Portable track area intelligent fence alarm and determination method thereof

Abstract

The application discloses a portable track area intelligent fence alarm and a determination method thereof. The alarm comprises a base and a vertical rod. The top of the vertical rod is integrated with a red flashing light and a camera with an LED light supplement. A laser emitter and a voice speaker are installed in the middle of the vertical rod. The base is sealed and has an intelligent recognition hardware built-in, which is connected with each component. The laser emitter projects a red laser strip in the shape of a Chinese character to form an alarm fence. The intelligent hardware runs a boundary recognition model and executes a digital twin early warning logic. A three-dimensional digital map is pre-stored. A dynamic track with geographical coordinates is generated by mapping the personnel in a two-dimensional image and the laser fence coordinates. Multiple risk areas are divided. A first-level risk voice warning is given. A zero-level danger triggers a flash and a high-frequency alarm. An encrypted evidence chain is uploaded to the cloud. The application provides reliable technical support for track area construction safety.

Description

Technical Field

[0001] This invention belongs to the field of machine vision inspection technology, and more specifically, relates to a portable intelligent fence alarm for track areas and its determination method. Background Technology

[0002] As rail transit projects continue to be constructed and their operation time increases, the frequency of construction work in the track area is increasing, leading to a surge in safety accidents in the track area. Because rail transit involves numerous specialties, the types and numbers of workers working in the track area are diverse, and they often move between different lines, resulting in unfamiliarity with the construction site environment and job duties. Consequently, workers frequently cross designated construction areas during construction, causing injuries ranging from falling objects and mechanical damage to serious accidents resulting in mass casualties.

[0003] To prevent construction workers from crossing boundaries in the track area, existing control methods mainly involve two approaches: conducting safety patrols of the construction area or placing flashing red warning lights. However, both methods have significant drawbacks. For example, personnel patrols are greatly influenced by subjective factors and are difficult to achieve full coverage. Furthermore, when construction areas overlap, it is easy to confuse the location of the construction boundary. Therefore, there is an urgent need to develop a portable intelligent fence alarm and judgment method that can proactively identify personnel crossing boundaries. This would enable timely and proactive alarms when workers cross the boundary in the track area, reminding them to work safely within the designated area and reducing safety accidents in the track area. Summary of the Invention

[0004] This invention aims to address the problems of construction workers easily crossing boundaries in rail transit track areas and the numerous loopholes in traditional manual patrols and red flashing light warnings. By combining portable intelligent alarm devices with laser fencing, machine vision, and digital twin technology, it proactively identifies boundary crossing behavior, issues tiered early warnings and alarms, and uploads encrypted evidence, thereby improving the level of safety management and control in track area construction and reducing the occurrence of safety accidents.

[0005] In view of the above-mentioned defects or improvement needs of the prior art, as a first aspect of the present invention, the present invention provides a portable intelligent fence alarm for track areas, comprising: The system comprises a base and a vertical pole mounted on the base. The top of the vertical pole integrates a red flashing light and a camera, with the camera equipped with an LED fill light for auxiliary shooting. A laser emitter and a speaker are installed in the middle of the vertical pole. The laser emitter projects a linear red laser strip that overlaps with the construction boundary line to form an alarm fence. The base integrates intelligent recognition hardware, which is built into a pre-reserved sealed cavity within the base and connected to the camera, laser emitter, red flashing light, and speaker. This intelligent recognition hardware is used to identify potential personnel crossing the boundary, determining whether construction personnel overlap with the laser projection line based on collected image information. Upon determining a boundary crossing, it outputs a control command to trigger a flashing red light and an alarm sounding the speaker. The intelligent recognition hardware is configured to execute boundary crossing warning linkage logic based on digital twins, including: A three-dimensional digital map of the construction environment of the track area is pre-stored, and the personnel positions and laser fence coordinates in the two-dimensional images captured by the camera are mapped onto the three-dimensional digital map to generate a dynamic trajectory with geographic coordinates. Based on the distance between personnel and the preset electronic fence in the 3D digital map, multiple risk zones are divided. When it is determined that someone has entered a Level 1 risk zone, a voice prompt will be played. When it is determined that someone has crossed the laser projection line into a Level 0 danger zone, a red flashing light and a high-frequency buzzer alarm will be triggered, and an encrypted evidence chain with timestamps and coordinates will be uploaded to the cloud management platform.

[0006] Furthermore, a circular hole is provided in the middle of the upright for leading out a power cord.

[0007] Furthermore, the intelligent recognition hardware is also equipped with a laser line enhancement and extraction module, the specific processing logic of which is as follows: The system acquires the raw image captured by the camera and converts it to the HSV color space. Based on the wavelength characteristics of the red laser, a dynamic threshold range is set to extract the red channel component. Morphological opening operations are used to remove noise from the image, and the pixel-level linear equation of the laser line is fitted using Hough transform. If the confidence level of the linear equation is detected to be lower than the preset threshold, the flicker frequency of the LED fill light and the projection power of the laser emitter are automatically adjusted until clear laser line features are extracted.

