Bridge pile construction process monitoring and data collaborative collection method based on multi-vision

Abstract

The application provides a multi-view visual bridge pile foundation construction process monitoring and data collaborative collection method, which comprises the following steps: S1: multiple groups of image collection units are arranged along the circumference of a pile hole, global calibration is completed, and a unified world coordinate system is established; S2: the multiple groups of image collection units are synchronously triggered to synchronously collect construction scene images; S3: the synchronous images are preprocessed, the turbidity of mud in the hole is evaluated in real time, and whether the imaging condition is met is judged; S4: according to the current process identified, a corresponding parameter extraction algorithm is adaptively called to extract key quality and state parameters corresponding to the process from three-dimensional point cloud and image data; S5: the operation data of construction equipment and external sensors are connected, the visual monitoring data and the equipment sensing data are aligned with a unified time reference, the process identifier is bound to generate a structured construction data package; and S6: when the monitoring parameters exceed the preset threshold, a hierarchical early warning is triggered, and the scene data corresponding to the early warning is stored.

Description

Technical Field

[0002] This invention belongs to the technical field of bridge pile foundation construction quality monitoring and digital construction, specifically a method for monitoring bridge pile foundation construction procedures and collaborative data acquisition based on multi-view vision. Background Technology

[0004] With the expansion of bridge construction scale and the increasing demands for digitalization in construction, the requirements for quality control efficiency and data integrity in pile foundation construction are becoming increasingly stringent. In recent years, research has introduced technologies such as machine vision and ultrasonic testing into the field of pile foundation construction monitoring to assist in the quality inspection of individual processes; however, existing technical solutions still have significant limitations.

[0005] In the field of borehole quality inspection, Chinese invention patent CN115930855A proposes to use ultrasonic sensors to obtain cross-sectional images of pile holes and stitch them together to form three-dimensional images, which are then compared with standards to complete acceptance. However, this patent is limited to a single process after borehole formation, relies on ultrasonic imaging, is greatly affected by mud, and is a post-inspection method that cannot be monitored in real time.

[0006] In the field of visual monitoring, Chinese invention patent CN119062314A proposes to generate a detection set by acquiring images and ranging data through multiple modules. However, this is limited to post-drilling acceptance, relies on the combination of two-dimensional images and ranging, lacks three-dimensional reconstruction, and does not coordinate with multi-source data such as drilling rigs and grouting equipment.

[0007] While ultrasonic testing can overcome the interference of mud media on optical imaging to some extent, its equipment is complex, the operation cycle is long, and the cross-sectional scanning method results in low testing efficiency. It is only suitable for the post-drilling acceptance stage and cannot achieve continuous monitoring during the construction process. Contact logging solutions require inserting the probe into the borehole, which carries the risk of probe burial. Furthermore, the logging data is one-dimensional cross-sectional information and cannot reconstruct the three-dimensional spatial morphology of the borehole wall and the spatial pose of the construction components.

[0008] In addition, the visual data from existing monitoring schemes are independent of multi-source data such as drilling rigs, grouting equipment, and mud testing. They lack a unified time benchmark and process dimension for integration, making it difficult to form a complete and traceable digital archive of construction quality and unable to support in-process quality control and post-event traceability analysis.

[0009] In existing pile foundation construction monitoring technologies, visual monitoring solutions are mainly limited by factors such as poor lighting conditions in the deep cavity of the pile hole, turbid mud medium, and severe obstruction by construction components. The imaging quality is difficult to meet the requirements of three-dimensional reconstruction. Therefore, existing technologies mostly adopt non-visual solutions such as ultrasonic and contact logging.

[0010] Therefore, how to effectively solve the problems of numerous blind spots, reliance on manual switching of procedures, insufficient accuracy of three-dimensional measurement, and fragmentation of multi-source data in existing pile foundation construction monitoring technologies are urgent technical issues that need to be addressed.

[0011] It should be noted that the above technical information is intended only to enhance the understanding of the overall background technology of the present invention, and should not be regarded as an admission or in any way implying that the above technical information constitutes prior art known to those skilled in the art. Summary of the Invention

[0013] To address the shortcomings in the aforementioned background technology, this invention proposes a method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision. The technical problems to be solved are: existing pile foundation construction monitoring technologies suffer from numerous blind spots, reliance on manual switching of procedures, insufficient accuracy in three-dimensional measurement, and fragmentation of multi-source data.

[0014] The technical solution of this application is as follows:

[0015] A method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision includes the following steps:

[0016] S1: Multiple image acquisition units are deployed around the pile hole and global calibration is completed to establish a unified world coordinate system. The global calibration specifically includes: performing intrinsic parameter calibration on a single image acquisition device to obtain lens distortion parameters and intrinsic parameter matrices; using a planar calibration target to acquire multiple sets of synchronous pose images at different depths and orientations within a common field of view, and solving for the relative extrinsic parameters between multiple devices; performing global optimization through bundle adjustment to obtain the optimized extrinsic parameters of each device; after calibration, a unified world coordinate system is established with the pile hole design center as the origin and the vertical direction as the elevation axis; the image acquisition devices are deployed around the pile hole through a support structure, and the device pose is adjusted so that the fields of view of adjacent devices overlap, and the common field of view of all devices covers the depth range to be measured inside the pile hole and the working area at the hole opening; a supplementary lighting module is set up to adapt to nighttime and low-light scenes inside the hole.

