A high-precision positioning system for underground construction sites

Abstract

The application relates to the technical field of underground engineering construction, and discloses a high-precision positioning system for an underground concealed engineering construction site. The system comprises a positioning signal receiving unit, an AR scene construction unit, a virtual superposition unit, a path planning unit and a dynamic calibration unit. The positioning signal receiving unit receives multi-frequency signals and processes position characteristics; the AR scene construction unit generates a dynamic spatial distribution diagram and realizes visualization; the virtual superposition unit realizes pixel-level fusion of virtual and real scenes; the path planning unit plans an obstacle avoidance path for a construction machine; and the dynamic calibration unit corrects fusion errors to guarantee positioning precision. The system effectively solves the positioning problem of underground concealed engineering construction, improves positioning precision, the degree of construction visualization and construction safety, and improves construction efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of underground engineering construction, and in particular to a high-precision positioning system for underground concealed engineering construction sites. Background Art

[0002] With the rapid advancement of urban modernization, underground concealed projects are playing an increasingly important role in urban infrastructure development. Projects such as subways, underground integrated pipeline corridors, and tunnels are constantly emerging. These projects operate in extremely complex environments, with underground spaces fraught with uncertainties. This places extremely high demands on positioning accuracy, safety, and efficiency during construction. However, existing positioning technologies face numerous challenges in addressing the challenges of underground concealed projects.

[0003] Traditional satellite positioning systems, such as GPS, rely on receiving satellite signals to determine location. However, in underground environments, thick layers of soil, rocks, and buildings strongly block and attenuate satellite signals, making it difficult for satellite signals to be transmitted stably and accurately to underground construction areas. Even in relatively shallow underground spaces, signals can suffer severe errors due to multiple reflections and refractions, resulting in positioning accuracy typically only reaching the meter level or even worse, which is simply unable to meet the demand for high-precision positioning of equipment and personnel in underground concealed projects. For example, during the construction of underground pipe corridors, various pipelines must be precisely laid. Positioning errors at the meter level can lead to problems such as incorrect pipeline connections and conflicts with other facilities, not only increasing construction costs but also affecting the quality and subsequent use of the entire project.

[0004] Positioning technologies based on ground-based base stations, such as mobile communication base station positioning, also face difficulties in underground environments. Complex underground structures and media, such as metal pipes and moist soil, absorb and scatter base station signals, causing signal propagation delays and distortion. This not only significantly reduces positioning accuracy but also limits signal coverage, making it prone to positioning blind spots. During large-scale underground construction, signal strength varies significantly across different areas, making it impossible to obtain effective positioning information in some areas, severely restricting the positioning reliability of construction personnel and equipment.

[0005] At construction sites, workers lack intuitive and accurate means of identifying the locations of hidden underground equipment and pipelines. Due to a lack of effective positioning and visualization technology, accidents involving accidental digging and contact with underground pipelines are common. These accidents not only damage underground pipelines, leading to serious consequences such as gas leaks, power outages, and communication failures, but also cause construction delays, resulting in significant economic losses and even threatening the lives of construction workers. For example, during underground construction in older urban areas, due to the complex layout of underground pipelines and a lack of accurate data, accidental digging is common, significantly impacting the lives of surrounding residents and the normal operation of the city.

[0006] Furthermore, existing methods for obstacle avoidance path planning for construction machinery are mostly simplistic, relying primarily on limited sensor information and basic algorithmic rules. In the complex and ever-changing underground construction environment, faced with dynamically emerging obstacles and intricate spatial layouts, these methods struggle to quickly and accurately plan appropriate obstacle avoidance paths. During operation, construction machinery is prone to collisions due to its inability to avoid obstacles in a timely manner. This not only damages the machinery and increases repair costs, but can also cause construction work to stall, severely impacting efficiency. Summary of the Invention

[0007] The purpose of the present invention is to provide a high-precision positioning system for underground concealed engineering construction sites to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a high-precision positioning system for underground concealed engineering construction sites, the system comprising:

[0009] A positioning signal receiving unit, configured to receive preset geographic coordinate data of underground concealed equipment and real-time multi-band positioning signals from the construction site, wherein the real-time multi-band positioning signals include ultra-wideband signals, lidar point cloud signals, and inertial navigation data;

[0010] An AR scene construction unit is used to perform three-dimensional spatial modeling on the real-time multi-band positioning signal through a synchronous positioning and mapping algorithm to generate a dynamic spatial distribution map of underground concealed equipment;

[0011] A virtual overlay unit is used to align the dynamic spatial distribution map of the underground hidden equipment with the preset geographic coordinate data in time and space, and to construct a pixel-level fusion mapping relationship between the virtual equipment outline and the real scene based on a deep learning model;

[0012] A path planning unit, configured to generate an obstacle avoidance path planning matrix for the construction machinery through a reinforcement learning algorithm based on a fusion mapping relationship between the virtual equipment profile and the real scene;

[0013] A dynamic calibration unit is configured to perform iterative correction on fusion errors of the ultra-wideband signal, the lidar point cloud signal and the inertial navigation data based on a Kalman filtering algorithm, and output calibrated construction positioning coordinates.

[0014] The execution steps of the AR scene construction unit include:

[0015] extracting a set of ground feature points in the lidar point cloud signal to construct an initial three-dimensional point cloud framework;

[0016] fusing phase difference data of the ultra-wideband signal and acceleration vectors of the inertial navigation data to generate dynamic pose correction parameters;

[0017] According to the dynamic pose correction parameters, the initial three-dimensional point cloud framework is optimized in pose to generate a dynamic spatial distribution map of the underground concealed equipment.

[0018] Preferably, the execution steps of the virtual superposition unit include:

[0019] The underground concealed equipment dynamic spatial distribution map is converted into a multi-scale feature tensor, and a semantic segmentation mask of the preset geographic coordinate data is extracted through a convolutional neural network;

[0020] Based on the attention mechanism, pixel correlation weights of the multi-scale feature tensor and the semantic segmentation mask are calculated to generate a virtual contour fusion coefficient;

[0021] According to the virtual contour fusion coefficient, the virtual device contour is rendered by non-linear interpolation to achieve sub-pixel level alignment with the real scene.

[0022] Preferably, the execution steps of the path planning unit further include:

[0023] A construction machinery kinematic constraint model is constructed, including joint angle constraints of a mechanical arm, movement speed constraints of a chassis, and obstacle avoidance safety distance thresholds;

[0024] The fusion mapping relationship between the virtual device contour and the real scene is input into a Markov decision process model to generate a state transition probability matrix;

[0025] The state transition probability matrix is iterated by a Q-learning algorithm to output a construction machinery obstacle avoidance path planning matrix.

[0026] Preferably, the execution steps of the dynamic calibration unit further include:

[0027] The time difference of arrival of the ultra-wideband signal is compensated for multipath effects to generate a first calibration factor;

[0028] Performing environmental noise suppression on the reflection intensity of the laser radar point cloud signal to generate a second calibration factor;

[0029] The first calibration factor, the second calibration factor, and the angular velocity deviation of the inertial navigation data are jointly optimized by using an extended Kalman filter algorithm to update the covariance matrix of the construction positioning coordinates.

[0030] Preferably, the execution steps of the positioning signal receiving unit include:

[0031] Performing time domain synchronization on the ultra-wideband signal, lidar point cloud signal, and inertial navigation data through a multi-band signal fusion algorithm;

[0032] Extracting carrier phase information of the ultra-wideband signal and constructing a signal propagation path loss model;

[0033] The attenuation coefficient of the signal propagation path loss model is dynamically adjusted according to the reflection intensity distribution of the lidar point cloud signal.

[0034] Preferably, the virtual overlay unit further includes:

[0035] Build a generative adversarial network, where the generator is used to predict the deformation parameters of the virtual device outline, and the discriminator is used to evaluate the lighting consistency between the real scene and the virtual outline;

[0036] The adversarial loss function of the generator and the discriminator is optimized through a gradient penalty mechanism to improve the robustness of the pixel-level fusion mapping relationship.