[0008] Furthermore, an attitude self-calibration mechanism is provided at the connection between the upright and the base: The base integrates a six-axis gyroscope sensor for real-time monitoring of the device's tilt angle and vibration frequency. When the tilt angle exceeds the preset threshold, the intelligent recognition hardware automatically pauses the alarm output and enters the calibration mode. It then uses an image recognition algorithm to re-detect the horizon or track level and calculates the projection transformation matrix to compensate for the digital twin mapping relationship in real time, thereby eliminating detection errors caused by device displacement.

[0009] Furthermore, the process of automatically adjusting the flicker frequency of the LED fill light and the projection power of the laser emitter specifically includes: The mean value of the grayscale histogram of the original image is calculated as the ambient brightness index. If the ambient brightness index is lower than the preset dark light threshold, it is determined to be a low-light environment. Light-coordinated control steps: In low-light environments, the flicker frequency of the LED fill light is synchronously locked to twice the camera frame rate, and the duty cycle is gradually increased using pulse width modulation until the average image brightness reaches the target range. A laser line signal-to-noise ratio feedback loop is constructed to calculate the gray-level variance ratio of the extracted laser pixel set and the background noise in real time. If the gray-level variance ratio is less than the judgment threshold, the driving current of the laser emitter is linearly increased by a preset step size to increase the projection power until the gray-level variance ratio meets the clarity requirements or reaches the hardware safe power limit.

[0010] Furthermore, in the process of mapping to the three-dimensional digital map, a visual positioning method based on feature point matching is used for coordinate transformation, specifically including: Extract SURF feature points from fixed reference objects in the orbital area and construct a local feature point database; calculate the principal direction and the Haar wavelet response of the neighborhood centered on the feature point for each extracted SURF feature point to generate a 64-dimensional or 128-dimensional feature description vector. The nearest neighbor and second nearest neighbor distance ratio method is used to perform preliminary matching between real-time image feature points and database feature points, and a distance ratio threshold is set to remove erroneous matching point pairs caused by blurry or duplicate textures; The minimum point set is randomly selected from the selected matching point pairs. The candidate homography matrix is ​​calculated using the direct linear transformation algorithm. The number of interior points is counted by calculating the reprojection error of all matching point pairs under this matrix. The process is iteratively updated until the number of interior points converges. The final homography matrix is ​​then solved by optimizing the least squares method using all interior point sets, thereby determining the pose of the camera.

[0011] Furthermore, in the process of generating the dynamic trajectory with geographic coordinates, an improved SORT multi-target tracking method is employed, specifically including: For each active human target trajectory, a Kalman filter based on a linear constant velocity model is used, based on the state estimation vector from the previous time step. ,in The coordinates of the detection box center are, Aspect ratio, For height, Given the rates of change of the corresponding state variables, predict the prior state at the current moment. and covariance matrix ; Calculate the association cost matrix between all detection boxes and prediction boxes in the current frame. For any element in the matrix The calculation formula is as follows: ;in Indicates the first Detection box With the Prediction boxes The intersection and union ratio, Represents the appearance feature vectors of both The cosine distance between them These are weighting coefficients used to balance the contributions of geometric position and appearance features; Cost matrix The input is fed into the Hungarian algorithm to find the matching subset with the minimum global cost. For successfully matched trajectory-detection pairs, the state of the Kalman filter is updated using the position information of the detection boxes, and the posterior state estimate is calculated. and covariance For unmatched trajectories, they are considered temporarily lost and retained in the buffer; for unmatched detections, they are considered new targets and new trajectories are initialized.

[0012] Furthermore, the process of uploading the encrypted evidence chain specifically includes: at the moment of boundary violation determination, capturing a video clip containing the boundary violation behavior and the corresponding three-dimensional coordinate trajectory data; generating a SHA-256 hash value using the hardware encryption chip built into the device, and performing RSA asymmetric encryption on the hash value; packaging the encrypted hash value with the original data, and synchronizing it to the cloud management platform through blockchain network nodes.

[0013] Furthermore, the division of the multi-level risk areas adopts a dynamic determination logic based on time to collision (TTC): Based on the continuous trajectory coordinates of personnel in a 3D digital map, the real-time movement velocity vector of personnel is calculated using the finite difference method. and the direction angle of movement Simultaneously, obtain the normal vector of the preset electronic fence or laser projection line. and the danger zone; Calculate the vertical projection distance from the operator's current location to the electronic fence or laser projection line. And combine the projection component of the velocity vector in the direction of the normal vector. Calculate time to collision value Among them, when If the person is moving away from or parallel to the fence, a risk assessment is not triggered. As people move toward the fence And the calculated time to collision value Less than the first preset time threshold When personnel are deemed to have entered a Level 1 risk area, an early warning signal is triggered. When the personnel's position coordinates are detected to cross the boundary of the laser projection line, and the velocity vector points towards the danger side... Furthermore, if the location is in the danger zone, it is determined that the person has entered the level zero danger zone, and an emergency alarm signal is triggered at this time; For moving horizontally only near the fence Or move in the opposite direction The system automatically maintains a safe status for personnel, thereby eliminating false alarms caused by personnel passing by or working with their backs to the fence.

