Arch bridge cable hoisting dynamic measurement and early warning method based on machine vision

Abstract

The invention discloses an arch bridge cable hoisting dynamic measurement and early warning method based on machine vision, and relates to the technical field of cable hoisting construction, and the arch bridge cable hoisting dynamic measurement and early warning method comprises a data acquisition module, an identification module, a measurement module and an early warning module. The invention aims to optimize the safety and efficiency in the hoisting process through accurate image processing and real-time data monitoring, and provides a non-contact and high-precision dynamic measurement and early warning method through fusion of machine vision and a deep learning technology, so that efficient space positioning of a hoisting point and a hoisted object in the cable hoisting construction process is realized, and the construction efficiency is improved. The method has the advantages of being rapid, accurate and stable, efficiency, operation and convenience in the arch bridge cable hoisting process are remarkably improved, and construction quality and safety can be further guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge construction, and in particular to a dynamic measurement and early warning method for cable hoisting of arch bridges based on machine vision, which is applicable to real-time monitoring, safety early warning and operation optimization in the cable hoisting construction of long-span arch bridges. Background Technique

[0002] With its superior adaptability to complex terrains, large lifting capacity, fast and efficient transportation efficiency, and excellent safety, the cable hoisting technology has become a key material transportation and construction auxiliary equipment in mountain bridge construction. However, in traditional cable hoisting construction, which relies on manual measurement and empirical judgment, there are mainly problems such as low measurement accuracy: it is difficult for manual use of total stations or laser rangefinders to continuously track dynamic targets (such as suspended objects, trolleys), and it is easily affected by the environment, resulting in many safety hazards; it is difficult to monitor the safety distance between construction workers and suspended objects in real time, and it is difficult to give timely warnings for speeding and over-limit operations, resulting in limited efficiency; manual data collection and processing takes a long time and cannot support real-time decision-making for high-frequency hoisting operations. Although the sensor-based monitoring systems in the prior art can partially solve the above problems, they have defects such as complex equipment layout, high cost, and being easily interfered by mechanical vibrations. These deficiencies make the positioning and monitoring means of traditional cable hoisting systems difficult to meet the requirements of high-precision positioning, safety guarantee, and automated operation in modern complex construction environments.

[0003] Therefore, in order to improve the construction efficiency and safety of cable hoisting, an intelligent cable hoisting positioning and regulation method is urgently needed. Conducting real-time monitoring during the cable hoisting construction process based on machine vision technology is a feasible solution. With the rapid development of target detection technology, even in small target monitoring tasks, modern algorithms can achieve high accuracy and reaction speed, and their real-time performance is sufficient to meet the requirements of construction monitoring. This method can replace traditional human eye monitoring, reduce the subjective errors caused by visual estimation, and significantly improve the positioning accuracy and stability.

[0004] The invention patent with the publication number CN118968414A discloses a monitoring method for the cable - car of an arch - bridge cable - hoisting based on visual geometry, which is a patent previously applied by the applicant of this application. The patent solution can accurately monitor the position of the crane, but its focus is only on positioning and data acquisition, and it fails to consider how to timely detect potential safety risks and respond. Due to the lack of real - time warning function, when a safety problem occurs, the system cannot immediately issue a warning, which may delay the handling of potential safety hazards. In addition, the solution involves multiple complex technical steps, such as camera calibration, distortion correction, image transformation, and coordinate system conversion, etc. These steps require complex mathematical calculations and computer vision algorithm processing, with relatively high technical requirements for on - site operators. In practical applications, the deployment of the system may require a long time and high technical support, and the operation difficulty of each step is relatively large, resulting in a high deployment cost and implementation difficulty.

[0005] Therefore, the present invention proposes a simplified non - contact, high - precision dynamic measurement and warning method by integrating machine vision and deep - learning technologies to solve the problems of inaccurate positioning, low monitoring efficiency, high potential safety hazards, and insufficient automation level in traditional technologies, significantly improving the overall efficiency and safety of cable - hoisting construction. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention provides a dynamic measurement and warning method for arch - bridge cable - hoisting based on machine vision. By introducing machine vision technology and a multi - module collaborative working mechanism, it has the ability to quickly and non - destructively perform real - time, accurate positioning, monitoring, and regulation of the suspension points and the state of the suspended object during cable - hoisting construction. This method integrates a data acquisition module, an identification module, a measurement module, and a warning module, and can efficiently complete functions such as target detection, environmental perception, and risk warning during the hoisting process, thus solving the problems of insufficient positioning accuracy, low monitoring efficiency, poor environmental adaptability, and insufficient construction safety guarantee existing in the prior art.