[0037] Preferably, the execution step of the path planning unit further includes:

[0038] A dynamic obstacle prediction module is introduced to perform spatiotemporal modeling of the trajectories of moving targets within the construction area using a long short-term memory network.

[0039] Conflict detection is performed on the spatiotemporal modeling result of the trajectory and the construction machinery obstacle avoidance path planning matrix to generate a real-time obstacle avoidance strategy update instruction.

[0040] Preferably, the execution step of the dynamic calibration unit further includes:

[0041] Construct multi-sensor fusion residual function:

[0042]

[0043] in, represents the residual function value, Indicates the The observation value of the class sensor, represents the state prediction model, represents the sensor noise variance, Indicates the number of sensor types;

[0044] The multi-sensor fusion residual function is subjected to nonlinear least squares optimization by the Levenberg-Marquardt algorithm to output optimal calibration parameters.

[0045] Preferably, the execution steps of the positioning signal receiving unit also include: sparsely representing the frequency hopping sequence of the ultra-wideband signal to construct a compressed sensing observation matrix; reconstructing the signal of the observation matrix through an orthogonal matching pursuit algorithm to extract sub-meter position features of the underground hidden equipment.

[0046] Preferably, the execution step of the AR scene construction unit further includes:

[0047] A semantic segmentation network is introduced to perform real-time recognition of underground pipeline categories in the lidar point cloud signal;

[0048] According to the pipeline category priority, the dynamic spatial distribution map of the underground concealed equipment is rendered in layers to generate a multi-resolution visualization model of the augmented reality scene.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] In the field of underground concealed engineering construction, the high-precision positioning system proposed by the present invention has brought significant positive impacts in many aspects. From the perspective of improving positioning accuracy, the positioning signal receiving unit receives ultra-wideband signals, lidar point cloud signals and inertial navigation data, and after a series of processing, including time domain synchronization, building a signal propagation path loss model and dynamically adjusting the attenuation coefficient, it can obtain underground concealed equipment.

[0051] Sub-meter positioning features are available. This represents a significant improvement over traditional positioning technologies. During underground pipeline construction, precise positioning minimizes pipe docking errors, reduces risks such as subsequent leaks, lowers maintenance costs, and improves the stability and reliability of the entire pipeline system. It avoids repeated excavation and laying due to inaccurate positioning, saving significant manpower, material resources, and time.

[0052] The AR scene construction unit plays a key role in construction visualization. Using simultaneous positioning and mapping algorithms, it generates a three-dimensional spatial model of real-time multi-band positioning signals, generating a dynamic spatial distribution map of hidden underground equipment. Using a semantic segmentation network, it identifies underground pipeline categories in real time and renders them layered according to category priority, creating a multi-resolution visualization model. Using AR, construction workers can intuitively and clearly visualize the specific location, direction, and category of hidden underground equipment and pipelines. This is crucial for construction planning and on-site operations, allowing them to plan safe and efficient construction routes in advance and avoid accidental excavation and damage to underground pipelines during construction. During tunnel construction, construction workers can use the visualization model to proactively understand the distribution of underground pipelines around the tunnel, rationally plan excavation directions and schedules, and effectively prevent damage to surrounding pipelines, thus ensuring construction safety and the normal operation of surrounding facilities.

[0053] The virtual overlay unit achieves spatiotemporal alignment of the dynamic spatial distribution map of underground concealed equipment with preset geographic coordinate data, and constructs a pixel-level fusion mapping relationship between the virtual equipment outline and the real scene. This function is of great value in construction simulation and actual construction guidance. Before construction, virtual overlay can be used to conduct virtual construction simulation. By simulating the interaction between construction equipment and the surrounding environment under different construction plans, potential problems can be identified in advance and construction plans can be optimized. During actual construction, construction personnel can use the intuitive display of virtual overlay to more accurately judge the relationship between construction operations and the surrounding environment, ensuring the accuracy of construction operations. For example, during the installation of large underground concealed equipment, virtual overlay can help construction personnel accurately determine the equipment installation position and angle, avoid collisions with surrounding structures during installation, and improve construction efficiency and installation quality.