[0014] As a second aspect of the present invention, a control method for a portable smart fence alarm for a track area is also provided, applied to the aforementioned portable smart fence alarm for a track area, comprising: S1. Using intelligent recognition hardware to load a pre-stored 3D digital map of the track area construction environment, obtain the coordinate position of the laser projection line projected by the laser emitter on the map; at the same time, collect images of the construction area in real time through a camera, map the personnel positions identified in the images onto the 3D digital map, and generate dynamic personnel trajectories with geographic coordinates. S2. Based on the personnel's dynamic trajectory, calculate the real-time distance between the personnel and the preset electronic fence or laser projection line, and divide the risk areas into multiple levels according to the distance threshold; the multiple risk areas include at least a level one risk area and a level zero danger area; S3. When it is determined that a person has entered a Level 1 risk area, the intelligent recognition hardware outputs the first control command to control the voice speaker to play a warning tone. S4. When a person is identified as crossing the laser projection line into a level-zero danger zone, the intelligent recognition hardware outputs a second control command, triggering a red flashing light and a high-frequency alarm from the voice speaker. At the same time, it generates an encrypted evidence chain with timestamps and coordinates and uploads it to the cloud management platform through the communication module.

[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The portable intelligent fence alarm for track areas of the present invention uses a base and a pole as the core main structure. A red flashing light and a camera equipped with an LED supplementary light are integrated at the top of the pole. A laser emitter and a voice speaker are installed in the middle of the pole. The base has a reserved sealed cavity for housing intelligent recognition hardware and is stably connected to each functional component. The laser emitter projects a straight red laser strip that coincides with the construction boundary line to form an alarm fence. The intelligent recognition hardware has a hazard identification model for personnel crossing the boundary. Based on the image information collected by the camera, it determines the overlap between the construction personnel and the laser projection line. When a boundary crossing is determined, a control command is output to trigger the red flashing light and the voice speaker to alarm. The integrated hardware structure realizes the construction boundary visualization fence construction, on-site image acquisition, and basic boundary crossing alarm. The device is portable and the sealed design is suitable for track area construction scenarios. All components work together to complete basic protection and alarm execution.

[0016] 2. The portable intelligent fence alarm for track areas of the present invention configures boundary crossing warning and linkage logic based on digital twins in the intelligent identification hardware, pre-stores a three-dimensional digital map of the track area construction environment, maps the personnel position in the two-dimensional image captured by the camera and the laser fence coordinates to the three-dimensional digital map, generates a dynamic trajectory of personnel with geographical coordinates, divides multiple risk zones according to the distance between personnel and the preset electronic fence, controls the voice speaker to play a warning sound when personnel enter the first-level risk zone, and triggers a red flashing light and a high-frequency buzzer alarm when crossing the laser projection line into the zero-level danger zone, and generates an encrypted evidence chain with timestamp and coordinates and uploads it to the cloud management platform, realizing graded warning alarm and safety evidence retention, accurately distinguishing risk levels, improving the pertinence of warning alarms, and facilitating construction safety management and behavior traceability.