[0017] S2: Controls multiple image acquisition units to simultaneously acquire construction scene images through a synchronous triggering mechanism;

[0018] S3: Preprocess the synchronous image, evaluate the mud turbidity in the borehole in real time and determine whether the imaging conditions are met; perform semantic recognition on the synchronous image to obtain the image semantic recognition result; when the imaging conditions are met, reconstruct the dense point cloud of the construction scene through a multi-view stereo vision algorithm, and extract the component shape and spatial position features from the dense point cloud as point cloud morphology features; input the image semantic recognition result, the available point cloud morphology features and the operating status signal from the construction equipment into the pile foundation construction process state machine with positive jump constraints, and automatically determine the current process in combination with the multi-frame confirmation mechanism;

[0019] S4: Based on the current process identified, the corresponding parameter extraction algorithm is adaptively invoked to extract the key quality and status parameters corresponding to the process from the image semantic recognition results, available point cloud morphological features, and the running status signal;

[0020] S5: Access the operating data of construction equipment and external sensors, align visual monitoring data and equipment sensor data with a unified time base, and generate structured construction data packages by binding process identifiers;

[0021] S6 triggers a tiered warning when the monitored parameters exceed the preset threshold, and retains the corresponding scenario data for the warning.

[0022] Furthermore, the image acquisition unit includes an image acquisition device whose pitch angle can be adjusted via an adjustable bracket. The adjustable bracket is connected to the ground or the protective cylinder via a bracket base. The image acquisition device is connected to a supplementary lighting module, and an optical target is provided on the inner wall of the protective cylinder.

[0023] Furthermore, in step S3, image preprocessing includes distortion correction, stereo correction, and image enhancement. Stereo correction keeps the corresponding epipolar lines of multiple images horizontally aligned and compresses the stereo matching search range. The three-dimensional point cloud reconstruction obtains a disparity map through stereo matching under epipolar constraints. After being converted into three-dimensional coordinate points, it undergoes multi-view registration and fusion, statistical filtering, and radius filtering for noise reduction to obtain a complete scene dense point cloud.

[0024] Furthermore, in step S3, the method for evaluating mud turbidity is as follows: an optical target with known contrast is fixed on the inner wall of the orifice casing, the target area is extracted from each frame of synchronized image and the gradient amplitude of the target edge is calculated, the gradient amplitude calibrated in clear water is used as the reference value, and when the measured gradient amplitude recovers to a preset proportion above the reference value, it is determined that the imaging conditions are met.

[0025] Further, in step S3, the process identification process is as follows: using a feature extraction model trained for the pile foundation construction scenario, the semantic information of construction objects such as drill bit, drill rod, rebar cage, grouting pipe, mud surface, and concrete surface are identified from the image, and the morphological and spatial location features of the corresponding objects are extracted from the three-dimensional point cloud; the identification results are input into the pile foundation construction process state machine, which is a finite state machine with preset standard process nodes and jump rules. The preset standard process nodes and forward jump rules conform to the construction specifications, allowing only jumps along the construction process sequence and prohibiting reverse jumps and jumps across processes; a multi-frame confirmation mechanism is set up, using multiple consecutive frames as a judgment window. When the process identification results of all frames in the window are consistent with the allowed jump processes of the current state machine, the process state switch is triggered.

[0026] Furthermore, in step S4, the parameter extraction methods corresponding to different processes are as follows:

[0027] Drilling process: By recognizing the drill string assembly and its movement status through image recognition, and combining the accessed drilling rig depth and drilling speed data, the stability of the drilling process is evaluated in real time;

[0028] Hole cleaning process: When the mud has settled and the imaging conditions are met, the hole wall point cloud is segmented and extracted from the 3D point cloud, the center and equivalent diameter of each section are fitted, the actual borehole axis is fitted by connecting the center of the sections, the verticality parameter is calculated, and the hole diameter variation curve is generated.

[0029] The process of lowering the reinforcing cage involves identifying the structural features or pre-marked features at the end of the reinforcing cage, calculating the three-dimensional spatial coordinates of the end of the reinforcing cage, and comparing them with the design coordinates to obtain the deviation of the cage top elevation and the deviation of the plane position.

[0030] The process of lowering the infusion conduit involves identifying the top and bottom positions of the conduit, calculating the lowering depth and bottom plane deviation, and verifying whether the distance between the bottom of the conduit and the bottom of the hole meets the construction requirements.