[0054] The path planning unit generates a construction machinery obstacle avoidance path planning matrix based on the fusion mapping relationship between virtual and real scenes, combined with a reinforcement learning algorithm. When constructing the kinematic constraint model for the construction machinery, factors such as the manipulator's joint angles, chassis movement speed, and obstacle avoidance safety distance threshold are fully considered. A dynamic obstacle prediction module is also introduced, utilizing a long-short-term memory network to perform spatiotemporal modeling of the trajectories of moving targets within the construction area, and to perform timely conflict detection and strategy updates. This enables construction machinery to quickly and intelligently plan obstacle avoidance paths in complex underground construction environments, effectively avoiding collisions with obstacles. In the presence of multiple dynamic obstacles on the construction site, the construction machinery can adjust its driving path in real time, ensuring its own safety while improving construction efficiency. This avoids equipment damage and construction delays caused by collisions, reduces construction costs, and improves the overall project progress.

[0055] The dynamic calibration unit, based on the Kalman filter algorithm, iteratively corrects the real-time multi-band positioning signal fusion error. By compensating for the multipath effect of the ultra-wideband signal arrival time difference, suppressing the environmental noise of the lidar point cloud signal reflection intensity, and jointly optimizing the calibration factor and the angular velocity deviation of the inertial navigation data using the extended Kalman filter algorithm, the covariance matrix of the construction positioning coordinates is updated. Furthermore, a multi-sensor fusion residual function is constructed and nonlinear least squares optimization is performed to obtain the optimal calibration parameters. These measures ensure the stability and reliability of the positioning system in complex underground environments. Even under various interference conditions, it can continue to provide accurate positioning data, providing solid positioning support for the entire construction process and ensuring the continuity and accuracy of construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a working principle diagram of the high-precision positioning system for underground concealed engineering construction sites according to the present invention;

[0057] Figure 2 Schematic diagram of the workflow for achieving pixel-level fusion for the virtual overlay unit;

[0058] Figure 3 Schematic diagram of the workflow for error correction of the dynamic calibration unit;

[0059] Figure 4 This is a flowchart of the signal processing of the positioning signal receiving unit. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] See also Figure 1-4 The present invention aims to provide a high-precision positioning system for underground concealed engineering construction sites to solve the problems of insufficient positioning accuracy and construction planning difficulties in underground concealed engineering construction. The specific implementation scheme and embodiments of the present invention are described in detail below.

[0062] The high-precision positioning system for underground concealed engineering construction sites of the present invention mainly includes a positioning signal receiving unit, an AR scene construction unit, a virtual overlay unit, a path planning unit and a dynamic calibration unit.

[0063] The positioning signal receiving unit is responsible for receiving the preset geographic coordinate data of underground concealed equipment and the real-time multi-band positioning signals from the construction site. The real-time multi-band positioning signals include ultra-wideband signals, lidar point cloud signals, and inertial navigation data. A multi-band signal fusion algorithm is used to synchronize these signals in the time domain, ensuring that different types of signals remain consistent in time for easier subsequent processing. Next, the carrier phase information of the ultra-wideband signal is extracted to construct a signal propagation path loss model, which is used to describe the energy attenuation of the signal during propagation. Based on the reflection intensity distribution of the lidar point cloud signal, the attenuation coefficient of the signal propagation path loss model is dynamically adjusted to improve the accuracy of the model, thereby more accurately obtaining the location information of the underground concealed equipment.