[0007] To achieve the above - mentioned purpose, the technical solution adopted by the present invention is:

[0008] A dynamic measurement and warning method for arch - bridge cable - hoisting based on machine vision, characterized in that it includes:

[0009] S1. Data acquisition module: Collect images of the hoisting area through a camera, perform distortion correction and orthographic projection transformation, generate scale parameters, and simultaneously collect environmental data (wind speed, wind direction);

[0010] S2. Identification module: Detect suspension points, suspended objects, and personnel based on deep - learning algorithms, and achieve cross - frame target tracking through target tracking algorithms, and output trajectory data;

[0011] S3. Measurement module: Map pixel coordinates to the world coordinate system in combination with a scale, and calculate speed, travel distance, and cumulative mileage in real time;

[0012] S4. Warning module: Dynamically compare measurement data with safety thresholds and trigger warning signals.

[0013] Preferably, the data acquisition module includes the following specific steps:

[0014] S1.1 Image acquisition unit: Used to set up a camera and adjust the field of view to ensure coverage of the pylon, cable, and the moving range of the carriage, and obtain the original image sequence {I};

[0015] S1.2 Orthorectification unit: Convert the tilted image I to an orthographic projection image I' through a homography matrix to eliminate perspective distortion;

[0016] S1.3 Scale calculation unit: Calculate the horizontal and vertical scale parameters S x , S y ;

[0017] S1.4 Environmental data acquisition unit: Collect real-time wind speed, wind direction, and weather data through a meteorological sensor.

[0018] Preferably, the orthorectification unit performs the following steps:

[0019] S1.2.1 Determine the world coordinate system O w-xyz : Establish the world coordinate system with the intersection point of the vertical plane where the traction cable is located and the inner side of the pylon facade;

[0020] S1.2.2 Feature point calibration and data preparation: Preset 4 high-contrast marking points on the pylon surface as reference feature points, and their planar coordinates in the world coordinate system are (i is the serial number of the feature point: i = 1, 2, 3, 4), and obtain their corresponding pixel coordinates through the image acquisition device (i is the serial number of the feature point: i = 1, 2, 3, 4);

[0021] S1.2.3 Distortion correction: Use the pre-calibrated camera internal parameter matrix K and distortion parameters dist, and correct the original image through the cv2.undistort function of OpenCV to generate the intermediate image I undistort ;

[0022] S1.2.4 Homography matrix calculation: Based on the world coordinates {P wi} and pixel coordinates {P pi} of the feature points, call the cv2.findHomography function of OpenCV, and solve the homography matrix H by minimizing the projection error through the least squares method, satisfying:

[0023]

[0024] The homography matrix H is a 3×3 projective transformation matrix used to describe the mapping relationship from an inclined plane to an orthographic plane.

[0025] S1.2.5 Orthographic image generation: For the intermediate image I undistort Apply the homography matrix H and perform a perspective transformation through the cv2.warpPerspective function in OpenCV to generate an orthographic projection image I', and its mathematical description is:

[0026] I' = warpPerspective(I undistort , H, (W, H))

[0027] Where: W and H are the width and height of the input image, and by default, they are the same as the size of the input image.

[0028] Preferably, the scale calculation unit performs the following steps:

[0029] S1.3.1 Feature point selection: Select two diagonal feature points P 1 (u 1 , v 1 ) and P 3 (u 3 , v 3 ) in the orthographic image I', and their corresponding plane coordinates in the world coordinate system are P w1 (x 1 , y 1 , 0), P w3 (x 3 , y 3 , 0).

[0030] S1.3.2 Adjust the image coordinate system O p-uv , unify the image coordinate system to and the world coordinate system O w-xyz , and adjust the image coordinate system to a coordinate system O p‘-uv with the same direction as the world coordinate system. The adjustment method is as follows:

[0031]

[0032] Where: (u k , v k ), (u' k , v' k ) respectively correspond to the image coordinates of the origin of the world coordinates before and after the transformation, and (u 0 , v 0 ) is the origin of the default image coordinate system;

[0033] S1.3.3 Calculate the image scales in the horizontal and vertical directions of the image, and their scales are calculated according to the following formulas respectively:

[0034] Horizontal scale: Vertical scale:

[0035] Where: (X, Y) are the physical coordinates of points i and j in the actual physical coordinate system, and (u’, v’) are the image coordinates of points i and j in the adjusted image coordinate system.

[0036] Preferably, the recognition module includes:

[0037] S2.1 Target detection unit: It is used to detect and recognize objects in the image on the orthorectified image, and adopts a deep learning-based target detection algorithm to perform multi-class target detection and positioning on the lifting points (D i ), lifted objects (W i ), and construction workers (R i ) in the orthorectified image (I’), and output the target category and pixel-level bounding box coordinates. Set the non-maximum suppression threshold iou = 0.45, filter out the low-quality detection boxes with confidence conf < 0.7, retain the high-confidence targets, extract the pixel coordinates (u, v) of the target center point as the positioning reference, and generate the detection result set Det = {class, (u, v), confidence}, where (u, v) are the pixel coordinates of the target center point, class is the target category, and confidence is the confidence of the detected target.);

[0038] S2.2 Target tracking unit: The target tracking unit is used to solve the problems of missing target identity association and discontinuous motion state in single-frame image detection, and realizes stable tracking of cross-frame target trajectories by fusing temporal information;

[0039] S2.3 Recognition and analysis unit: It is used to receive the wind speed magnitude, the end coordinates of the installed segment, and the spatial coordinates of the lifted object output by the data acquisition unit and the target detection unit, and dynamically calculate and set the equipment operation safety parameters based on the input data.

