intrusion prevention and hidden danger self-inspection early warning system, and belongs to the technical field of road maintenance safety prevention and control. The system adopts a layered framework of a sensing layer, a transmission layer, a platform layer and an application layer, double AI cameras are used for realizing accurate identification of vehicle intrusion, hidden dangers and'three violation 'behaviors, and a'identification-early warning-supervision-rectification-tracing' management closed loop is constructed by matching an acousto-optic early warning device and a double transmission mode. The system can be quickly deployed in various operation vehicles or fixed structures, adapts to complex outdoor environments, has the advantages of high recognition precision, quick response and high compatibility, is suitable for maintenance scenes such as highways, national and provincial trunk lines, urban roads and the like, and can greatly improve the safety management level of maintenance operation.

Description

Technical Field

[0001] This invention relates to the field of highway maintenance safety prevention and intelligent management technology, specifically to a highway maintenance site vehicle intrusion prevention and hidden danger self-inspection and early warning system that integrates AI video real-time perception, multi-dimensional early warning, and remote collaborative management. It is applicable to various maintenance operation scenarios such as expressways, national and provincial trunk roads, and urban roads, and can realize full-process intelligent supervision of risk prevention and control of vehicle intrusion outside the work area and internal hidden dangers and "three violations" behaviors. Background Technology

[0002] As a core infrastructure of the transportation system, highway maintenance is crucial for ensuring road capacity, extending road lifespan, and reducing traffic accident rates. However, maintenance sites are typically located in open traffic environments, with the work area adjacent to the traffic lanes, posing extremely high safety risks. Currently, highway maintenance sites face two major core safety issues:

[0003] First, there is the risk of private vehicles entering the maintenance work area. In the past, private vehicles have frequently entered the maintenance work area due to excessive speed, improper operation, or failure to warn in time, which directly threatens the lives of the workers, causes damage to maintenance equipment, and delays in the work progress.

[0004] Secondly, there is insufficient prevention and control of hidden dangers and "three violations" (illegal command, illegal operation, and violation of labor discipline) in the work area. Problems such as non-standard equipment placement, missing safety warning signs, workers not wearing protective equipment as required, and illegal operation of equipment can easily lead to secondary safety accidents.

[0005] Traditional on-site safety management in highway maintenance relies primarily on manual supervision, which suffers from inherent drawbacks such as limited monitoring scope, poor real-time performance, high labor costs, and susceptibility to human error. Existing early warning devices are mostly single-function designs, such as standalone warning lights or ordinary cameras, lacking AI-powered intelligent recognition capabilities and unable to proactively assess vehicle intrusion risks and potential hazards in the work area. Some similar systems suffer from poor mobility, low recognition accuracy, delayed early warning response, and insufficient hardware and software compatibility. Furthermore, they fail to establish a closed-loop management system encompassing "identification-early warning-supervision-rectification," making it difficult to meet the dynamic safety management needs of modern highway maintenance operations.

[0006] With the rapid development of technologies such as artificial intelligence, the Internet of Things, and cloud computing, intelligent monitoring has become a trend in highway maintenance safety management. While some existing technical solutions attempt to use cameras combined with simple algorithms for monitoring, they generally suffer from the following shortcomings:

[0007] First, it has poor mobility, with most equipment being fixedly installed, making it difficult to adapt to maintenance operation scenarios of different locations and scales;

[0008] Second, the recognition accuracy is insufficient. The algorithm model lacks special training for the complex environment of the maintenance site, resulting in a high rate of misjudgment and missed judgment for vehicle intrusion risks, hidden dangers, and violations of safety regulations.

[0009] Third, the early warning methods are limited, relying mainly on light or sound warnings, which have limited effectiveness and fail to achieve coordination between on-site early warning and remote reminders;

[0010] Fourth, the system lacks compatibility and scalability, has poor hardware and software synergy, and is difficult to upgrade functions and adapt to scenarios according to actual needs.

[0011] Fifth, it lacks a complete management loop, only achieving the "identification-early warning" stage, without involving subsequent management processes such as hazard rectification, behavior correction, and data statistical analysis.

[0012] Therefore, developing an AI video perception and early warning system with high mobility, high-precision recognition, multi-dimensional early warning, and full-process management capabilities to address the shortcomings of existing technologies and improve the safety control level and management efficiency of highway maintenance sites has significant practical significance and application value. Summary of the Invention

[0013] This invention aims to overcome the shortcomings of existing technologies and provide an AI video perception system for preventing vehicle intrusion and for self-inspection and early warning of potential hazards at highway maintenance sites, specifically addressing the following technical problems:

[0014] 1. To achieve accurate identification and real-time early warning of social vehicles entering the work area, solving the problems of delayed early warning and low identification accuracy of traditional prevention and control methods, and reducing the risk of safety accidents caused by vehicles entering the work area;

[0015] 2. To achieve intelligent, all-time identification and timely reminders of environmental hazards and personnel's "three violations" within the work area, solving the problems of low efficiency and easy omissions in manual supervision, and promoting self-inspection and self-correction of hazards and real-time control of "three violations";

[0016] 3. Improve the system's mobility and environmental adaptability, enabling it to be quickly deployed in different types and locations of maintenance operations, and adapt to complex outdoor climate and lighting conditions;

[0017] 4. Construct a complete management closed loop of "identification-early warning-supervision-rectification-traceability" to realize functions such as remote real-time monitoring, historical data traceability, and statistical analysis, thereby improving the level of precision in maintenance safety management;

[0018] 5. Ensure the stability and compatibility of the system's software and hardware working together, possess good scalability, and support subsequent algorithm optimization, function upgrades, and hardware expansion.