[0017] 3. The portable intelligent fence alarm for track areas of the present invention, through the configuration of a laser line enhancement and extraction module, an attitude self-calibration mechanism, a low-light environment adaptation mechanism, a GPS-free visual positioning method, an improved multi-target tracking method, and a dynamic false alarm judgment logic, accurately extracts laser line features and automatically adjusts the supplementary light and laser projection power. It relies on a six-axis gyroscope to monitor the device's attitude and compensate for detection errors. It completes the coordinate transformation from two-dimensional image to three-dimensional map through SURF feature point matching, uses an improved SORT algorithm to track the dynamic trajectory of personnel, judges risks based on time-to-collision logic and eliminates false alarms, and uploads the evidence chain synchronously through hardware encryption and blockchain. It is adapted to complex scenarios in track areas without GPS, in low light, and where equipment is easily displaced, improving the accuracy of boundary crossing detection and the stability of device operation, and reducing the probability of invalid alarms. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of a portable smart fence alarm for a track area according to an embodiment of the present invention; Figure 2This is a schematic diagram illustrating the triggering of an alarm according to an embodiment of the present invention; Figure 3 This is a flowchart of a method according to an embodiment of the present invention; In all the accompanying drawings, the same reference numerals indicate the same technical features, specifically: 1-Red flashing light, 2-LED fill light, 3-Camera, 4-Laser emitter, 5-Speaker, 6-Standing pole, 7-Power cord, 8-Bracket, 9-Base, 10-Intelligent recognition hardware, 11-Six-axis gyroscope sensor. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0020] Example 1 Please refer to Figure 1 This embodiment 1 provides a portable smart fence alarm for a track area, comprising: The system comprises a base 9 and a pole 6 mounted on the base 9. The top of the pole 6 integrates a red flashing light 1 and a camera 3, with the camera 3 equipped with an LED fill light 2 for auxiliary shooting. A laser emitter 4 and a speaker 5 are installed in the middle of the pole 6. The laser emitter 4 projects a straight red laser strip that overlaps with the construction boundary line to form an alarm fence. The base 9 integrates intelligent identification hardware 10, which is housed within a pre-reserved sealed cavity and connected to the camera 3, laser emitter 4, red flashing light 1, and speaker 5. The intelligent identification hardware 10 is used to identify potential personnel crossing the boundary, determining whether construction personnel overlap with the laser projection line based on collected image information. Upon determining a boundary crossing, it outputs a control command to trigger the red flashing light 1 and the speaker 5 to sound an alarm. The intelligent identification hardware 10 is configured to execute boundary crossing warning linkage logic based on digital twins, including: A three-dimensional digital map of the construction environment of the track area is pre-stored, and the personnel positions and laser fence coordinates in the two-dimensional images captured by the camera are mapped onto the three-dimensional digital map to generate a dynamic trajectory with geographic coordinates. Based on the distance between personnel and the preset electronic fence in the 3D digital map, multiple risk zones are divided. When it is determined that someone has entered a Level 1 risk zone, a voice prompt will be played. When it is determined that someone has crossed the laser projection line into a Level 0 danger zone, a red flashing light and a high-frequency buzzer alarm will be triggered, and an encrypted evidence chain with timestamps and coordinates will be uploaded to the cloud management platform.

[0021] This embodiment 1 further elaborates on the above content.

[0022] During the equipment deployment and self-calibration phase, the operator first places the portable smart fence alarm for the track area at a predetermined location beside the track. After the equipment is powered on, the six-axis gyroscope sensor 11 integrated inside the base 9 (located at the connection between the upright 6 and the base 9 for attitude self-calibration) immediately monitors the tilt angle and vibration frequency of the equipment in real time. If the detected tilt angle exceeds a preset threshold, the intelligent recognition hardware 10 automatically pauses the alarm output and enters calibration mode. At this time, the camera 3 acquires an image of the scene in front, and then re-detects the horizon or track horizontal line through an image recognition algorithm, calculates the projection transformation matrix, and performs real-time compensation for the digital twin mapping relationship, thereby eliminating the detection error caused by equipment displacement and ensuring the accuracy of subsequent monitoring benchmarks.

[0023] During the operation monitoring phase, laser emitter 4 projects a straight red laser stripe that overlaps with the construction boundary line, forming a visible physical alarm fence. Simultaneously, camera 3 continuously captures raw images of the monitored area at a high frame rate. To cope with the complex lighting environment of the track area, the laser line enhancement and extraction module built into the intelligent recognition hardware 10 begins operation. The captured images are converted to the HSV color space, and a dynamic threshold is set based on the wavelength characteristics of the red laser to extract the red channel component. Noise is removed using morphological opening operations, and the pixel-level linear equation of the laser line is fitted using Hough transform. If the confidence level of the detected laser line is insufficient, the flicker frequency of the LED fill light 2 and the projection power of the laser emitter 4 are automatically adjusted. Specifically, the mean value of the grayscale histogram of the raw image is first calculated as the ambient brightness index. If this index is lower than the preset dark light threshold, it is determined to be a low-light environment, and the light-coordinated control step is initiated: the flicker frequency of the LED fill light 2 is synchronously locked to twice the camera frame rate, and the duty cycle is gradually increased using pulse width modulation until the average image brightness reaches the target range. Simultaneously, a laser line signal-to-noise ratio feedback loop is constructed to calculate the grayscale variance ratio of the extracted laser pixel set and the background noise in real time. If the ratio is less than the judgment threshold, the driving current of the laser emitter 4 is linearly increased by a preset step size to improve the projection power until the clarity requirements are met, ensuring that the fence position can still be locked in low-light or high-dust environments.

[0024] To achieve the mapping from 2D images to 3D space, a visual localization method based on feature point matching is used to determine the camera pose. During the initialization or mapping phase, images are first acquired of fixed reference objects in the track area environment (such as fixed markers on tunnel walls, trackside mileage markers, or fixed structural components). A robust feature detection algorithm is then used to detect key points in the images; these key points possess scale and rotation invariance. For each detected feature point, the principal direction is first determined by calculating the Haar wavelet response within its neighborhood to achieve rotation invariance. Subsequently, a square neighborhood aligned with this principal direction is constructed around the feature point, and this neighborhood is divided into... The system calculates the Haar wavelet responses in the horizontal and vertical directions within each sub-region. and By performing a Gaussian weighted summation on these response values ​​and their absolute values, a summation containing... Feature description vector of each component Its form can be expressed as In scenarios requiring higher resolution, the sub-regions can be further subdivided into more refined gradient distribution statistics to generate 128-dimensional feature description vectors, thereby constructing a highly robust local feature point database.