[0040] Preferably, the target tracking unit performs the following operations:

[0041] (4) Cross-frame identity association: Based on the target position information output by the single-frame detection unit, bind the detection boxes of the same target in consecutive video frames to reduce identity jumps caused by occlusion and lighting changes;

[0042] (5) Motion state smoothing: Dynamically predict the target displacement speed and acceleration through a multi-target motion model, and perform weighted fusion with the detection results to output the smoothed target coordinate sequence. The historical trajectory of each target is Traj(ID) = {(u 1 , v 1 , t 1 ), (u 2 , v 2 , t 2 ),..., (u n , v n , t n )};

[0043] (6) Trajectory integrity maintenance: Perform short-term trajectory backtracking and state recovery on lost targets to avoid trajectory breaks caused by short-term occlusion, and at the same time eliminate invalid targets that exceed the preset number of lost connection frames (such as 30 frames).

[0044] Preferably, the recognition and analysis unit performs the following operations:

[0045] Dynamically update the safety threshold based on the existing current hoisting progress and set the safety threshold:

[0046] (1) Operating stroke threshold L max : The maximum translational distance of the initial hoisting point on the hoisting reference plane. The calculation of the operating stroke L max is determined by the following formula:

[0047] L max,n = X installed,n-1 - X ori - ΔX safe

[0048] where: X installed，n-1 is the coordinate of the end of the (n - 1)th installed segment (dynamically updated), and ΔX safe is the lateral safety distance (the value should be determined considering the shape characteristics of the hoisted object)

[0049] (2) Operating speed threshold V th : Set in gears according to the operating conditions, satisfying:

[0050]

[0051] where: Q 额定 : The rated load of the equipment, depending on the design load-bearing capacity of the equipment.

[0052] (3) Safety radius threshold S safe , and the construction rule of its safety area:

[0053] Personnel safety S person : With the position of the construction personnel as the center of the circle, Sperson = 3m to construct a safety restricted area for personnel;

[0054] Safe radius S of the suspended object load : With the geometric center of the projection of the outer contour of the suspended object as the center, dynamically calculate the safe radius based on the type of suspended object: S load = max{3m, 0.5D envelope} where D envelope is the diameter of the minimum circumscribed circle of the projection of the outer contour of the suspended object on the frontal plane;

[0055] Safe radius S of the lifting point hook : With the projection point of the hook center as the center, S hook = 1.5m to restrict personnel from approaching the dangerous operation point;

[0056] Preferably, the measurement module includes:

[0057] Position measurement unit in S3.1: Map the target pixel coordinates to the world coordinate system through scales S x , S y to complete coordinate positioning by converting the pixel coordinates of the object to the world coordinate system;

[0058] State measurement unit in S3.2: The state measurement unit is used to calculate the motion state parameters of the lifting point and the suspended object in real time based on the object coordinates obtained in S3.1 and feedback the parameters to the operator interface;

[0059] Preferably, the position measurement unit performs the following calculations:

[0060] Coordinate increment calculation in S3.1.1, obtain the pixel coordinate increment of the recognized object relative to the origin of the world coordinate system, the coordinate increment in the horizontal direction is Δu = u - u' 0 , and in the vertical direction is Δv = v - v' 0 where, (u' 0 , u' 0 ) is the pixel coordinates of the origin of the world coordinate system in the image,

[0061] World coordinate mapping in S3.1.2, perform linear transformation calculation based on the scale parameter to obtain the world coordinate of the recognized object relative to the origin of the coordinate system, and calculate in the following way:

[0062]

[0063] Thus, the pixel positioning coordinates of the recognized object are transformed from the image coordinate system O p ′ to the world coordinate system O w , and the world coordinate (X ob j, Y obj , 0) of the recognized object is obtained.

[0064] S3.1.3 Position calculation: It is used to provide real-time feedback on the horizontal displacement of the lifted object from its initial position during a single operation. The travel L of the hook and the load is calculated through coordinates. i The operating travel is the horizontal movement span L i = X obj _-X ori where (X ori , Y obj ) are the coordinates of the identified object at the initial lifting point, and (X base , Y base ) are the coordinates of the lifting base surface.