[0031] Concrete pouring process: Combining image texture and color features with point cloud elevation change features, supplemented by prior information on the location of the grout outlet of the duct and the pouring time, spatial constraints are applied to the interface candidate region, the concrete liquid surface and the lower end position of the pouring duct are segmented and identified, the concrete surface elevation and duct burial depth parameters are calculated, and the concrete filling coefficient is estimated by combining the pouring flow data.

[0032] Furthermore, in the hole cleaning process, the hole wall point cloud segmentation adopts a region growth strategy under the prior constraint of cylindrical shape: the radius search range is set with the borehole center axis as a reference, and points that meet the design radius of the pile hole are selected from the point cloud near the hole opening as seed points. The growth criteria are the angle between the point cloud normal and the horizontal plane and the distance deviation from the point cloud to the axis. The hole wall point cloud is extracted layer by layer in the depth direction. According to the hole diameter change curve, abnormal shapes such as reduced diameter, expanded diameter, and irregular hole shape are automatically identified, and the hole depth coordinates are calibrated in combination with the drilling rig depth data. In the steel cage lowering process, the coaxiality of the upper and lower steel cage sections is monitored simultaneously. In the concrete pouring process, the mud surface and the concrete surface are distinguished by the geometric shape constraint of the approximately horizontal liquid surface.

[0033] Furthermore, in step S5, a globally unified time reference is established through a network time protocol or a precise time protocol to synchronize and calibrate the clocks of all access devices; and the visual monitoring data and device sensing data with different sampling frequencies are aligned to a unified time axis by using nearest neighbor timestamp matching combined with linear interpolation.

[0034] Furthermore, in step S5, the structured construction data package is indexed by pile number and timestamp, and includes three types of data fields: process type, visual monitoring parameters, and equipment operating parameters. The process identifier enables the classification and integration of all construction data according to the process dimension. The data fusion process simultaneously performs multi-source data cross-validation, and the validation rules are adaptively matched with the current process type. When the deviation between the visual monitoring parameters and the equipment sensing parameters exceeds the preset calibration threshold, a calibration prompt is triggered.

[0035] Furthermore, in step S6, the process status and monitoring parameters are visualized in real time. The displayed content includes a 3D point cloud model, parameter change curves, process status indicators, and deviation dashboards. The early warning adopts a hierarchical triggering mechanism, setting multiple threshold levels based on relevant construction specifications and project quality control objectives. When the deviation of the monitoring parameter reaches the early warning threshold, an early warning level prompt is triggered; when it exceeds the alarm threshold, an alarm level prompt is triggered. When an early warning is triggered, the image, point cloud snapshot, and corresponding parameter data at the time of the early warning are automatically recorded to form a traceable early warning record.

[0036] The specific beneficial effects of this invention include:

[0037] 1. The surround multi-view vision deployment scheme solves the problem of severe occlusion and monitoring blind spots in the deep cavity scenario of pile holes by overlapping and covering the field of view of multiple devices. It can completely acquire the three-dimensional morphology of the hole wall, the spatial pose of the construction components and the working status of the hole opening, realize three-dimensional monitoring of the entire process from hole formation to grouting, and improve the accuracy and adaptability of visual measurement in the complex deep cavity scenario.

[0038] 2. Construct a joint recognition mechanism that combines image semantics, point cloud morphology, and process state machine. With the forward flow of construction technology as the logical constraint and a multi-frame confirmation mechanism, it can effectively eliminate misidentification caused by construction interference such as temporary hoisting and personnel obstruction. It can automatically complete the judgment of the entire process and adaptively switch the parameter extraction logic without the need for manual switching of monitoring modes, thus realizing continuous automated monitoring of the entire process.

[0039] 3. By using a unified time benchmark, visual monitoring data and multi-source sensor data can be accurately aligned. At the same time, process identifiers are bound to the fused data, which solves the problems of fragmentation, single dimension and difficulty in traceability of traditional construction data. It forms a structured data chain integrating time, process and parameter, which can be directly used for construction process playback, quality traceability and digital archive archiving.

[0040] 4. Continuous real-time monitoring of key quality parameters for each process is conducted, and a graded early warning mechanism is set up. When quality risks such as excessive verticality, abnormal hole diameter, rebar cage misalignment, or non-compliance of duct burial depth occur, an early warning can be triggered in a timely manner and on-site data can be retained to assist on-site personnel in handling the situation and control quality risks during the construction process.

[0041] 5. During the mud settling stage after hole cleaning, the turbidity of the medium inside the hole will decrease significantly. At this time, by combining adaptive supplemental lighting and image enhancement technology, clear images of the hole wall can be obtained and three-dimensional reconstruction can be completed. Attached Figure Description

[0043] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the overall process of the monitoring method described in this invention.

[0045] Figure 2 This is a schematic diagram of the logical framework for process identification and parameter extraction in this invention;

[0046] Figure 3 This is a schematic diagram of the information flow for data acquisition and collaborative fusion in this invention;

[0047] Figure 4 This is a schematic diagram of the on-site layout in a specific embodiment of the present invention.