[0025] During the device operation and real-time positioning phase, the camera captures current real-time images and repeats the feature extraction and description process described above to generate a real-time feature set. To achieve accurate association between the real-time images and the pre-stored database, the system uses the nearest neighbor to second nearest neighbor distance ratio method for initial matching. For any feature point in the real-time image... The system searches the database for the two feature points with the closest Euclidean distance. and System calculation and distance and with distance By setting a distance ratio threshold (Usually the value is between 0.6 and 0.8), when the condition is met When the matching point pair is determined to be a valid match, this step effectively eliminates fuzzy matches caused by repetitive textures within the tunnel (such as continuous segment joints) or changes in lighting, retaining a set of matching point pairs with high confidence. .

[0026] Based on the filtered set of matching points The homography matrix from the image plane to the 3D map plane is calculated using a random sampling consistency algorithm. Homography matrix It is The matrix describes the projection transformation relationship between the two planes. The algorithm first starts from... A minimum set of points (i.e., 4 pairs of non-collinear matching points) is randomly selected from the dataset, and a candidate homography matrix is ​​solved using the direct linear transformation algorithm. Subsequently, the system calculates the set. All points in Reprojection error under transformation Its calculation formula is ,in For image points, These are the corresponding map points. If there is an error... Less than the preset geometric error threshold If a point is found to be an interior point, it is determined to be an interior point. The algorithm iterates through the above process, counting the number of interior points generated in each iteration, until the number of interior points converges or the maximum number of iterations is reached. Finally, using all identified interior point sets, the optimal homography matrix is ​​solved by nonlinear optimization using the least squares method. This matrix describes the rotational and translational pose of the camera's optical center in the world coordinate system of the orbital area, mapping the personnel's position and the laser fence coordinates onto a pre-stored 3D digital map, providing a geometric reference for subsequent personnel boundary crossing determination.

[0027] In the target tracking and trajectory generation process, an improved SORT multi-target tracking method is employed. For each active personnel target trajectory, a Kalman filter based on a linear constant velocity model is used, based on the state estimation vector from the previous time step. ,in The coordinates of the detection box center are, Aspect ratio, For height, Given the rates of change of the corresponding state variables, predict the prior state at the current moment. and covariance matrix ; Calculate the association cost matrix between all detection boxes and prediction boxes in the current frame. For any element in the matrix The calculation formula is as follows: ;in Indicates the first Detection box With the Prediction boxes The intersection and union ratio, Represents the appearance feature vectors of both The cosine distance between them These are weighting coefficients used to balance the contributions of geometric position and appearance features; Cost matrix The input is fed into the Hungarian algorithm to find the matching subset with the minimum global cost. For successfully matched trajectory-detection pairs, the state of the Kalman filter is updated using the position information of the detection boxes, and the posterior state estimate is calculated. and covariance For unmatched trajectories, they are considered temporarily lost and retained in the buffer; for unmatched detections, they are considered new targets and new trajectories are initialized.

[0028] Based on the generated trajectory, a dynamic time-to-collision determination logic is used to delineate risk areas. The real-time movement velocity vector of personnel is calculated using the finite difference method. And the direction angle of movement, to obtain the normal vector of the electronic fence. Calculate the vertical projection distance from the operator's current position to the fence. and the projection component of the velocity vector in the direction of the normal vector. Then calculate the time to collision value. .

[0029] For moving horizontally only near the fence Or move in the opposite direction The system automatically maintains a safe state assessment, thereby eliminating false alarms caused by personnel passing by or working with their backs to the fence. When personnel move towards the fence... And time to collision value When the time threshold is less than the first preset time threshold, it is determined that the person has entered a level one risk zone, and the intelligent recognition hardware 10 controls the voice speaker 5 to play a warning tone; if the personnel's position coordinates are detected to cross the boundary of the laser projection line and the velocity vector is pointing towards the danger side... If the location is within the danger zone, then the person is considered to have entered a level zero danger zone. At this point, refer to... Figure 2 The intelligent identification hardware 10 outputs control commands to trigger a red flashing light 1 and a high-frequency buzzer alarm to warn personnel to evacuate. Simultaneously, it initiates the encrypted evidence chain upload process: at the moment of boundary violation determination, it captures video clips containing the boundary violation behavior and corresponding 3D coordinate trajectory data; it uses the device's built-in hardware encryption chip to generate a SHA-256 hash value and performs RSA asymmetric encryption on the hash value; finally, it packages the encrypted hash value and the original data, and synchronizes them to the cloud management platform through blockchain network nodes, completing the entire process of intelligent monitoring and evidence solidification.