[0019] To achieve the above objectives, the present invention adopts the following technical solution: an AI video perception system for preventing vehicle intrusion and a self-inspection and early warning system for hidden dangers at highway maintenance sites. This system employs a layered architecture design, including a perception layer, a transmission layer, a platform layer, and an application layer. Each layer works collaboratively to achieve functions such as vehicle intrusion identification and early warning, self-inspection and early warning of hidden dangers and violations of traffic regulations, and remote monitoring and management. The specific technical solution is as follows:

[0020] (a) Perception layer

[0021] The perception layer, serving as the system's data acquisition and on-site early warning terminal, integrates functions such as data acquisition, intelligent recognition, and audible and visual early warning. It includes a first AI camera, a second AI camera, a controllable flashing alarm light, a loudspeaker, a liftable pole, and a fixed mounting bracket. The specific structure and parameters are as follows:

[0022] 1. First AI Camera: Specifically designed for identifying unauthorized vehicles, this camera is deployed in the upstream transition zone of the maintenance work area to monitor the real-time operation status of passing vehicles. It employs a high-definition image sensor with a resolution of at least 1080P and a frame rate of at least 30fps, enabling high-speed dynamic object capture. It accurately acquires vehicle outlines, colors, license plates, and other feature information, as well as their trajectory and real-time speed data. A built-in high-performance AI chip supports local operation of complex vehicle trajectory and speed analysis algorithms, providing rapid data processing and risk assessment capabilities. The casing is waterproof, dustproof, and anti-interference designed with a protection rating of at least IP67. Its operating temperature range is -30℃ to +60℃, making it adaptable to complex outdoor environments such as high temperatures, low temperatures, rain, snow, and sandstorms.

[0023] 2. Second AI Camera: Dedicated to identifying potential hazards and violations of safety regulations in the work area, this camera is deployed on the work area side to cover the environment and personnel dynamics within the work area. It has a resolution of at least 1080P, a wide-angle lens, and a horizontal field of view of at least 120°, enabling wide coverage of the work area. It has a built-in infrared night vision module that automatically switches to night vision mode in low-light conditions (illuminance below 0.1 lux) to ensure clear image acquisition. It integrates algorithms for identifying potential hazards in the work area and for identifying violations of safety regulations, providing local real-time analysis and feature matching capabilities. It also features IP67 waterproof and dustproof ratings and a wide operating temperature range of -30℃ to +60℃, adapting to the complex environment of the maintenance site.

[0024] 3. Controllable Strobe Alarm Light: As an on-site visual warning device, it has a long, flat shell structure. The main body of the shell is made of dark gray hard engineering plastic, which combines lightweight and impact resistance. Red LED light-emitting areas are set at both ends along the length, and a blue LED light-emitting area is set in the middle. Each light-emitting area uses a concave rectangular light-transmitting mask made of high-transmittance PC material to enhance light penetration. Each light-emitting area has multiple high-brightness LED beads built in, with a brightness of no less than 8000mcd per bead. It supports switching between multiple light-emitting modes such as constant light, strobe, and burst flash. The strobe frequency can be adjusted within the range of 5-20Hz to ensure the warning effect at different distances and in different environments. The bottom of the warning light is fixed to the mounting base by two symmetrically distributed small connecting brackets. The mounting base is a cylindrical metal part with a rust-proof surface and is coaxially connected to the top of the liftable rod. The connection method is bolt fastening to ensure structural stability.

[0025] 4. Loudspeakers: Two loudspeakers are configured, corresponding to the first and second AI cameras respectively, to provide directional warnings. The loudspeakers adopt a frustum-shaped structure, with the mouth diameter larger than the rear diameter (mouth diameter 15-20cm, rear diameter 8-12cm). They are integrally injection molded with the rear of the camera housing, resulting in a compact structure. The mouth faces horizontally outward to ensure directional propagation of warning messages, minimizing interference within the work area. The loudspeakers support multiple sound alarm modes, with sound intensity adjustable from 60-120dB. They have pre-stored multiple sets of warning messages, including "You are speeding, slow down," "Construction ahead, please change lanes immediately," "Caution: Danger in work area," "You have violated regulations, please rectify immediately," "Vehicle has entered from the side, please evacuate immediately," and "Do not work without safety protective equipment," which can automatically switch playback according to different warning scenarios. The loudspeakers have a built-in waterproof and dustproof mesh, with a protection level of no less than IP65, suitable for outdoor environments.

[0026] 5. Liftable Rod: Utilizing a hydraulic lifting column structure, this is a retractable, multi-section cylindrical metal rod made of high-strength aluminum alloy, combining lightweight design with load-bearing capacity. The outer surface features a fine, vertical striped texture, enhancing both structural strength and aesthetics. The column comprises a fixed lower section and a retractable upper section, with the lower section having a larger diameter than the upper section (8-10cm for the lower section and 5-7cm for the upper section). The retractable section can smoothly rise and fall axially, with a maximum lifting height of 4m and a lifting speed of 0.1-0.2m / s. The height can be adjusted remotely or via local buttons to meet the monitoring coverage needs of different work scenarios. The lower section of the column has two annular metal reinforcing hoops on its outer side. These hoops are made of stainless steel, have a smooth surface, and are coaxially fitted with the column, secured with bolts to enhance the column's wind and impact resistance. The bottom of the column connects to a cylindrical hydraulic drive chamber, a gray, sealed shell containing a hydraulic pump, solenoid valves, controllers, and other lifting control components. The bottom of the drive chamber has anti-slip and shock-absorbing pads to reduce vibration during equipment operation.