[0065] Preferably, the state measurement unit performs the following calculations:

[0066] (3) Space-time state solution module:

[0067] 3) Based on the trajectory data Tra j (ID) = {(x 1 , y 1 , t 1 ), (x 2， y 2 , t 2 ),..., (x n , y n , t n )}, extract the coordinates (x t1 , y t1 , t 1 ) and (x t2 , y t2 , t 2 ) of two adjacent frames at a frequency of 1 frame / second, and calculate the speed where v x is the horizontal speed, representing the direction and rate of horizontal movement, v y is the vertical movement speed, representing the lifting rate of the lifted object, and its resultant speed is used for overspeed alarm

[0068] 4) Cumulative mileage: Based on the sum of the absolute values of the displacements in the horizontal direction (X-axis), it represents the total travel of the lifted object:

[0069]

[0070] (4) State feedback module

[0071] Horizontal speed v x : A positive value indicates forward movement, and a negative value indicates backward movement.

[0072] Vertical speed v y : A positive value indicates upward movement, and a negative value indicates downward movement.

[0073] Single - trip L i : It shows the horizontal displacement of the suspended object from the initial position and is used to determine whether the target point has been reached.

[0074] Preferably, the warning module includes:

[0075] S4.1 Safety rule library unit: Set the safety thresholds of the system, and pre - define the motion over - limit threshold (L max ), speed threshold V th , wind speed threshold w safe and safety radius S safe ;

[0076] S4.2 Risk analysis unit: Real - time monitor the displacement L i , speed v x , v y , the distance d of the personnel and wind speed w wind , and trigger corresponding warning information according to the trigger rules;

[0077] S4.3 Warning output unit: Generate a hierarchical warning signal according to the trigger rules and output reminder information through the man - machine interface or the sound and light device.

[0078] Preferably, the setting method of the safety threshold is:

[0079] (1) Motion over - limit threshold:

[0080] Maximum travel L max : It is determined based on the span between the starting point X start and the ending point X end of the hoisting path, and a certain safety distance ΔX safe， is reserved to satisfy L max，n = X installed，n-1 - X start - ΔX safe

[0081] (2) Speed over - limit threshold: V th

[0082] (3) Wind speed threshold:

[0083] Continuous safe wind speed w safe = 5.5m / s (level 4 wind);

[0084] Instantaneous gust wind speed threshold w gust，th = 1.5wsafe;

[0085] (4) Safety radius: S safe

[0086] Personnel restricted area: S person = 3m;

[0087] Hoisting Area: S load = max{3m, 0.5D envelope}

[0088] Lifting Point Restricted Area: S hook = 1.5

[0089] Preferably, the triggering rules for the risk analysis are as follows:

[0090] Rule 1: If L i > L max , trigger the horizontal displacement over - limit alarm;

[0091] Rule 2: If the distance d between personnel / lifting object < S safe , trigger the safety area intrusion alarm;

[0092] Rule 3: If ∣v x ∣> v x，max or ∣v y ∣> vy,max, trigger the speed over - limit alarm;

[0093] Rule 4: If w wind > the preset continuous wind speed threshold and the duration is over - limit, trigger the continuous wind speed alarm;

[0094] Rule 5: If w gust > the preset gust threshold, trigger the instantaneous gust alarm.

[0095] Preferably, the warning output unit includes a hierarchical warning mechanism:

[0096] First - level warning: Generate a prompt signal and feedback parameter anomalies through the human - machine interface;

[0097] Second - level warning: Generate a signal to reduce the equipment speed and limit the hoisting speed;

[0098] Emergency braking: Generate a shutdown signal and trigger the safety locking mechanism.

[0099] Preferably, the corresponding relationship between the hierarchical warning mechanism of the warning output unit and the triggering rules is as follows:

[0100] First - level warning (prompt signal): Triggered by Rule 1 (horizontal displacement over - limit), highlight the abnormal parameters on the human - machine interface and issue a low - frequency buzzer prompt;

[0101] Second - level warning (equipment speed reduction): Triggered by Rule 2 (safety area intrusion), Rule 3 (speed over - limit), limit the towing speed of the trolley to the preset safety value (such as 50% of v x，max ), and activate the audible and visual alarm device;

[0102] Emergency braking: Triggered by Rule 4 (continuous wind speed exceeding limit) and Rule 5 (instantaneous gust exceeding limit), immediately cut off the power of the traction cable, activate the mechanical brake device, broadcast the emergency evacuation instructions through the public address system, and suspend the hoisting operation.

[0103] Compared with the prior art, the present invention provides a dynamic measurement and early warning method for the cable hoisting of arch bridges based on machine vision, having the following beneficial effects:

[0104] 1. High-precision non-contact dynamic measurement. Through orthorectification and scale calibration techniques, eliminate image distortion and perspective error, and significantly improve the dynamic measurement accuracy of the position and height of the lifted object; based on the object detection and tracking algorithm, achieve continuous and stable tracking of the lifted object and construction personnel in complex environments (such as occlusion and illumination changes), reducing the risk of missed detection and misjudgment in manual monitoring.