[0048] Explanation of icon numbers:

[0049] 1-Image acquisition device; 2-Fill light module; 3-Adjustable bracket; 4-Bracket base; 5-Optical target; 6-Pile hole. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the core concept of the present invention and the following embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] These embodiments are provided to make the application thorough and complete, and to fully express the scope of the application to those skilled in the art. It should be noted that, unless otherwise specifically stated, the relative arrangement of components and steps, material composition, numerical expressions, and values ​​illustrated in these embodiments should be interpreted as merely exemplary and not as limiting.

[0053] A method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision includes the following steps:

[0054] Step S1: Deployment and Global Calibration of Multi-View Vision System

[0055] S11. On-site deployment: Multiple sets of adjustable brackets 3 are fixed on the hardened ground or protective casing and other stable structures around the pile hole 6 to be monitored. Each set of adjustable brackets 3 is fixed to the ground or protective casing through the bracket base 4. Image acquisition equipment 1 and supplementary lighting module 2 are installed on the brackets to form a multi-view image acquisition unit.

[0056] Preferably, three or more adjustable supports 3 can be set up and evenly distributed around the circumference of the pile hole 6. Adjust the height of the adjustable supports 3 and the pitch angle of the image acquisition device 1 so that the optical axis of each device faces the inside of the pile hole 6; based on the designed diameter D of the pile hole and the estimated hole depth... Select the horizontal field of view α and vertical field of view β of the image acquisition device, such that the device installation height h and pitch angle θ satisfy:

[0057]

[0058] in, The maximum depth to be scanned; the angle between the optical axes of adjacent devices in the horizontal direction is less than α / 2, ensuring that the overlap rate of adjacent images at the bottom of the hole is not less than 30%;

[0059] The common field of view of all equipment covers the depth range to be measured and the working area at the borehole opening, and the overlapping area of ​​adjacent fields of view is continuous and uninterrupted throughout the entire depth range of the pile hole; the adjustable bracket 3 adopts a rigid fixing method to reduce the impact of construction vibration on imaging stability.

[0060] S12. Single device intrinsic parameter calibration: The intrinsic parameters of a single image acquisition device 1 are calibrated using a planar calibration target. The intrinsic parameter matrix and distortion coefficient of the lens are obtained by solving the problem, eliminating pixel distortion caused by lens optical distortion, and providing an accurate intrinsic parameter basis for subsequent 3D reconstruction.

[0061] S13. Global Extrinsic Parameter Calibration: The planar calibration target is placed within the common field of view of multiple devices, and synchronous images at different depths and orientations are acquired. The relative extrinsic parameters between each image acquisition device 1 are solved using a multi-camera global calibration algorithm. As an optimization, in addition to acquiring target images at the borehole opening, the calibration target can be lowered to different depths within the pile hole 6 via suspension, acquiring synchronous images at multiple depths to participate in the calibration calculation, ensuring consistent calibration accuracy across the entire depth range within the pile hole 6. During the calibration calculation, bundle adjustment is used to globally optimize all calibration parameters, with reprojection error as the optimization objective. Simultaneously, a rigid constraint on the target plane is introduced to reduce cumulative calibration error and improve the stability of the extrinsic parameter calibration.

[0062] S14. Coordinate system establishment: With the design center of pile hole 6 as the coordinate origin, the vertical upward axis as the elevation axis, and the horizontal axis along the longitudinal and transverse directions of the bridge as the plane coordinate axis, a unified world coordinate system is established to provide a unified coordinate reference for subsequent three-dimensional reconstruction and parameter calculation.

[0063] Step S2: Synchronous Image Acquisition

[0064] S21. Synchronous Trigger Configuration: A synchronous trigger controller is used to connect all image acquisition devices 1, set the acquisition frame rate, and control all devices to expose synchronously, ensuring that each set of images corresponds to the construction scene at the same moment, and avoiding the impact of timing deviation on the accuracy of 3D reconstruction.

[0065] S22, Illumination Adaptive Adjustment: The supplementary lighting module 2 automatically adjusts the output brightness according to the ambient light intensity. When there is insufficient light at night or in the hole, supplementary lighting is turned on. Visible light or infrared supplementary lighting can be used to ensure that the acquired image features are clear and the contrast is moderate.

[0066] S23. Image Transmission: The acquired multi-channel synchronous images are transmitted to the edge computing terminal via a wired network for subsequent processing.

[0067] Step S3: 3D Scene Reconstruction and Automatic Process Recognition

[0068] S31. Image Preprocessing: The edge computing terminal sequentially performs distortion correction, stereo correction, and image enhancement on the received multi-channel synchronous images. Lens distortion is corrected based on intrinsic parameters, and stereo correction is performed based on extrinsic parameters to align the corresponding epipolar lines horizontally, compressing the search range for stereo matching. Local image contrast is enhanced by limiting the adaptive histogram equalization algorithm, improving details in dark areas under uneven lighting within the aperture, and providing texture gradient support for subsequent stereo matching.