[0030] In addition, this embodiment also provides a simplified implementation method, as follows: Please refer to Figure 1 as well as Figure 2It consists of a pole 6 and functional modules integrated at different heights on the pole (red flashing light 1, LED supplementary light 2, camera 3, laser emitter 4, voice speaker 5, power cord 7, bracket 8, base 9, and intelligent recognition hardware 10 (computing power box) integrated inside the base 9). Each component is assembled in a modular manner to achieve active recognition and audible and visual alarms for personnel crossing boundaries. Its specific structural design, component parameters, assembly relationships, and core module collaborative logic are detailed below: In this simplified embodiment, the base 9 serves as the core load-bearing component of the device. It is manufactured using high-strength ABS engineering plastic through a one-piece injection molding process, with overall dimensions of 300mm × 300mm × 150mm and a single base weight of ≤1.2kg, balancing portability and structural strength. A 3mm thick rubber anti-slip pad is attached to its bottom, with a diamond-shaped anti-slip texture on the surface, effectively preventing the device from shifting due to construction vibrations or personnel collisions.

[0031] In this simplified embodiment, the intelligent identification hardware 10 (computing power box) is built into a reserved sealed chamber inside the base 9. The chamber measures 150mm × 80mm × 50mm, and four symmetrical 8mm diameter heat dissipation holes are opened on the side wall of the chamber to prevent hardware short circuits or performance degradation. The intelligent identification hardware 10 has a built-in processor and storage unit, supports edge computing, and can independently run the "personnel boundary crossing hazard identification model." The model is designed for track area construction scenarios, with high identification accuracy and low latency. The computing power box connects to various functional modules through multiple interfaces to realize data reception, analysis, and command output.

[0032] In this simplified embodiment, the bracket 8 is fixed to the base 9 and the upright 6, forming a stable 45° triangular structure, which can further improve the stability of the device and prevent the upright from tipping over.

[0033] In this simplified embodiment, the upright 6 is made of round metal tubing, the length of which can be adjusted according to the construction scenario (typically 1.2m-1.8m). It has moisture-resistant and corrosion-resistant properties, making it suitable for environments such as railway tracks and tunnels. The bottom of the upright is machined with external threads, which are screwed into the internal thread interface of the base for fixation, ensuring verticality.

[0034] In this simplified embodiment, the power cord 7 is led out from the round hole in the middle of the pole, mainly because the track area may be damp and waterlogged. To prevent water from entering the equipment, the power cord is not led out from the bottom. Its waterproof rating meets the construction requirements of the track area.

[0035] In this simplified embodiment, a fixed speaker 5 is built into the pole 6 400mm from the base. The speaker power meets the voice transmission requirements in noisy environments in the track area, has pre-stored special warning voices for the track area, can be played in a loop, and supports volume adjustment to adapt to different noise intensity scenarios.

[0036] In this simplified embodiment, the laser emitter 4 is built 200mm above the speaker and emits a red, straight laser line to form a continuous laser boundary line at the construction boundary. The projection distance meets the requirements of conventional construction. The emitter can be finely adjusted to ensure that the laser line coincides with the construction boundary marking line, thus forming a construction fence.

[0037] In this simplified example, camera 3 can cover an area within 10m of the construction boundary. With the help of LED fill light 2, it supports low-light shooting and is suitable for tunnels, nighttime and other scenarios. The camera has a wide-angle shooting range to ensure coverage of the construction boundary and no blind spots.

[0038] In this simplified example, LED fill light 2 mainly serves as a fill light, providing an auxiliary light source for the camera to adapt to the low lighting environment of the track area.

[0039] In this simplified example, the red flashing light 1 uses a 12V LED light source, emits red light, and has an impact-resistant casing to withstand impacts from gravel and tools during construction. It includes both flashing and strobe modes. Under normal circumstances, it flashes at a fixed frequency to meet the passive warning requirements of the track area; when personnel are detected crossing the boundary, the red flashing light receives an alarm command and can switch to strobe mode to remind workers to return to a safe construction area.

[0040] Example 2 Please refer to Figure 3 This embodiment 2 provides a control method for a portable smart fence alarm for railway tracks, applied to the aforementioned portable smart fence alarm for railway tracks, including: S1. Using intelligent recognition hardware 10, load the pre-stored three-dimensional digital map of the track area construction environment, and obtain the coordinate position of the laser projection line projected by laser emitter 4 on the map; at the same time, use camera 3 to collect images of the construction area in real time, map the personnel positions identified in the images onto the three-dimensional digital map, and generate dynamic personnel trajectories with geographic coordinates. S2. Based on the personnel's dynamic trajectory, calculate the real-time distance between the personnel and the preset electronic fence or laser projection line, and divide the risk areas into multiple levels according to the distance threshold; the multiple risk areas include at least a level one risk area and a level zero danger area; S3. When it is determined that a person has entered a Level 1 risk area, the intelligent identification hardware 10 outputs the first control command to control the voice speaker 5 to play a warning tone. S4. When it is determined that a person has crossed the laser projection line into the level zero danger zone, the intelligent identification hardware 10 outputs a second control command, triggering the red flashing light 1 to flash and the voice speaker 5 to sound a high-frequency alarm. At the same time, it generates an encrypted evidence chain with timestamps and coordinates and uploads it to the cloud management platform through the communication module.