[0105] 2. Multi-dimensional risk early warning and active protection. Integrate visual positioning, environmental sensing and motion parameters, support multi-rule joint triggering and hierarchical early warning mechanisms (prompting / speed reduction / emergency braking), and achieve closed-loop control from risk identification to active intervention; through the setting of dynamic safety thresholds (such as travel, speed, safety radius), effectively reduce the accident risks such as personnel straying into restricted areas and lifted object collisions. A customizable safety rule library is added, which can be flexibly adjusted according to the needs of specific projects. The setting of the safety rule library is more flexible and easy to manage, further improving the applicability and practicality of the system, and does not require professional technicians to perform cumbersome adjustments, making the system easier to operate and maintain.

[0106] 3. Optimization of construction efficiency and economic benefits. Adopt a non-contact machine vision solution, streamline camera calibration and image preprocessing, reduce the complex layout requirements of traditional sensor networks, and reduce equipment deployment and maintenance costs; fully automated monitoring throughout the process reduces manual dependence and supports real-time data feedback and operation process optimization. Description of the Drawings

[0107] Figure 1 : System architecture diagram of the present application.

[0108] Figure 2 : Layout diagram of feature points in Embodiment 1 of the present application.

[0109] Figure 3 : Coordinate positioning diagram in Embodiment 1 of the present application.

[0110] Figure 4 : Safety early warning diagram in Embodiment 2 of the present application. Detailed Implementation Modes

[0111] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0112] The present invention proposes a dynamic measurement and early warning method for the cable hoisting of arch bridges based on machine vision. By using machine vision means to replace the human eye to construct a machine vision measurement system, the target detection algorithm and scale correction are used to accurately detect the positions of the hoisting points and the hoisted objects during the cable hoisting process and calculate the real-world coordinates at this time. Further calculate the operating state of the hoisting according to the measurement and positioning results, and combine the early warning module to realize the risk early warning and regulation during the hoisting process. This method can quickly and efficiently realize the real-time and accurate positioning and regulation of the cable hoisting construction process, so that the construction personnel can intuitively describe the specific positions of the hoisting points and the hoisted objects at this time, and ensure the safe construction according to the early warning prompts. Using this method can greatly improve the construction efficiency and safety, and the monitoring results are intuitive and accurate.

[0113] Embodiment 1

[0114] Before the camera device works, the Zhang-Zhengyou calibration method is used to calibrate the camera to obtain internal parameter information such as its internal parameter matrix K and distortion parameters D. Before the cable hoist trolley is implemented, the camera device is fixed and erected, and its focal length and related parameters are adjusted to make the picture clear and stable. Adjust the camera's field of view range so that the camera's shooting range can cover the entire working plane of the cable system. The left and right sides of the picture should include two pylons, and the entire cable and the longitudinal maximum movement range of the trolley should be covered in the up and down directions. Start the camera to collect video, and perform distortion correction on the original images collected by the camera. After the cable hoist trolley device runs, continue to use the camera device to shoot the hoisting area. Figure 2 Four high-reflectivity marking points are preset on the pylon elevation as feature points, and the origin of the world coordinate system is set at the lower left marking point, and its world coordinate P The corresponding default pixel coordinates are Use the findHomography function of OpenCV to calculate the homography matrix H, and perform perspective transformation on the collected original frame images to convert the originally obliquely shot images into orthographic images. On the orthographic hoisting image, select feature points P1 and P3 to calculate the scale of the image after perspective transformation, where P 1 The world coordinates and default pixel coordinates of are and P 4 The world coordinates and pixel coordinates are and The scale of the image after perspective transformation is S x = 0.224315 m / pixel, S y = 0.225122 m / pixel. In the recognition model, the YOLOv5s architecture is adopted for the object detection model. The training set contains 2,000 labeled images (three types of targets: hook, suspended object, and personnel). The mAP@0.5 of the test set reaches 85.3%. Its tracking algorithm is based on DeepSORT multi-object tracking. The Kalman filter parameters are Q = 0.01 (process noise) and R = 0.1 (observation noise). The maximum number of lost frames is 30 frames. The object detection unit receives the image in the hoisting area transmitted by the data acquisition unit. For example Figure 3 , the image contains two types: the hoisting point and the suspended object. The image pixel coordinates corresponding to hoisting point 1, hoisting point 2, and the hoisting segment are respectively and Then, the coordinates of the hoisting point and the suspended object relative to the world coordinate origin are calculated through the scale as and The pixel coordinates of the suspended object corresponding to the initial hoisting point are Then, it is converted to the world coordinate relative to the world coordinate origin as Then, the current travel of the suspended object L = 105.74 - 122.19 = -16.45 m (direction to the left), thus completing the spatial positioning of the recognition object.