[0027] 6. Fixed Mounting Bracket: Used to fix the entire system to the work vehicle or on-site fixed structure, including two sets of symmetrically distributed metal buckle structures. Each set of buckles consists of an arc-shaped clamping plate and a locking bolt. The arc-shaped clamping plate is made of stainless steel with an anti-slip rubber pad on the inside, which fits and conforms to the outer cylindrical surface of the lower section of the hydraulic lifting column, enhancing clamping friction and preventing loosening. The fixing end of the buckle is connected to the metal guardrail of the pickup truck cargo box, dump truck guardrail, or on-site fixed bracket by bolts. The bolts pass through the reserved holes in the guardrail or the mounting holes of the bracket to achieve locking, adapting to different specifications of fixed carriers. The two sets of buckles are distributed along the axial direction of the column, with a spacing of 30-50cm, respectively fixing the "middle part of the lower section of the column" and the "upper part of the drive compartment", ensuring that the column is in a vertical position and preventing tilting or shaking. The opening angle of the arc-shaped clamping plate of the mounting bracket can be finely adjusted by the locking bolt, adapting to columnar carriers with a diameter of 5-12cm, realizing rapid fixing in multiple positions and scenarios.

[0028] 7. Sound and light cone (auxiliary early warning device): It adopts high-brightness LED beads, with high light intensity and obvious warning effect; it supports multiple sound alarm modes and the sound intensity is adjustable, which can clearly convey warning information in different environments; it has wireless receiving function, which can receive the early warning signal transmitted by the camera in a timely manner and quickly issue sound and light warnings; at the same time, it is waterproof and drop-resistant, adaptable to outdoor working environment, and can be flexibly placed around the work area to enhance the early warning coverage.

[0029] (ii) Transport Layer

[0030] The transport layer is responsible for data transmission between the perception layer and the platform layer, ensuring the real-time performance, stability, and security of the data. The specific configuration is as follows:

[0031] 1. Transmission method: It adopts a dual transmission mode combining 4G / 5G wireless network and wired network. The 4G / 5G wireless network is given priority to achieve flexible deployment. In the operation scenario where conditions permit, it can be switched to wired network to improve transmission stability.

[0032] 2. Transmission Equipment: Equipped with a high-performance 4G / 5G wireless router and a gigabit wired network switch. The wireless router supports multi-band (2.4GHz / 5GHz) communication, has strong signal coverage and anti-interference performance, a maximum transmission rate of no less than 1Gbps, and supports dual-SIM dual-standby function to ensure uninterrupted network signal. The wired network switch has 4-8 gigabit Ethernet ports to meet the needs of multiple devices accessing simultaneously and supports plug and play.

[0033] 3. Data Security: The data is encrypted using SSL / TLS encryption protocol during transmission to prevent theft or tampering; it has a data retransmission mechanism that automatically retransmits critical data from the interruption period when the network connection is restored after an interruption, ensuring data integrity.

[0034] 4. Equipment compatibility: The transmission equipment is waterproof and dustproof, with a protection level of not less than IP65. It can be integrated with the sensing layer equipment and is suitable for outdoor operating environments.

[0035] (III) Platform Layer

[0036] The platform layer serves as the core control and data processing center of the system. It utilizes cloud computing technology to build a distributed cloud platform, possessing functions such as data storage, data processing, algorithm model management, and system management, as detailed below:

[0037] 1. Data Storage Module: Adopting a distributed storage architecture, it integrates hard disk arrays and cloud storage resources to centrally store images, videos, vehicle operation data (trajectory, speed, license plate information), hazard identification data (location, type, characteristics, discovery time), and "three violations" behavior identification data (personnel information, behavior type, occurrence time, image evidence) collected by the perception layer. It supports the classified storage of structured and unstructured data. Video data adopts the H.265 encoding format to reduce storage space. The data retention time can be set according to needs (1-90 days), and it supports quick query and retrieval of historical data.

[0038] 2. Data Processing Module: Configured with high-performance computing nodes, it performs real-time analysis and processing of data uploaded from the transmission layer, including vehicle feature extraction, trajectory fitting, speed calculation, hazard feature matching, and personnel behavior and posture analysis; it adopts a collaborative processing mode of edge computing and cloud computing, with the perception layer completing preliminary identification and risk assessment locally, and the platform layer performing in-depth analysis and data aggregation, thereby improving processing efficiency and reducing network bandwidth consumption;

[0039] 3. Algorithm Model Management Module: Responsible for uploading, updating, optimizing, and testing AI algorithm models, including vehicle trajectory and speed analysis algorithms, work area hazard identification algorithms, and personnel violation identification algorithms.