[0069] S32. Online Turbidity Assessment: An optical target 5 with known contrast is fixed to the inner wall of the orifice casing. The target area is extracted from each frame of synchronized image, and the gradient amplitude of the target edge is calculated. A gradient reference value calibrated in clear water before orifice cleaning is set. When the measured gradient amplitude recovers to more than 80% of the reference value, it is determined that the mud turbidity in the orifice meets the imaging conditions required for three-dimensional reconstruction, allowing the subsequent multi-view stereo vision reconstruction process to be triggered; otherwise, only two-dimensional image semantic analysis is performed, and dense point clouds are not generated.

[0070] S33. Multi-view 3D reconstruction: Under epipolar constraints, a semi-global stereo matching algorithm is used to calculate the disparity map. Census transformation is used as the matching cost, and sub-pixel level disparity is obtained by multi-path cost aggregation. After converting the disparity into 3D coordinate points, it is uniformly transformed to the world coordinate system based on extrinsic parameters. Outlier noise is removed by statistical filtering and radius filtering to obtain the scene dense point cloud.

[0071] S34. Intelligent Process Recognition: The edge computing terminal acquires basic operational status signals in real time, such as drilling rig lifting / lowering, mud pump start / stop, and grouting equipment working status, through a multi-source data fusion gateway. When a dense point cloud is available, the geometric dimensions and spatial location of the construction object are extracted from the point cloud. Regardless of whether a dense point cloud is available, the trained feature extraction model identifies semantic information such as drill bit, drill rod, rebar cage, grouting pipe, and various liquid surfaces from the image. The above semantic information, point cloud morphological features, and equipment status signals are input into the pile foundation construction process state machine, and the current construction process is determined by combining the construction process logic.

[0072] The standard process nodes preset in the pile foundation construction process state machine include drilling, primary hole cleaning, lowering of the reinforcing cage, extension of the reinforcing cage, secondary hole cleaning, lowering of the guide pipe, concrete pouring, and completion of pouring. The jump between nodes strictly follows the construction process sequence and only allows jumps in the forward direction of the construction process. Reverse jumps and jumps across processes are prohibited.

[0073] The state switching must meet two conditions simultaneously: first, the current frame must identify a typical combination of features of the target process; second, the identification results of multiple consecutive frames must remain consistent, thereby eliminating misjudgments caused by interference such as temporary hoisting or personnel obstruction during construction.

[0074] The feature extraction model is based on an instance segmentation network architecture and is trained on an annotated image dataset containing construction scenarios such as drilling, hole cleaning, rebar cage lowering, duct lowering, and concrete pouring. The training data is labeled using instance-level masks of construction objects as annotation units, with category labels including drill bit, drill rod, rebar cage segment, duct, mud surface, and concrete surface. During annotation, a polygon-dense annotation strategy is used for the boundaries of the surface categories to ensure that the boundary features between the surface and the borehole wall are fully represented in the training data. The model training employs a transfer learning strategy, initializing with pre-trained weights from a general object detection dataset and fine-tuning all parameters on a pile foundation construction scenario dataset.

[0075] For example, the process will only proceed to the lowering of the rebar cage if a complete sequence of rebar cage segment hoisting, docking, and lowering is continuously detected and the current state is in the hole cleaning completion stage.

[0076] Step S4 Adaptive Parameter Extraction

[0077] S41. Drilling process parameter extraction: The position and movement of the drill bit and drill rod are identified by multi-view image recognition. Combined with the drilling depth, drilling pressure and drilling speed data accessed by the edge computing terminal, the drilling speed and borehole stability are evaluated in real time. No three-dimensional point cloud reconstruction and geometric parameter measurement of the borehole wall are performed at this stage.

[0078] S42. Hole Cleaning Process Parameter Extraction: When the process state machine determines that the hole cleaning process has begun and the online turbidity assessment determines that the imaging conditions are met, a full-depth scan of the hole wall is initiated. The hole wall point cloud is segmented and extracted from the overall point cloud. Horizontal sections are cut at equal elevation intervals along the depth direction. The center of each section is fitted with the equivalent diameter. The actual borehole axis is fitted and the verticality is calculated to generate a hole diameter variation curve.

[0079] The borehole wall point cloud segmentation is achieved through a point cloud semantic segmentation model or a region growing algorithm based on geometric features. Specifically, a region growing strategy under the prior constraint of cylindrical morphology is adopted: taking the borehole center axis as a reference, a radius search range is set, and points that meet the design radius of the pile hole are selected from the point cloud near the borehole opening as seed points. The growth criteria are the angle between the point cloud normal and the horizontal plane and the distance deviation of the point cloud from the axis. The borehole wall point cloud is extracted layer by layer in the depth direction. During the growth process, the point cloud in each depth layer is fitted with a cross-sectional circle. When the fitting residual exceeds a set threshold, the automatic growth of that depth segment is paused and marked as an area to be manually reviewed, thus balancing the degree of automation and data reliability.