[0064] The AR scene construction unit uses a simultaneous positioning and mapping algorithm to perform three-dimensional spatial modeling of multi-band signals and generate a dynamic spatial distribution map of underground concealed equipment. The specific steps are as follows: First, extract the ground feature point set from the lidar point cloud signal to construct an initial three-dimensional point cloud framework, providing the infrastructure for subsequent modeling. The phase difference data of the ultra-wideband signal is integrated with the acceleration vector of the inertial navigation data to generate dynamic pose correction parameters. These parameters are used to adjust the posture of the initial three-dimensional point cloud framework. Based on the dynamic pose correction parameters, the initial three-dimensional point cloud framework is optimized in pose, ultimately generating an accurate dynamic spatial distribution map of underground concealed equipment.

[0065] The virtual overlay unit spatially and temporally aligns the dynamic spatial distribution map of underground hidden equipment with preset geographic coordinate data, and constructs a pixel-level fusion mapping relationship between the virtual device outline and the real scene based on a deep learning model. Specifically, the dynamic spatial distribution map of underground hidden equipment is converted into a multi-scale feature tensor, and a convolutional neural network is used to extract the semantic segmentation mask of the preset geographic coordinate data. The attention mechanism is used to calculate the pixel-wise association weights between the multi-scale feature tensor and the semantic segmentation mask to generate a virtual outline fusion coefficient. Based on the virtual outline fusion coefficient, the virtual device outline is rendered using nonlinear interpolation to achieve sub-pixel alignment with the real scene, allowing the virtual device outline to be more accurately integrated with the real scene.

[0066] The path planning unit generates an obstacle avoidance path planning matrix for the construction machinery using a reinforcement learning algorithm, based on the fused mapping relationship between the virtual equipment outline and the real scene. First, a kinematic constraint model for the construction machinery is constructed, including constraints on the manipulator arm joint angles, chassis movement speed, and obstacle avoidance safety distance thresholds, to ensure the safety and feasibility of the construction machinery during movement. The fused mapping relationship between the virtual equipment outline and the real scene is input into a Markov decision process model to generate a state transition probability matrix. Using a Q-learning algorithm, policy iteration is performed on the state transition probability matrix, outputting the obstacle avoidance path planning matrix for the construction machinery and planning a reasonable obstacle avoidance path for the construction machinery.

[0067] The dynamic calibration unit iteratively corrects the fusion error of the ultra-wideband signal, the laser radar point cloud signal and the inertial navigation data based on a Kalman filtering algorithm, and outputs the calibrated construction positioning coordinates. The specific execution steps include: compensating the multipath effect of the time difference of arrival of the ultra-wideband signal to generate a first calibration factor; suppressing the environmental noise of the reflection intensity of the laser radar point cloud signal to generate a second calibration factor; and updating the covariance matrix of the construction positioning coordinates by jointly optimizing the first calibration factor and the second calibration factor with the angular velocity deviation of the inertial navigation data through the extended Kalman filtering algorithm, thereby continuously improving the accuracy of the positioning coordinates.

[0068] Embodiment 1:

[0069] In this embodiment, the specific implementation of the virtual superposition unit is further described. In addition to the basic operations described above, the virtual superposition unit also constructs a generative adversarial network. The generator is used to predict the deformation parameters of the virtual device profile, which determines the shape change of the virtual device profile in different scenarios, making it more consistent with the morphology of the real device. The discriminator is used to evaluate the lighting consistency of the real scene and the virtual profile, ensuring that the virtual device profile is natural and coordinated when integrated into the real scene.

[0070] The adversarial loss function of the generator and the discriminator is optimized through a gradient penalty mechanism. The adversarial loss function can be represented as: wherein is the total adversarial loss function value, is the standard loss function of the generative adversarial network, used to measure the difference between the virtual profile generated by the generator and the real scene; is the gradient penalty coefficient, used to control the weight of the gradient penalty term in the total loss function; is the gradient penalty term, and the calculation formula is wherein represents the expectation, is the sample obtained by sampling, is the sample distribution, is the output of the discriminator to the sample , is the gradient of the discriminator output with respect to the sample , represents norm. By continuously optimizing this adversarial loss function, the robustness of the pixel-level fusion mapping relationship is improved, and the fusion of the virtual device profile and the real scene can remain stable and accurate even in complex construction site environments.