[0041] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A portable intelligent fence alarm for a track area, characterized in that it includes: The base (9) and the pole (6) installed on the base (9) are equipped with a red flashing light (1) and a camera (3) at the top of the pole (6). The camera (3) is equipped with an LED fill light (2) for auxiliary shooting. A laser emitter (4) and a voice speaker (5) are installed in the middle of the pole (6). The laser emitter (4) is used to project a red laser strip that overlaps with the construction boundary line to form an alarm fence. The base (9) is equipped with intelligent identification hardware (10). The intelligent identification hardware (10) is built into the sealed cavity reserved in the base (9) and is connected to the camera (3), laser emitter (4), red flashing light (1) and voice speaker (5). The intelligent identification hardware (10) is used to identify the hidden danger of personnel crossing the boundary. Based on the collected image information, it determines whether the construction personnel overlap with the laser projection line and outputs control commands to trigger the red flashing light (1) to flash and the voice speaker (5) to alarm when the boundary is determined. The intelligent identification hardware (10) is configured to execute boundary crossing warning linkage logic based on digital twins, including: A three-dimensional digital map of the construction environment of the track area is pre-stored, and the personnel positions and laser fence coordinates in the two-dimensional images captured by the camera are mapped onto the three-dimensional digital map to generate a dynamic trajectory with geographic coordinates. Based on the distance between personnel and the preset electronic fence in the 3D digital map, multiple risk zones are divided. When it is determined that someone has entered a Level 1 risk zone, a voice prompt will be played. When it is determined that someone has crossed the laser projection line into a Level 0 danger zone, a red flashing light and a high-frequency buzzer alarm will be triggered, and an encrypted evidence chain with timestamps and coordinates will be uploaded to the cloud management platform.

2. The portable intelligent fence alarm for track area according to claim 1 is characterized in that the middle part of the pole (6) is provided with a round hole for leading out a power line (7).

3. The portable intelligent fence alarm for track areas according to claim 1, characterized in that the intelligent identification hardware (10) is further equipped with a laser line enhancement and extraction module, the specific processing logic of which is: The original image captured by the camera (3) is obtained and converted to the HSV color space; the dynamic threshold range is set according to the wavelength characteristics of the red laser, and the red channel component is extracted; noise in the image is removed by morphological opening operation, and the pixel-level linear equation of the laser line is fitted by Hough transform; if the confidence of the linear equation is detected to be lower than the preset threshold, the flicker frequency of the LED fill light (2) and the projection power of the laser emitter (4) are automatically adjusted until clear laser line features are extracted.

4. A portable intelligent fence alarm for a track area according to claim 1, characterized in that a posture self-calibration mechanism is provided at the connection between the pole (6) and the base (9): The base (9) integrates a six-axis gyroscope sensor (11) for real-time monitoring of the device's tilt angle and vibration frequency. When the tilt angle exceeds the preset threshold, the intelligent recognition hardware (10) automatically pauses the alarm output and enters the calibration mode. It re-detects the horizon or track horizontal line through the image recognition algorithm, calculates the projection transformation matrix to compensate the digital twin mapping relationship in real time, so as to eliminate the detection error caused by the device displacement.

5. The portable intelligent fence alarm for a track area according to claim 1, characterized in that, The process of automatically adjusting the flicker frequency of the LED fill light (2) and the projection power of the laser emitter (4) specifically includes: The mean value of the grayscale histogram of the original image is calculated as the ambient brightness index. If the ambient brightness index is lower than the preset dark light threshold, it is determined to be a low-light environment. Light-coordinated control steps: In low-light environments, the flicker frequency of the LED fill light (2) is synchronously locked to twice the frame rate of the camera (3), and the duty cycle is gradually increased by pulse width modulation until the average brightness of the image reaches the target range. Construct a laser line signal-to-noise ratio feedback loop to calculate the grayscale variance ratio of the extracted laser pixel set and the background noise in real time. If the grayscale variance ratio is less than the judgment threshold, the driving current of the laser emitter (4) is linearly increased by a preset step size to improve the projection power until the grayscale variance ratio meets the clarity requirements or reaches the hardware safety power limit.