[0040] Vehicle trajectory and speed analysis algorithm: By processing the vehicle image sequence captured by the first AI camera, the YOLO target detection algorithm is used to extract vehicle features, and the Kalman filter target tracking algorithm is combined to achieve continuous tracking of the vehicle and fit the vehicle trajectory; based on the monocular vision ranging principle, combined with the preset geographical location information and safety zone boundary of the work area, the distance between the vehicle and the work area and the real-time speed are calculated; a vehicle intrusion risk assessment model is established, and speed thresholds (preset according to road speed limit standards and work area safety requirements, which can be manually adjusted) and distance thresholds are set. When the vehicle speed exceeds the threshold and the distance from the work area is less than the threshold, it is judged as a risky intrusion vehicle and the warning mechanism is triggered;

[0041] Hazard identification algorithm for work area: Establish a feature library containing common hazards such as non-standard equipment placement, missing or damaged safety warning signs, messy temporary power lines, potholes on the road without enclosure, and disorderly stacking of work materials. Use CNN convolutional neural network algorithm to scan and match the environmental images of the work area captured by the second AI camera in real time. When the matching degree exceeds the preset threshold (85%), it is automatically identified as a hazard in the work area and the relevant information of the hazard is recorded.

[0042] The algorithm for identifying personnel violations ("three violations") establishes a feature database containing violations such as not wearing a safety helmet, not wearing a safety belt, not wearing a reflective vest, operating maintenance equipment improperly, crossing safety barriers without authorization, leaving one's post, and staying in dangerous areas. It uses a human pose estimation algorithm (OpenPose) to extract key skeletal points of workers, and combines this with a behavior analysis algorithm to determine whether a person's behavior conforms to the "three violations" characteristics. When the matching degree exceeds a preset threshold (80%), it is identified as a "three violations," triggering an alert mechanism. The algorithm model management module supports online upgrades and can continuously optimize algorithm parameters based on feedback data from actual application scenarios to improve recognition accuracy.

[0043] 4. System Management Module: Includes user management, device management, data management, and access control functions.

[0044] User management: Supports user registration, login, password modification, account cancellation and other operations, and allows the creation of accounts with different roles such as administrator, job supervisor, and regular operator;

[0045] Equipment Management: Real-time monitoring of hardware devices such as AI cameras, controllable flashing alarm lights, loudspeakers, and lifting poles in the perception layer; displaying online status, operating parameters, and fault information of the devices; supporting remote control of device start / stop, parameter adjustment, fault reset, and other operations.

[0046] Data Management: Supports data query, filtering, export, deletion and other operations. It can query relevant data by time, location, event type and other conditions, and generate data backup and recovery plans.

[0047] Access control: Assign different operation permissions to users with different roles to ensure system data security and operational compliance.

[0048] (iv) Application Layer

[0049] The application layer provides users with a visual interface and function entry points, including PC software and mobile apps, supporting Android and iOS operating systems, and possessing the following core functions:

[0050] 1. Real-time monitoring function: After logging into the system, users can view the video footage of the maintenance work area in real time through the client, including the vehicle operation status in the upstream transition area captured by the first AI camera and the internal environment and personnel dynamics of the work area captured by the second AI camera; it supports single-view and multi-view switching display, and the camera angle (horizontal rotation angle 0-360°, vertical rotation angle -90° to +90°) and focal length can be remotely adjusted to achieve detailed viewing of different areas of the work area;

[0051] 2. Early Warning and Reminder Function: When the system identifies a vehicle entering the work area at risk, a hidden danger in the work area, or a violation of safety regulations, the application layer pushes early warning information to the user through terminal pop-ups, sound prompts, vibration alerts, etc. The early warning information includes the event type, time of occurrence, location, and relevant image / video screenshots. Users can view the early warning details through the client and issue a one-click response command.

[0052] 3. Video playback function: Supports querying and replaying work area videos within a specified time period by time, location, event type, etc. Video playback supports fast forward, slow motion, pause, screenshot, recording and other operations, making it easy for users to trace past work situations and analyze the causes of accidents or the rectification of hidden dangers.

[0053] 4. Data statistical analysis function: The platform automatically performs statistical analysis on the collected data such as the number of vehicle intrusion warnings, the number and types of hidden dangers identified, the number and types of violations, and the rectification rate of hidden dangers, and generates various statistical reports and analysis charts (such as bar charts, pie charts, line charts, etc.). Users can view and export reports through the client to provide data support for maintenance safety management decisions.

[0054] 5. Remote control function: Users can remotely control the sensing layer hardware devices through the client, including: controlling the raising and lowering of the lifting pole and its positioning; adjusting the illumination mode and flashing frequency of the controllable strobe alarm light; adjusting the sound intensity of the loudspeaker and switching the warning voice; controlling the rotation and zoom of the camera; and starting and stopping the sound and light warning function, etc.

[0055] 6. Rectification closed-loop management function: For identified hazards and violations of safety regulations in the work area, users can issue rectification instructions through the client, designate the person responsible for rectification and the rectification deadline. After the rectification is completed, the person responsible can upload supporting materials such as rectification photos and videos. Users can review and confirm the rectification status, forming a closed-loop management process of "identification-reminder-rectification-review-archiving".

[0056] 7. System settings function: Supports user-defined parameters such as warning thresholds (e.g., vehicle speeding thresholds, intrusion distance thresholds), data retention time, and warning reminder methods to adapt to the personalized needs of different maintenance operation scenarios.