[0080] When fitting the borehole centerline, piecewise linear fitting or overall spatial linear fitting is used to calculate the overall verticality and piecewise verticality of the entire hole, respectively. The equivalent diameters of each section are sorted by depth to generate a full-depth borehole diameter variation curve, automatically identifying abnormal hole shapes such as reduced diameter and enlarged diameter. For irregular hole wall sections, the equivalent diameter, maximum diameter, and minimum diameter are calculated simultaneously to identify defects such as cloverleaf holes. During parameter extraction, borehole depth coordinates are calibrated using drilling rig depth data to improve the consistency between visual measurements and equipment sensor data.

[0081] S43. Extraction of parameters for the lowering of the reinforcing cage: Identify inherent structural features such as the reinforcing ring and the end of the main reinforcement bars at the top of the reinforcing cage, or pre-set marker features, locate the spatial coordinates of these features in the 3D point cloud, compare them with the designed cage top coordinates, and calculate the cage top elevation deviation and planar offset; During the extension of the reinforcing cage, simultaneously monitor the coaxiality of the connection between the upper and lower sections of the reinforcing cage, providing quantitative data support for the quality control of the coaxiality of the reinforcing cage connection.

[0082] S44. Extraction of parameters for the lowering of the conduit: Identify the positional characteristics of the top and bottom ends of the grouting conduit, calculate the lowering depth and the offset of the bottom plane, verify whether the suspended distance of the bottom of the conduit meets the construction process requirements, and avoid the conduit being too close or too far from the bottom of the hole, which would affect the grouting quality.

[0083] S45. Concrete pouring process parameter extraction: Combining image color and texture features with point cloud elevation change features, the position of the slurry outlet of the guide pipe is used as the spatial center constraint, the cumulative volume of pouring is used as the time constraint, and the geometric shape constraint of the approximately horizontal liquid surface is used to distinguish the mud liquid surface and the concrete liquid surface, and the concrete surface elevation is calculated; the position of the lower end of the pouring guide pipe is identified to obtain the bottom elevation of the guide pipe, and the difference is the burial depth of the guide pipe; the real-time concrete surface elevation is cross-compared with the theoretical elevation calculated by integrating the pouring flow rate, and the filling coefficient calculation is calibrated.

[0084] Because the mud inside the borehole is turbid, it is difficult to accurately distinguish the interface between concrete and mud by relying solely on the color features of the image. This step can combine prior information such as the position of the guide pipe, the duration of grouting, and the cumulative grouting volume to constrain the interface position and smooth the recognition results, thereby further improving the accuracy and stability of interface recognition under complex working conditions.

[0085] Step S5: Multi-source data collaborative fusion

[0086] S51. Multi-source data access: Access the operating data of the construction equipment control system and various external sensors through the multi-source data fusion gateway, including drilling pressure, drilling speed and depth data of the drilling rig, flow rate and pressure data of the grouting equipment, as well as mud indexes and environmental parameters.

[0087] S52. Unified Time Base: A globally unified time base is established through a precise time protocol or a network time protocol. Image acquisition device 1 and construction equipment controller synchronize exposure and mark timestamps via hardware triggering; for third-party sensors without hardware synchronization interfaces, the multi-source data fusion gateway uses the timestamp in the data packet to perform nearest neighbor matching and linear interpolation based on the arrival time, aligning the multi-source data to a unified time axis. The time alignment error is controlled within the order of hundreds of milliseconds, meeting the real-time requirements of construction monitoring.

[0088] S53. Data Alignment and Tag Binding: Nearest neighbor matching combined with linear interpolation is used to align visual monitoring data and equipment sensor data from different sampling frequencies onto a unified timeline. Each set of fused data is bound to the current process identifier, generating a structured construction data package indexed by pile location number and ordered by timestamp, achieving the classification and integration of all construction data according to process dimensions. Data can be stored locally on the edge computing terminal or uploaded to a remote data management system.

[0089] S54. Multi-source data cross-validation: The depth parameters obtained from visual monitoring are compared with the depth parameters output by the encoder of the construction equipment. A calibration prompt is triggered when the deviation between the elevation of the bottom of the duct monitored by visual monitoring and the travel data of the hoisting equipment exceeds 0.1m; a calibration prompt is also triggered when the deviation between the elevation of the cage top monitored by visual monitoring and the hoisting travel data exceeds 30mm, or when the deviation between the elevation of the concrete surface monitored by visual monitoring and the elevation calculated from the integral of the pouring flow rate exceeds 0.2m, reminding on-site personnel to check the equipment status. The system can automatically reduce the weight of low-quality data in the fusion and verification process based on data reliability rating, ensuring the stability of data fusion.

[0090] Step S6: Real-time display and over-limit warning

[0091] S61. Multi-dimensional visualization display: Through on-site display and early warning terminals and mobile devices, the system displays real-time 3D point cloud models, current process status, real-time values ​​and change curves of various parameters, and parameter deviation dashboards, allowing managers to intuitively grasp the construction status.