6. A portable intelligent fence alarm for a track area according to claim 1, characterized in that, in the process of mapping to the three-dimensional digital map, a visual positioning method based on feature point matching is used for coordinate transformation, specifically including: Extract SURF feature points from fixed reference objects in the orbital area and construct a local feature point database; calculate the principal direction and the Haar wavelet response of the neighborhood centered on the feature point for each extracted SURF feature point to generate a 64-dimensional or 128-dimensional feature description vector. The nearest neighbor and second nearest neighbor distance ratio method is used to perform preliminary matching between real-time image feature points and database feature points, and a distance ratio threshold is set to remove erroneous matching point pairs caused by blurry or duplicate textures; The minimum point set is randomly selected from the selected matching point pairs. The candidate homography matrix is ​​calculated using the direct linear transformation algorithm. The number of inliers is counted by calculating the reprojection error of all matching point pairs under this matrix. The inlier number is iteratively updated until it converges. The final homography matrix is ​​solved by the least squares method using all inlier sets, thereby determining the pose of the camera (3).

7. A portable intelligent fence alarm for a track area according to claim 1, characterized in that, In the process of generating dynamic trajectories with geographic coordinates, an improved SORT multi-target tracking method is employed, specifically including: For each active human target trajectory, a Kalman filter based on a linear constant velocity model is used, based on the state estimation vector from the previous time step. ,in The coordinates of the detection box center are, Aspect ratio, For height, Given the rates of change of the corresponding state variables, predict the prior state at the current moment. and covariance matrix ; Calculate the association cost matrix between all detection boxes and prediction boxes in the current frame. For any element in the matrix The calculation formula is as follows: ;in Indicates the first Detection box With the Prediction boxes The intersection and union ratio, Represents the appearance feature vectors of both The cosine distance between them These are weighting coefficients used to balance the contributions of geometric position and appearance features; Cost matrix The input is fed into the Hungarian algorithm to find the matching subset with the minimum global cost. For successfully matched trajectory-detection pairs, the state of the Kalman filter is updated using the position information of the detection boxes, and the posterior state estimate is calculated. and covariance For unmatched trajectories, they are considered temporarily lost and retained in the buffer; for unmatched detections, they are considered new targets and new trajectories are initialized.

8. The portable smart fence alarm for a track area according to claim 1, characterized in that the process of uploading the encrypted evidence chain specifically includes: At the moment of boundary violation determination, capture the video clip containing the boundary violation behavior and the corresponding three-dimensional coordinate trajectory data; The device uses its built-in hardware encryption chip to generate a SHA-256 hash value and performs RSA asymmetric encryption on the hash value. The encrypted hash value is then packaged with the original data and synchronized to the cloud management platform through blockchain network nodes.

9. A portable intelligent fence alarm for track areas according to claim 1, characterized in that the division of the multi-level risk areas adopts a dynamic determination logic based on time-to-collision (TTC) criteria: Based on the continuous trajectory coordinates of personnel in a 3D digital map, the real-time movement velocity vector of personnel is calculated using the finite difference method. and the direction angle of movement Simultaneously, obtain the normal vector of the preset electronic fence or laser projection line. and the danger zone; Calculate the vertical projection distance from the operator's current location to the electronic fence or laser projection line. And combine the projection component of the velocity vector in the direction of the normal vector. Calculate time to collision value ;in, when If the person is moving away from or parallel to the fence, a risk assessment is not triggered. As people move toward the fence And the calculated time to collision value Less than the first preset time threshold When personnel are deemed to have entered a Level 1 risk area, an early warning signal is triggered. When the personnel's position coordinates are detected to cross the boundary of the laser projection line, and the velocity vector points towards the danger side... Furthermore, if the location is in the danger zone, it is determined that the person has entered the level zero danger zone, and an emergency alarm signal is triggered at this time; For moving horizontally only near the fence Or move in the opposite direction The system automatically maintains a safe status for personnel, thereby eliminating false alarms caused by personnel passing by or working with their backs to the fence.

10. A control method for a portable intelligent fence alarm for a track area, applied to the portable intelligent fence alarm for a track area as described in claim 1, specifically characterized by comprising: S1. Using intelligent recognition hardware (10) to load a pre-stored three-dimensional digital map of the track area construction environment, obtain the coordinate position of the laser projection line projected by the laser emitter (4) in the map; at the same time, collect images of the construction area in real time through the camera (3), map the personnel positions identified in the images to the three-dimensional digital map, and generate a dynamic trajectory of personnel with geographical coordinates. S2. Based on the personnel's dynamic trajectory, calculate the real-time distance between the personnel and the preset electronic fence or laser projection line, and divide the risk areas into multiple levels according to the distance threshold; the multiple risk areas include at least a level one risk area and a level zero danger area; S3. When it is determined that a person has entered a Level 1 risk area, the intelligent identification hardware (10) outputs the first control command to control the voice speaker (5) to play a warning sound. S4. When the person is judged to have crossed the laser projection line into the level zero danger zone, the intelligent identification hardware (10) outputs the second control command, triggers the red flashing light (1) to flash and the voice speaker (5) to sound a high frequency alarm, and generates an encrypted evidence chain with timestamp and coordinates, and uploads it to the cloud management platform through the communication module.

Images