[0057] Compared with the prior art, the present invention has the following significant advantages:

[0058] 1. Precise and efficient safety control: Through the specialized division of labor of dual AI cameras and the optimized AI algorithm model, the system can achieve high-precision identification of risks of social vehicles entering the work area, hidden dangers, and violations of safety regulations, with a false judgment rate of less than 5% and a false judgment rate of less than 3%. The system adopts a sound and light coordinated directional early warning method, with an on-site early warning response time of less than 1 second. At the same time, it links remote reminders to effectively improve the early warning effect and significantly reduce the incidence of safety accidents.

[0059] 2. High mobility and environmental adaptability: It adopts a liftable pole and multi-position fixed bracket design, which can be quickly fixed to pickup trucks, dump trucks and other work vehicles or on-site fixed structures. The deployment time is no more than 30 minutes, which is suitable for maintenance work scenarios of different locations and scales. The hardware has a high protection level, a wide temperature range and infrared night vision function, and can operate stably in complex environments such as high temperature, low temperature, rain, snow, sand and dust, and night.

[0060] 3. Complete closed-loop management process: Construct a closed-loop management process of "identification-early warning-supervision-rectification-traceability" to achieve full-chain control from risk identification to rectification and archiving. Combined with remote monitoring, video playback and data statistical analysis functions, it reduces the cost of manual supervision and improves the level of refinement and intelligence of maintenance safety management.

[0061] 4. Good system stability and compatibility: It adopts a layered architecture design and dual transmission mode, with strong software and hardware synergy, stable and reliable data transmission, and continuous operation without failure for no less than 1,000 hours; it has good scalability, supports online upgrade of algorithm models, expansion of functional modules and expansion of hardware equipment, and can be adapted to more maintenance operation scenarios and management needs according to actual needs.

[0062] 5. High ease of operation: The client software and mobile APP have a simple and intuitive interface and a simple operation process. Managers can view the work area, receive early warning information, and issue management instructions anytime and anywhere. Operators can quickly complete equipment deployment and daily operation without the need for professional technical training.

[0063] 6. Wide range of applications: Applicable to daily maintenance, emergency repair, and special projects of various types of roads such as highways, national and provincial trunk roads, and urban roads. It can provide rapid prevention and control solutions for small-scale temporary operations, as well as meet the routine safety management needs of large-scale maintenance projects, and has broad application prospects. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the overall architecture of the AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites, as described in this invention.

[0065] Figure 2 This is a three-dimensional structural diagram of the sensing layer hardware of the present invention;

[0066] Figure 3 This is a schematic diagram illustrating the deployment and working scenario of the present invention at a maintenance work site.

[0067] In the diagram: 1. Perception Layer, 11. First AI Camera, 12. Second AI Camera, 13. Controllable Flashing Alarm Light, 14. Loudspeaker, 15. Liftable Pole, 16. Fixed Mount, 17. Mounting Base; 2. Transmission Layer, 21. 4G / 5G Wireless Router, 22. Wired Network Switch; 3. Platform Layer, 31. Data Storage Module, 32. Data Processing Module, 33. Algorithm Model Management Module, 34. System Management Module; 4. Application Layer, 41. PC Software, 42. Mobile App. Detailed Implementation

[0068] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] The present invention will be further described in detail below with reference to specific embodiments:

[0070] (I) System Deployment

[0071] 1. Preliminary preparation: Based on the actual conditions such as the size of the maintenance work area, geographical location, work time, and traffic flow, determine the system deployment location (prioritize the anti-collision vehicle or fixed bracket on the upstream transition zone side of the work area), check the stability of the deployment location and the network signal coverage; prepare installation tools (wrench, screwdriver, level, etc.), fixing bolts, waterproof tape and other auxiliary materials.

[0072] 2. Hardware installation

[0073] Fixed bracket installation: Attach the arc-shaped clamping pieces of the two sets of metal buckles to the cargo box guardrail or fixed bracket of the work vehicle, adjust the buckle position to ensure that the two sets of buckles are 30-50cm apart vertically, use bolts to pass through the reserved holes in the guardrail and the fixing ends of the buckles, tighten the bolts to fix the fixed bracket 16, and use a level to correct the levelness of the bracket.

[0074] Installation of the liftable rod: Connect the lower section of the hydraulic lifting column to the drive compartment, ensuring a firm connection. Then, place the lower section of the column into the arc-shaped clamping plate of the fixing seat 16, adjust the verticality of the column, and tighten the locking bolts to make the liftable rod 15 stand vertically with the bottom anti-slip and shock-absorbing pads in close contact with the installation surface.

[0075] Top-mounted equipment installation: Fix the controllable strobe alarm light 13 to the mounting base 17 at the top of the lifting rod 15 via the connecting bracket, ensuring that the luminous area faces the upstream and downstream direction of the work area; fix the first AI camera 11 and the second AI camera 12 to the side wall of the mounting base 17 via the L-shaped metal bracket, so that the first AI camera 11 faces the upstream transition area of ​​the work area and the second AI camera 12 faces the inside of the work area, adjust the camera angle and focus to ensure that the monitoring coverage meets the requirements; connect the two loudspeakers 14 to the corresponding cameras respectively, ensuring that the loudspeaker openings face the horizontal outward.

[0076] Auxiliary equipment deployment: Place sound and light cones flexibly in key locations around the work area to ensure that wireless signals can receive the early warning signals of the sensing layer 1.