[0071] In addition to the basic path planning steps, the path planning unit also incorporates a dynamic obstacle prediction module. This module uses a long short-term memory (LSTM) network to perform spatiotemporal modeling of the trajectory of moving objects within the construction area. LSTM networks have the ability to memorize long-term information and process time series data, effectively capturing the movement trends of moving objects.

[0072] The spatiotemporal modeling results of the trajectory are used to perform conflict detection with the construction machinery's obstacle avoidance path planning matrix. If a conflict is detected, a real-time obstacle avoidance strategy update instruction is generated, enabling the construction machinery to avoid dynamic obstacles in a timely manner and ensure construction safety.

[0073] Example 2:

[0074] This embodiment describes in detail the application of the multi-sensor fusion residual function in the dynamic calibration unit. In the dynamic calibration unit, a multi-sensor fusion residual function is constructed: ,in, Represents the residual function value, which reflects the degree of difference between the sensor observation value and the state prediction model. The smaller the value, the closer the observed value is to the predicted value, and the better the calibration effect; Indicates the The observation values ​​of the sensor include ultra-wideband signal sensor, lidar sensor, inertial navigation sensor, etc. Obtain corresponding measurement data according to different sensor types; Represents a state prediction model, which is based on the state variables of the system Predicting sensor observations; Represents the sensor noise variance, which is used to measure the The noise level of the sensor measurement data. The larger the noise variance, the higher the uncertainty of the measurement data. Indicates the number of sensor types.

[0075] The nonlinear least squares optimization of the multi-sensor fusion residual function is performed using the Levenberg-Marquardt algorithm. The Levenberg-Marquardt algorithm is an iterative algorithm for solving nonlinear least squares problems, which combines the advantages of the Gauss-Newton method and the gradient descent method. In each iteration, by adjusting the state variables , so that the residual function The optimal calibration parameters are then outputted gradually. These optimal calibration parameters are used to further optimize the fusion of ultra-wideband signals, lidar point cloud signals, and inertial navigation data, improving the accuracy of construction positioning coordinates.

[0076] Example 3:

[0077] This embodiment is used to describe the processing of the frequency hopping sequence of the ultra-wideband signal in the positioning signal receiving unit. In the positioning signal receiving unit, the frequency hopping sequence of the ultra-wideband signal is sparsely represented to construct a compressed sensing observation matrix. Assume that the frequency hopping sequence of the ultra-wideband signal is ,in is the sequence length. Through a specific sparse representation method, it is converted into a sparse vector , making ,here is a sparse basis matrix.

[0078] Constructed compressed sensing observation matrix satisfy Satisfy the finite isometry property (RIP). According to the compressed sensing theory, under certain conditions, a small number of observations can be obtained. Reconstruct the original signal The signal of the observation matrix is ​​reconstructed by the orthogonal matching pursuit algorithm to extract the sub-meter position characteristics of underground hidden equipment. The iterative formula of the orthogonal matching pursuit algorithm is:

[0079]

[0080] in, It is The sparse vector estimate at iterations, is the update vector for each iteration, is the observation vector. Through continuous iteration, it gradually approaches the sparse vector of the original ultra-wideband signal , and then extract the sub-meter location features of underground hidden equipment to improve positioning accuracy.

[0081] Example 4:

[0082] This embodiment describes the application of a semantic segmentation network in an AR scene construction unit. In the AR scene construction unit, a semantic segmentation network is introduced to perform real-time identification of underground pipeline categories in lidar point cloud signals. The semantic segmentation network can adopt a classic neural network architecture such as U-Net. The lidar point cloud signal data is input, and after a series of convolution, pooling, upsampling and other operations of the network, the underground pipeline category label corresponding to each point cloud is output.

[0083] Assume the input of the network is , after a series of convolutional layers and pooling layers After processing, the feature map is obtained . Then through the upsampling layer and deconvolution layers Process the feature map and finally output the predicted category label The network loss function can use the cross entropy loss function: ,in is the sample size, is the number of categories, It is a sample Belong to category The true label (0 or 1), is the network prediction sample Belong to category probability.