[0092] S62. Adaptive Threshold Setting: Set graded thresholds according to the construction specifications implemented in the project. For example, a verticality deviation of 80% of the allowable value in the specification can be set as a warning threshold, and 100% can be set as an alarm threshold; a hole diameter deviation exceeding the design value by ±5% triggers a warning, and exceeding ±8% triggers an alarm; a rebar cage plane deviation exceeding 20mm triggers a warning, and exceeding 40mm triggers an alarm; a conduit burial depth of less than 2m or more than 6m triggers a warning, and less than 1m or more than 8m triggers an alarm; thresholds are dynamically loaded through configuration files and support on-site adjustment.

[0093] S63. Tiered early warning trigger: When the deviation of the monitored parameter reaches the early warning threshold, an early warning prompt is triggered; when the deviation exceeds the alarm threshold, an alarm prompt is triggered. The early warning adopts a combination of on-site audio-visual prompts and remote message push. The early warning information is simultaneously output to the on-site display early warning terminal, and the corresponding handling suggestions for the corresponding process are pushed simultaneously.

[0094] S64. Early Warning Data Retention: When an early warning is triggered, the image, point cloud snapshot, and corresponding parameter data at the time of the early warning are automatically recorded to form a traceable early warning record, which is stored locally or uploaded to a remote data management system for easy subsequent analysis and archiving.

[0095] System architecture description for implementing the above methods

[0096] To implement the aforementioned method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, the following layered system architecture can be adopted. Logically, the system is divided into a field edge and a remote management end, as shown in the overall architecture below. Figure 4 As shown.

[0097] The on-site edge terminal corresponds to the on-site processing logic of steps S1 to S6 of the execution method. The core hardware includes a multi-view image acquisition unit (including image acquisition device 1, supplementary lighting module 2, adjustable bracket 3, and bracket base 4), a synchronous trigger controller, an edge computing terminal, a multi-source data fusion gateway, and an on-site display and early warning terminal, which is responsible for image acquisition and access, real-time data preprocessing, 3D reconstruction, process identification and parameter extraction, and on-site early warning output. The remote management terminal corresponds to data storage, summary analysis, and file management functions. The core is the remote data management system, which undertakes the uploading and subsequent application of structured construction data packages.

[0098] The hardware can be divided into: a perception layer (image acquisition device 1, supplementary lighting module 2, and various external sensors), a network and edge computing layer (multi-source data fusion gateway and edge computing terminal), an execution and display layer (on-site display and early warning terminal and audible and visual early warning device), and a remote service layer (cloud server and data storage device corresponding to the remote data management system). Each software functional module interacts through a unified data bus and interface to achieve data flow and functional collaboration.

[0099] All aspects not detailed in this invention are conventional technical means known to those skilled in the art.

[0100] The above content shows and describes the basic principles, main features, and beneficial effects of the present invention. The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or 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 method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, characterized in that: Includes the following steps: S1: Deploy multiple image acquisition units around the pile hole (6) and complete global calibration to establish a unified world coordinate system; S2: Controls multiple image acquisition units to simultaneously acquire construction scene images through a synchronous triggering mechanism; S3: Preprocess the synchronous image, evaluate the mud turbidity in the borehole in real time and determine whether the imaging conditions are met; perform semantic recognition on the synchronous image to obtain the image semantic recognition result; when the imaging conditions are met, reconstruct the dense point cloud of the construction scene through a multi-view stereo vision algorithm, and extract the component shape and spatial position features from the dense point cloud as point cloud morphology features; input the image semantic recognition result, the available point cloud morphology features and the operating status signal from the construction equipment into the pile foundation construction process state machine with positive jump constraints, and automatically determine the current process in combination with the multi-frame confirmation mechanism; S4: Based on the current process identified, the corresponding parameter extraction algorithm is adaptively invoked to extract the key quality and status parameters corresponding to the process from the image semantic recognition results, available point cloud morphological features, and the running status signal; S5: Access the operating data of construction equipment and external sensors, align visual monitoring data and equipment sensor data with a unified time base, and generate structured construction data packages by binding process identifiers; S6 triggers a tiered warning when the monitored parameters exceed the preset threshold, and retains the corresponding scenario data for the warning.

2. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision as described in claim 1, characterized in that: The image acquisition unit includes an image acquisition device (1) whose pitch angle can be adjusted by an adjustable bracket (3). The adjustable bracket (3) is connected to the ground or the protective cylinder through a bracket base (4). The image acquisition device (1) is connected to a supplementary light module (2), and an optical target (5) is provided on the inner wall of the protective cylinder.

3. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S3, image preprocessing includes distortion correction, stereo correction, and image enhancement. Stereo correction keeps the corresponding epipolar lines of multiple images horizontally aligned and compresses the stereo matching search range. The three-dimensional point cloud reconstruction obtains a disparity map through stereo matching under epipolar constraints. After being converted into three-dimensional coordinate points, it undergoes multi-view registration and fusion, statistical filtering, and radius filtering for noise reduction to obtain a complete scene dense point cloud.

4. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S3, the mud turbidity is evaluated as follows: an optical target with known contrast is fixed on the inner wall of the orifice casing, the target area is extracted from each frame of synchronized image and the gradient amplitude of the target edge is calculated, the gradient amplitude calibrated in clear water is used as the reference value, and when the measured gradient amplitude recovers to a preset proportion above the reference value, the imaging conditions are determined to be met.

5. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S3, the process identification process is as follows: using a feature extraction model trained for pile foundation construction scenarios, the semantic information of construction objects such as drill bits, drill rods, rebar cages, grouting pipes, mud surfaces, and concrete surfaces is identified from the image. At the same time, the morphological and spatial location features of the corresponding objects are extracted from the 3D point cloud. The identification results are input into the pile foundation construction process state machine. The process state machine is a finite state machine with preset standard process nodes and jump rules. Its preset standard process nodes and forward jump rules conform to the construction specifications. Jumping is only allowed along the construction process sequence, and reverse jumps and cross-process jumps are prohibited. A multi-frame confirmation mechanism is set up, with multiple consecutive frames as the judgment window. When the process identification results of all frames in the window are consistent with the allowed jump processes of the current state machine, the process state switch is triggered.

6. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S4, the parameter extraction methods for different processes are as follows: Drilling process: By recognizing the drill string assembly and its movement status through image recognition, and combining the accessed drilling rig depth and drilling speed data, the stability of the drilling process is evaluated in real time; Hole cleaning process: When the mud has settled and the imaging conditions are met, the hole wall point cloud is segmented and extracted from the 3D point cloud, the center and equivalent diameter of each section are fitted, the actual borehole axis is fitted by connecting the center of the sections, the verticality parameter is calculated, and the hole diameter variation curve is generated. The process of lowering the reinforcing cage involves identifying the structural features or pre-marked features at the end of the reinforcing cage, calculating the three-dimensional spatial coordinates of the end of the reinforcing cage, and comparing them with the design coordinates to obtain the deviation of the cage top elevation and the deviation of the plane position. The process of lowering the infusion conduit involves identifying the top and bottom positions of the conduit, calculating the lowering depth and bottom plane deviation, and verifying whether the distance between the bottom of the conduit and the bottom of the hole meets the construction requirements. Concrete pouring process: Combining image texture and color features with point cloud elevation change features, supplemented by prior information on the location of the grout outlet of the duct and the pouring time, spatial constraints are applied to the interface candidate region, the concrete liquid surface and the lower end position of the pouring duct are segmented and identified, the concrete surface elevation and duct burial depth parameters are calculated, and the concrete filling coefficient is estimated by combining the pouring flow data.

7. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision as described in claim 6, characterized in that: In the hole cleaning process, the hole wall point cloud segmentation adopts a region growth strategy under the prior constraint of cylindrical shape: the radius search range is set with the borehole center axis as a reference, and points that meet the design radius of the pile hole are selected from the point cloud near the hole opening as seed points. The growth criteria are the angle between the point cloud normal and the horizontal plane and the distance deviation from the point cloud to the axis. The hole wall point cloud is extracted layer by layer in the depth direction. According to the hole diameter change curve, abnormal shapes such as reduced diameter, expanded diameter, and irregular hole shape are automatically identified, and the hole depth coordinates are calibrated in combination with the drilling rig depth data. In the steel cage lowering process, the coaxiality of the upper and lower steel cage sections is monitored simultaneously. In the concrete pouring process, the mud surface and the concrete surface are distinguished by the geometric shape constraint of the approximately horizontal liquid surface.

8. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S5, a globally unified time reference is established through a network time protocol or a precise time protocol, and the clocks of all access devices are synchronized and calibrated. By using nearest neighbor timestamp matching combined with linear interpolation, visual monitoring data and device sensing data with different sampling frequencies are aligned to a unified time axis.

9. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S5, the structured construction data package is indexed by pile number and timestamp, and includes three types of data fields: process type, visual monitoring parameters, and equipment operating parameters. The process identifier enables the classification and integration of all construction data according to the process dimension. The data fusion process simultaneously performs multi-source data cross-validation, and the validation rules are adaptively matched with the current process type. When the deviation between the visual monitoring parameters and the equipment sensing parameters exceeds the preset calibration threshold, a calibration prompt is triggered.

10. The method for monitoring and collaboratively acquiring data on bridge pile foundation construction procedures based on multi-view vision, as described in claim 1 or 2, is characterized in that: In step S6, the process status and monitoring parameters are visualized in real time. The display includes a 3D point cloud model, parameter change curves, process status indicators and deviation dashboards. The early warning adopts a hierarchical triggering mechanism. Multiple thresholds are set according to the corresponding construction specifications and project quality control objectives. When the deviation of the monitoring parameter reaches the early warning threshold, an early warning prompt is triggered. When the deviation exceeds the alarm threshold, an alarm prompt is triggered. When an alert is triggered, the system automatically records the image, point cloud snapshot, and corresponding parameter data at the time of the alert, forming a traceable alert record.

Images