[0077] Wiring connection: Connect the wiring inside the first AI camera 11, the second AI camera 12, the controllable flashing alarm light 13, the loudspeaker 14 and the lifting pole 15, ensuring that the wiring is secure and wrapping the wiring with waterproof tape to waterproof it; connect the data transmission equipment (4G / 5G wireless router 21, wired switch 22) to the sensing layer 1 device to complete network access.

[0078] 3. Software Deployment: Deploy the platform layer 3 system management software on the cloud server, configure modules such as data storage module 31, data processing module 32, and algorithm model management module 33, import the hidden danger feature library and the "three violations" behavior feature library, and set early warning thresholds (such as vehicle speeding threshold, intrusion distance threshold, and feature matching degree threshold); install the computer software 41 or mobile APP 42 on the terminal devices such as computers and mobile phones of management personnel, and complete user registration and permission allocation.

[0079] (II) System Debugging

[0080] 1. Hardware debugging

[0081] Adjustment of the lifting pole: Start the hydraulic drive chamber and test the lifting function of the lifting pole 15 through remote control and local button respectively to ensure smooth lifting, accurate positioning, and stable stay at the preset height of 4m. Check whether the lifting speed meets the requirement of 0.1-0.2m / s.

[0082] Camera debugging: Start the first AI camera 11 and the second AI camera 12 to test the image acquisition quality, ensure that the picture is clear and without distortion, adjust the camera rotation angle and focal length, and verify the monitoring coverage; test the high-speed dynamic capture capability of the first AI camera 11, and check the vehicle feature extraction effect by simulating a fast-moving vehicle scenario; test the wide-angle coverage and infrared night vision function of the second AI camera 12, and verify the image clarity in night or dark room environments.

[0083] Debugging of sound and light warning equipment: Activate the controllable strobe alarm light 13, test the switching effect of different light-emitting modes (constant light, strobe, strobe), and check the brightness and penetration of the LED light-emitting area; activate the loudspeaker 14, test the adjustment effect of different sound intensities, play the preset warning voice, and verify the voice clarity and directional propagation effect.

[0084] Fixed stability test: Simulate outdoor wind load (by blowing air through a fan) and slight collision to check whether the fixed bracket 16 is securely fixed, whether the lifting rod 15 has no obvious tilt or shaking, and whether the hardware connection is reliable.

[0085] 2. Software and Algorithm Debugging

[0086] Algorithm recognition accuracy debugging: Simulate different scenarios (such as social vehicles approaching the work area at excessive speed, vehicles changing lanes and entering the work area at risk, equipment not being placed in a standardized manner, personnel not wearing safety helmets, and equipment being operated in violation of regulations, etc.) to test the recognition effect of the AI ​​algorithm model in the algorithm model management module 33, record the recognition accuracy, false positive rate and false negative rate, and optimize the model by adjusting the algorithm parameters and supplementing the feature library data to ensure that the recognition accuracy is not less than 95%.

[0087] Data transmission debugging: Test the transmission stability of the wireless network of 4G / 5G wireless router 21 and the wired network of wired network switch 22, check the real-time transmission of data such as images, videos, and warning signals, and verify the effectiveness of data encryption transmission and retransmission mechanisms.

[0088] Client-side function debugging: Test the real-time monitoring, early warning, video playback, and remote control functions of the computer software 41 and the mobile APP 42 to ensure that the data interaction between the client and the platform layer 3 and the perception layer 1 is normal, the early warning information is pushed in a timely manner, and the remote control command is responded to quickly.

[0089] 3. Collaborative Debugging: Simulate complete early warning scenarios, such as controlling a vehicle to approach the work area at a speed exceeding a preset threshold, verifying whether the first AI camera 11 accurately identifies the risk of intrusion, whether it triggers the controllable strobe alarm light 13 (red LED light area flashing) and the loudspeaker 14 (playing "You are speeding, slow down" and "Construction ahead, please change lanes immediately"), and whether the computer software 41 and the mobile APP 42 receive the early warning information in a timely manner; simulate a scenario where workers are not wearing safety helmets, verifying whether the second AI camera 12 accurately identifies the "three violations", whether it triggers the controllable strobe alarm light 13 (blue LED light area flashing) and the loudspeaker 14 (playing "Do not work without safety protective equipment", and whether the computer software 41 and the mobile APP 42 push reminder information and issue rectification instructions normally.

[0090] (III) System Operation

[0091] 1. Vehicle intrusion identification and early warning process

[0092] Data Acquisition: The first AI camera 11 collects vehicle images and video data in real time in the upstream transition zone of the maintenance work area, obtains vehicle location information through built-in sensors, and continuously transmits it to the local AI chip and platform layer 3.

[0093] Data analysis and risk assessment: The vehicle trajectory and speed analysis algorithm in the local AI chip's algorithm model management module 33 extracts vehicle features, tracks vehicle trajectory, calculates real-time speed, and, combined with the safety zone boundary of the work area, determines whether a vehicle has a tendency to intrude. When the vehicle speed exceeds a preset threshold (e.g., the road speed limit is 60km / h, and the preset overspeed threshold is 40km / h) and the distance from the work area is less than the intrusion distance threshold (e.g., 300m), it is determined to be a vehicle with intrusion risk.