[0115] Example 2

[0116] The current ambient wind speed W wind = 1.7 m / s, which is less than the threshold of level 5 wind (W safe = 5.5 m / s), and the wind speed warning is not triggered. The current hoisting segment is restricted by the already installed segment 3 on the left, and its horizontal travel is limited. Take the hoisting end point at the top of this installation node. The corresponding pixel coordinates of the suspended object are Converted to the world coordinate relative to the world coordinate origin as The top coordinate X of the already installed segment 3 installed,3 = 94.86 m (dynamically updated). The reference coordinate X of the initial hoisting point ori = 122.19 m. The horizontal length L of segment 4 segment = 20 m. The lateral safety distance is taken as 1 m. Then, the horizontal travel threshold L max，4 = X installed，3 - X ori - ΔX safe = 94.86 + 1 - 115.86 + 20 / 2 = -15 (the negative sign indicates the limiting direction of moving to the left), Figure 4 For the middle segment 4, the horizontal travel |L| < |L max,5|, the travel limit warning is not triggered. When the boundary of the hoisted segment approaches the travel limit area, the abnormal parameters are highlighted on the human-machine interface, and a low-frequency beeping prompt is issued.

[0117] In summary, the method proposed by the present invention can efficiently and real-time monitor the states of the lifting points and the lifted objects. By using the perspective transformation method to convert the image into an orthographic image, the pixel scale error in the image is effectively corrected. The target detection algorithm is used to accurately locate the target objects in the image, and the center point of the target is used as the coordinate representation point, so as to directly calculate its movement. Compared with the traditional manual visual estimation method and the method of arranging sensors, this method significantly reduces the use and maintenance costs of expensive equipment, and at the same time avoids human subjective errors. Its feedback in the lifting state monitoring is faster and more efficient, which not only improves the efficiency of the construction process, but also significantly improves the safety. In addition, even in the scenario of oblique photography, the method of the present invention can still achieve high-precision positioning and measurement, with wide applicability and reliability.

[0118] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0119] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic measurement and early warning method for arch bridge cable hoisting based on machine vision, characterized by: include; S1. Data acquisition module: collects images of the hoisting area through the camera, performs distortion correction and orthographic projection transformation, generates scale parameters, and simultaneously collects environmental data including wind speed, wind direction and weather; S2. Recognition module: detects hanging points, hanging objects and personnel based on deep learning algorithms, realizes cross-frame target tracking through target tracking algorithms, and outputs trajectory data; S3. Measurement module: maps pixel coordinates to the world coordinate system in combination with the scale, and calculates speed, travel distance and accumulated mileage in real time; S4. Early warning module: dynamically compares measurement data with safety thresholds and triggers early warning signals.

2. According to claim 1, a machine vision-based dynamic measurement and early warning method for arch bridge cable hoisting, characterized in that: The data acquisition module includes the following specific steps: S1.1 Image acquisition unit: used to set up the camera and adjust the field of view to ensure that the tower, cable and sports car range are covered, and obtain the original image sequence {I}; S1.2 Orthorectification unit: convert the oblique image I into the orthophoto projection image I' through the homography matrix to eliminate the perspective distortion; S1.3 Scale calculation unit: Calculate the horizontal and vertical scale parameters S based on feature point calibration x , S y ; S1.4 Environmental data collection unit: collects real-time wind speed, wind direction and weather data through meteorological sensors.

3. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 2 is characterized in that: The orthorectification unit performs the following steps: S1.2.1 Determine the world coordinate system O w-xyz : Establish the world coordinate system with the intersection of the vertical plane where the traction cable is located and the inner side of the facade of the cable tower; S1.2.2 Feature point calibration and data preparation: Four high-contrast markers are preset on the tower surface as reference feature points, and their plane coordinates in the world coordinate system are And obtain the corresponding pixel coordinates through the image acquisition device Where i is the feature point number: i = 1, 2, 3, 4; S1.2.3 Distortion correction: Use the pre-calibrated camera intrinsic parameter matrix K and distortion parameter dist to correct the original image through OpenCV's cv2.undistort function to generate the intermediate image I undistort ; S1.2.4 Homography matrix calculation: Based on the world coordinates of the feature points {P wi } and pixel coordinates {P pi }, call OpenCV's cv2.findHomography function, and solve the homography matrix H by minimizing the projection error using the least squares method, satisfying: The homography matrix H is a 3×3 projection transformation matrix, which is used to describe the mapping relationship from the inclined plane to the orthographic plane; S1.2.5 Orthophoto generation: for intermediate image I undistort Apply the homography matrix H and perform perspective transformation through OpenCV's cv2.warpPerspective function to generate an orthographic projection image I', which is mathematically described as: I'=warpPerspective(I undistort ,H,(W,H)) Where: W, H are the width and height of the input image, which are consistent with the input image size by default.

4. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 3 is characterized in that: The scale calculation unit performs the following steps: S1.3.1 Feature point selection: Select two diagonal feature points P1 (u1, v1) and P3 (u3, v3) in the orthophoto image I', and their corresponding world coordinate plane coordinates are P w1 (x1, y1, 0), P w3 (x3, y3, 0); S1.3.2 Adjusting the image coordinate system O p-uv , unify the image coordinate system and the world coordinate system O w-xyz , adjust the image coordinate system to be consistent with the world coordinate system direction O p‘-uv , the adjustment method is as follows: Where: (u k , v k )、(u' k ,v' k ) respectively correspond to the image coordinates of the world coordinate origin before and after the transformation, and (u0, v0) is the origin of the default coordinate system of the image; S1.3.3 Calculate the image scale in the horizontal and vertical directions of the image. The scales are calculated according to the following formulas: Horizontal scale: Vertical scale: Where: (X, Y) is the physical coordinates of point i and point j in the actual physical coordinate system, and (u', v') is the image coordinates of point i and point j in the adjusted image coordinate system.

5. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 4 is characterized in that: The identification module comprises: S2.1 Target detection unit: used to detect and identify objects in the orthophoto image, using a deep learning-based target detection algorithm to detect the hanging point (D i )、Hanging objects(W i ) and construction personnel (R i ) performs multi-category target detection and positioning, outputs the target category and pixel-level bounding box coordinates; sets the non-maximum threshold iou=0.45, filters the low-quality detection frames with confidence conf<0.7, retains high-confidence targets, extracts the target center pixel coordinates (u, v) as the positioning reference, and generates a detection result set Det={class, (u, v), confidence}, where (u, v) is the target center pixel coordinates, class is the target category, and confidence is the confidence of the detected target; S2.2 Target tracking unit: The target tracking unit is used to solve the problem of missing target identity association and discontinuous motion state in single-frame image detection, and realizes stable tracking of cross-frame target trajectory by fusing time series information; S2.3 Identification and analysis unit: used to receive the wind speed, the end coordinates of the installed segment, and the spatial coordinates of the suspended object output by the data acquisition unit and the target detection unit, and dynamically calculate and set the equipment operation safety parameters based on the input data.

6. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 5 is characterized in that: The target tracking unit performs the following operations: (1) Cross-frame identity association: Based on the target position information output by the single-frame detection unit, the detection frames of the same target in consecutive video frames are bound to each other to reduce identity jumps caused by occlusion and lighting changes. (2) Motion state smoothing: The target displacement velocity and acceleration are dynamically predicted through the multi-target motion model, and weighted fused with the detection results to output a smoothed target coordinate sequence. The historical trajectory of each target is Traj(ID) = {(u1, v1, t1, (u 2， v2, t2), ..., (u n , v n , t n )}; (3) Trajectory integrity maintenance: Perform short trajectory backtracking and state recovery on lost targets to avoid trajectory breakage caused by short-term occlusion, and eliminate invalid targets that exceed the preset number of lost frames.

7. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 6 is characterized in that: The identification and analysis unit performs the following operations: dynamically updates the safety threshold based on the current hoisting progress, and sets the safety threshold: (1) Operation stroke threshold L max : Maximum translation distance of the initial lifting point of the lifting reference surface, operating stroke L max The calculation of is determined by the following formula: L max，n =X installed，n-1 -X ori -ΔX safe ; Where: X installed，n-1 is the dynamically updated end coordinate of the installed (n-1)th segment, ΔX safe The horizontal safety distance is determined by considering the shape characteristics of the hoisted object; (2) Operating speed threshold V th : The gear positions are set according to the working conditions to meet the following requirements: Where: Q 额定 : The rated load of the equipment depends on the design load-bearing capacity of the equipment. (3) Safety radius threshold S safe , the construction rules of its safe area are: Personnel safety person : With the construction worker's position as the center, S person =3m, to build a safety restricted area for personnel; Safety radius of hanging objects S load : Take the geometric center of the projection of the outer contour of the hoisted object as the center of the circle, and dynamically calculate the safety radius based on the type of hoisted object: S load =max{3m, 0.5D envelope }, where D envelope It is the minimum circumscribed circle diameter of the projection of the outer contour of the hanging object on the rotation plane; Safety radius of lifting point S hook : With the projection point of the hook center as the center of the circle, S hook =1.5m, restricting personnel from approaching dangerous work points.

8. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 7 is characterized in that: The measuring module comprises: S3.1 Position measurement unit: The target pixel coordinates are measured by the scale S x , S y Map to the world coordinate system, convert the pixel coordinates of the object into the world coordinate system to complete coordinate positioning; S3.2 State measurement unit: The state measurement unit is used to calculate the motion state parameters of the hanging point and the hanging object in real time according to the object coordinates obtained in S3.1, and feed the parameters back to the operator interface; the position measurement unit performs the following calculations: S3.1.1 Coordinate increment calculation, obtain the pixel coordinate increment of the identified object relative to the world coordinate origin, the horizontal coordinate increment is Δu=u-u'0, and the vertical coordinate increment is Δu=v-v'0, where (u'0, u'0) is the pixel coordinate of the origin of the world coordinate system in the image; S3.1.2 World coordinate mapping: Based on the scale parameter, linear transformation is performed to calculate the world coordinates of the identified object relative to the origin of the coordinate system. The calculation is performed in the following way: Thus, the pixel location coordinates of the recognized object can be obtained from the image coordinate system O p 'Transform to world coordinate system O w , get the world coordinates of the identified object (X obj , Y obj , 0); S3.1.3 Positioning calculation: used to provide real-time feedback on the horizontal displacement of the object from the initial position during a single operation, and to calculate the travel L of the hook and load through coordinates. i , the running stroke is the horizontal moving span L i =X obj _-X ori , where (X ori , Y obj ) is the coordinate of the initial lifting point, (X base , Y base ) is the coordinate of the lifting foundation surface; The state measurement unit performs the following calculations: (1) Space-time state calculation module: 1) Based on the trajectory data Tra output by the target tracking unit j (ID) = {(x1, y1, t1), (x 2， y2, t2), ..., (x n ,y n , t n )}, extract the coordinates of two adjacent frames at a frequency of 1 frame / second (x t1 ,y t1 , t1) and (x t2 ,y t2 , t2), calculation speed where v x is the horizontal velocity, which represents the speed in the horizontal moving direction, v y is the vertical moving speed, which represents the lifting rate of the hanging object. Used for overspeed alarm; 2) Cumulative mileage: Based on the sum of the absolute values ​​of horizontal (X-axis) displacement, it represents the total travel of the hoisted object: (2) Status feedback module: Horizontal speed v x : Positive value means forward movement, negative value means backward movement; Vertical speed v y : Positive value means rising, negative value means falling; Single trip L i : Displays the horizontal displacement of the hanging object from the initial position, which is used to determine whether it has reached the target point.

9. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 8, characterized in that: The early warning module comprises: S4.1 Safety rule base unit: Set the safety threshold of the system and predefine the motion limit threshold (L max ), speed threshold V th , wind speed threshold w safe And safety radius S safe ; S4.2 Risk analysis unit: real-time monitoring of displacement L i , speed v x 、v y , personnel distance d and wind speed w wind , trigger corresponding warning information according to the triggering rules; S4.3 Warning output unit: Generates graded warning signals according to triggering rules and outputs warning information through human-computer interaction interface or sound and light device; The safety threshold is set as follows: (1) Excessive exercise threshold: Maximum stroke L max : Based on the starting point X of the lifting path start With the end point X end The span is determined and a certain safety distance ΔX is reserved safe， Meet L max，n =X installed，n-1 -X start -ΔX safe ; (2) Speed ​​overlimit threshold: V th ; (3) Wind speed threshold: Continuous safe wind speed w safe =5.5m / s (level 4 wind); Instantaneous gust wind speed threshold w gust，th =1.5wsafe; (4) Safety radius: S safe ; Personnel restricted area: S person =3m; Hanging restricted area: S load =max{3m, 0.5D envelope }; Hanging point restricted area: S hook =1.5; The triggering rules of the risk analysis are: Rule 1: If L i >L max , triggering the horizontal displacement over-limit alarm; Rule 2: If the distance between person and suspended object is d safe , triggering a safe area intrusion alarm;​ Rule 3: If |v x ∣>v x，max or |v y |>vy,max, triggers speed over limit alarm; Rule 4: If w wind > If the preset continuous wind speed threshold exceeds the duration limit, the continuous wind speed alarm will be triggered; Rule 5: If w gust >Preset gust threshold to trigger instantaneous gust alarm.

10. The method for dynamic measurement and early warning of arch bridge cable hoisting based on machine vision according to claim 9, characterized in that: The warning output unit includes a hierarchical warning mechanism: Level 1 warning: Generates a warning signal and feedbacks parameter abnormalities through the human-computer interaction interface; Second level warning: Generates equipment deceleration signal to limit hoisting speed; Emergency brake: generates a stop signal and triggers the safety locking mechanism; The corresponding relationship between the hierarchical warning mechanism and the triggering rules of the warning output unit is as follows: Level 1 warning signal: Rule 1 is triggered by horizontal displacement exceeding the limit, abnormal parameters are highlighted on the human-machine interface, and a low-frequency buzzer is emitted to warn; Secondary warning device speed reduction: triggered by rule 2 safety zone intrusion and rule 3 speed limit exceeding, the sports car towing speed is limited to the preset safety value and the sound and light alarm device is activated; Emergency braking: triggered by continuous wind speed exceeding the limit in Rule 4 or instantaneous gust exceeding the limit in Rule 5, the traction rope power is immediately cut off, the mechanical brake device is started, the emergency evacuation order is broadcast through the broadcasting system, and the lifting operation is suspended.

Images