[0094] Warning Trigger: The first AI camera 11 immediately sends a warning signal to the platform layer 3 through the transmission layer 2. The platform layer 3 simultaneously triggers on-site warnings and remote reminders: the red LED light area of ​​the controllable strobe alarm light 13 starts the strobe mode, and the loudspeaker 14 plays warning voice messages such as "You are speeding, slow down" and "Construction ahead, please change lanes immediately" in a loop. At the same time, the platform layer 3 pushes warning information to the computer software 41 and mobile APP 42 of the management personnel, including the real-time location, speed, and screenshot of the running trajectory of the vehicle.

[0095] Subsequent handling: Managers can view the warning details through computer software 41 or mobile APP 42, remotely adjust the focal length of the first AI camera 11 to track the target vehicle, and coordinate on-site personnel to take additional warning measures when necessary; after the vehicle leaves the risk area, the system automatically stops the warning and records the relevant data of the warning event to the data storage module 31.

[0096] 2. Self-inspection and early warning process for potential hazards and violations of safety regulations in the work area

[0097] Data Acquisition: The second AI camera 12 collects environmental images and video data of workers' behavior in the work area in real time and continuously transmits them to the local AI chip and platform layer 3.

[0098] Feature recognition and judgment: The local AI chip runs the work area hazard recognition algorithm and the "three violations" behavior recognition algorithm in the algorithm model management module 33, and matches the collected environmental features and personnel behavior features with the preset feature library; when the environmental feature matching degree exceeds 85%, it is judged as a work area hazard, and the location, type, discovery time and image evidence of the hazard are recorded; when the personnel behavior feature matching degree exceeds 80%, it is judged as a "three violations" behavior, and the personnel, behavior type, occurrence time and video evidence of the behavior are recorded, and the relevant data are synchronously stored in the data storage module 31.

[0099] Alert Trigger: The second AI camera 12 sends an alert signal to the platform layer 3 via the transmission layer 2. The platform layer 3 triggers on-site alerts and remote push notifications: the controllable strobe alarm light 13 activates the strobe mode in the blue LED light area, and the loudspeaker 14 plays the corresponding warning voice according to the event type (e.g., for hidden dangers, it plays "Hidden danger found in the work area, please rectify it in time"; for violations, it plays "You have violated regulations, please rectify it immediately"); at the same time, the platform layer 3 pushes alert information to the management personnel's computer software 41 and mobile APP 42, including event details and supporting materials.

[0100] Rectification closed-loop management: Managers issue rectification instructions through computer software 41 or mobile APP 42, designating the person responsible for rectification and the rectification deadline; after rectification is completed, the person responsible uploads rectification photos, videos and other materials through the client, and managers review and confirm them. After the review is approved, the system records the rectification results to the data storage module 31, forming closed-loop management; if rectification is not completed on time, the system will continuously remind and record the overdue situation.

[0101] 3. Remote monitoring and management process

[0102] Real-time monitoring: Managers can log in to the system via computer software 41 or mobile APP 42, select the corresponding maintenance work area, and view the video footage captured by the first AI camera 11 and the second AI camera 12 in real time. By adjusting the camera angle and focus, they can focus on key areas of the work area (such as equipment stacking area, worker concentration area, and vehicle traffic intersection area).

[0103] Video playback: As needed, managers can query historical video data in data storage module 31 by time, event type, and other conditions to review the playback, trace the operation, and analyze the causes of hidden dangers or violations.

[0104] Data statistical analysis: The data processing module 32 of platform layer 3 regularly compiles statistics on data such as the number of vehicle intrusion warnings, the number and type distribution of hazard identifications, the number of violations and personnel distribution, and the hazard rectification rate, generating weekly reports, monthly reports and analysis charts. Managers can view the reports through computer software 41 or mobile APP 42 to understand the safety status of the maintenance work area and optimize management strategies.

[0105] Remote control: According to actual needs, managers can remotely control the height of the lifting pole 15, adjust the monitoring angle and focus of the first AI camera 11 and the second AI camera 12, switch the illumination mode of the controllable flashing alarm light 13, adjust the sound intensity of the loudspeaker 14, and start and stop related equipment through computer software 41 or mobile APP 42.

[0106] (iv) System Maintenance

[0107] 1. Routine maintenance: Regularly (recommended once a week) check the appearance and fixation of the hardware devices in the perception layer 1, clean the lenses of the first AI camera 11 and the second AI camera 12, the light-transmitting cover of the controllable flashing alarm light 13 and the dustproof net of the loudspeaker 14 to ensure that the equipment is working properly; check whether the wiring connections are firm and whether the waterproofing is intact, and replace aging or damaged wiring and components in a timely manner.

[0108] 2. Software maintenance: Regularly (recommended once a month) update the AI ​​algorithm models and feature libraries in the algorithm model management module 33 of platform layer 3, and optimize algorithm parameters based on feedback data from actual application scenarios; back up the system data in the data storage module 31 to prevent data loss; check the running status of the computer software 41 and the mobile APP 42, and fix vulnerabilities and update the version in a timely manner.

[0109] 3. Fault Handling: A fault early warning mechanism is established. The system monitors the operating status of the hardware devices in the perception layer 1 and the software operation in the platform layer 3 in real time. When a fault is detected (such as device offline, data transmission interruption, or algorithm recognition anomaly), a fault reminder is promptly pushed to the administrator's computer software 41 and mobile APP 42. For common faults, a fault troubleshooting guide is provided (such as checking the network connection and power supply in the transmission layer 2 when the device is offline, and checking the algorithm parameters and feature library of the algorithm model management module 33 when identifying anomalies). Administrators can quickly troubleshoot and handle the faults according to the guide. Complex faults are repaired on-site by technical personnel.

[0110] 4. Personnel Training: Regularly conduct system operation training for management and operation personnel of maintenance units. The content includes deployment of sensing layer 1 equipment, operation of computer software 41 and mobile APP 42, daily maintenance, and troubleshooting of common faults, to ensure that relevant personnel are proficient in using the system and give full play to the system's safety prevention and management functions.

[0111] This invention organically integrates AI video perception technology, sound and light collaborative early warning technology, and remote management technology to achieve intelligent, real-time, and closed-loop management of vehicle intrusion prevention and hidden danger control, as well as self-inspection of "three violations" at highway maintenance sites. It effectively solves the pain points of traditional maintenance safety management, improves the safety and management efficiency of maintenance operations, and has significant value for promotion and application.

[0112] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A highway maintenance site AI video perception system for preventing vehicle intrusion and for self-inspection and early warning of potential hazards, characterized in that: The system adopts a layered architecture design, including a perception layer (1), a transmission layer (2), a platform layer (3), and an application layer (4). The perception layer (1) is used for data acquisition and on-site early warning. The transmission layer (2) is used for data transmission between the perception layer (1) and the platform layer (3). The platform layer (3) is the core control and data processing center. The application layer (4) provides a visual operation interface. The four layers work together to realize the functions of vehicle intrusion identification and early warning, self-inspection and early warning of hidden dangers and "three violations" behaviors, and remote monitoring and management.

2. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 1, characterized in that: The perception layer (1) includes a first AI camera (11), a second AI camera (12), a controllable strobe alarm light (13), a loudspeaker (14), a liftable pole (15), and a fixed bracket (16); the first AI camera (11) is deployed on the upstream transition zone side of the work area, with a resolution of not less than 1080P, a frame rate of not less than 30fps, and a protection level of not less than IP67; the second AI camera (12) is deployed on the work area side, with a horizontal field of view of not less than 120°, and a built-in infrared night vision module.

3. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 2, characterized in that: The controllable strobe alarm light (13) has a long, flat shell structure with red LED light-emitting areas at both ends and a blue LED light-emitting area in the middle. It supports switching between constant light, strobe, and burst flash modes, and the strobe frequency can be adjusted within the range of 5-20Hz. The bottom of the controllable strobe alarm light (13) is fixed to the mounting base (17) through a connecting bracket. The mounting base (17) is coaxially connected to the top of the liftable rod (15).

4. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 2, characterized in that: Two loudspeakers (14) are configured, which are in the shape of a frustum and are respectively linked with the first AI camera (11) and the second AI camera (12). The sound intensity can be adjusted in the range of 60-120dB. Multiple sets of warning voices are pre-stored and can be automatically switched and played according to the warning scenario. The loudspeaker (14) and the rear end of the camera housing are integrally injection molded, and the loudspeaker mouth faces the horizontal outer side.

5. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 2, characterized in that: The liftable pole (15) is a hydraulic lifting column structure made of high-strength aluminum alloy. The maximum lifting height can reach 4m, and the lifting speed is 0.1-0.2m / s. The height can be adjusted by remote control or local button. The lower section of the liftable pole (15) is provided with two ring-shaped metal reinforcing hoops, and the bottom end is connected to the cylindrical hydraulic drive chamber.

6. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 2, characterized in that: The fixed bracket (16) includes two sets of symmetrically distributed metal buckle structures. Each set of buckles consists of an arc-shaped clamping piece and a locking bolt. The inner side of the arc-shaped clamping piece is provided with an anti-slip rubber pad. The two sets of buckles are distributed axially along the liftable rod (15) with a spacing of 30-50cm, which is suitable for columnar carriers with a diameter of 5-12cm.

7. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 1, characterized in that: The transmission layer (2) adopts a dual transmission mode combining 4G / 5G wireless network and wired network, and is configured with a 4G / 5G wireless router (21) and a gigabit wired network switch (22); the 4G / 5G wireless router (21) supports multi-band communication and the maximum transmission rate is not less than 1Gbps; the gigabit wired network switch (22) has 4-8 gigabit Ethernet ports; the transmission process adopts SSL / TLS encryption protocol and has a data retransmission mechanism.

8. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 1, characterized in that: The platform layer (3) uses cloud computing technology to build a distributed cloud platform, including a data storage module (31), a data processing module (32), an algorithm model management module (33) and a system management module (34); the algorithm model management module (33) includes a vehicle running trajectory and speed analysis algorithm, a work area hidden danger identification algorithm, and a personnel "three violations" behavior identification algorithm.

9. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 8, characterized in that: The vehicle trajectory and speed analysis algorithm adopts the YOLO target detection algorithm and the Kalman filter target tracking algorithm; the work area hidden danger identification algorithm adopts the CNN convolutional neural network algorithm; the personnel "three violations" behavior identification algorithm adopts the OpenPose human posture estimation algorithm; the algorithm model management module (33) supports online upgrade function.

10. The AI ​​video perception system for preventing vehicle intrusion and self-inspection and early warning of potential hazards at highway maintenance sites according to claim 1, characterized in that: The application layer (4) includes computer software (41) and mobile APP (42), which supports real-time monitoring, early warning reminders, video playback, data statistical analysis, remote control and rectification closed-loop management functions, and is compatible with Android and iOS operating systems.

Images