[0084] The dynamic spatial distribution map of underground hidden equipment is rendered in layers based on pipeline category priority. For example, important and dangerous pipelines are given higher priority for rendering, allowing construction workers to more intuitively focus on key information when viewing the augmented reality scene. Multi-resolution visualization models of the augmented reality scene are generated to facilitate viewing by personnel with different needs.

[0085] Example 5:

[0086] This embodiment describes in detail the specific process of multi-scale feature tensor and semantic segmentation mask processing in the virtual overlay unit. In the virtual overlay unit, the dynamic spatial distribution map of underground hidden equipment is converted into a multi-scale feature tensor. Assume that the dynamic spatial distribution map of underground hidden equipment is , through a series of convolution operations and pooling operations, feature tensors of different scales are obtained For example, for the scale The feature tensor of , which can be obtained by convolution kernel size , the step size is It is processed by the convolutional layer.

[0087] Extracting semantic segmentation masks of preset geographic coordinate data through convolutional neural networks . Assume that the structure of the convolutional neural network is , enter the preset geographic coordinate data After convolution layer by layer, the semantic segmentation mask is output .

[0088] The pixel association weights between the multi-scale feature tensor and the semantic segmentation mask are calculated based on the attention mechanism. Assume that the calculation formula of the attention mechanism is:

[0089]

[0090] in, is the feature tensor Medium pixels and semantic segmentation masks Medium pixels The correlation score of The function can use calculation methods such as dot product and cosine similarity; is the pixel association weight, which represents the feature tensor Medium pixels and semantic segmentation masks Medium pixels By calculating the weights, the virtual contour fusion coefficient is generated, and then the virtual device contour is rendered more accurately by nonlinear interpolation, achieving sub-pixel alignment with the real scene.

[0091] Example 6:

[0092] This embodiment is used to describe the specific application of the construction machinery kinematic constraint model and related algorithms in the path planning unit. In the path planning unit, the construction machinery kinematic constraint model is constructed. The robot arm joint angle constraint can be expressed as: ,in Indicates the robot arm The angle of the joint, and These are the minimum and maximum values ​​of the joint angle, respectively, to ensure that the robotic arm does not exceed its movable range during movement to avoid mechanical damage.

[0093] The chassis moving speed constraint is: , is the moving speed of the chassis, and These are the minimum and maximum moving speeds allowed by the chassis, ensuring the safety and stability of the construction machinery during movement.

[0094] The obstacle avoidance safety distance threshold is set to , when the distance between the construction machine and the virtual equipment outline or other obstacles is less than , the obstacle avoidance mechanism is triggered.

[0095] The fusion mapping relationship between the virtual device profile and the real scene is input into the Markov decision process model. The Markov decision process can be represented by a five-tuple Indicates that is the state space, i.e., all possible positions and postures of the construction machinery at the construction site; is the action space, i.e., all the movements and manipulations that the construction machine can take; is the state transition probability, indicating that in state Next action After transfer to state probability; is the reward function used to measure the Next action Rewards received; is a discount factor, used to balance the importance of current rewards and future rewards.

[0096] The state transition probability matrix is iterated by the Q-learning algorithm. The update formula of the Q-learning algorithm is:

[0097]

[0098] wherein, is the Q value of performing action in state , is a learning rate, controlling the step size of each update; is the new state transitioned to after performing action , is the optimal action in the new state . Through continuous iteration, the construction machinery obstacle avoidance path planning matrix is output, and a safe and efficient obstacle avoidance path is planned for the construction machinery.

[0099] It should be noted that in this article, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that includes a list of elements does not only include those elements, but also includes other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements, and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A high-precision positioning system for underground concealed engineering construction sites, characterized in that: include: A positioning signal receiving unit, configured to receive preset geographic coordinate data of underground concealed equipment and real-time multi-band positioning signals from the construction site, wherein the real-time multi-band positioning signals include ultra-wideband signals, lidar point cloud signals, and inertial navigation data; An AR scene construction unit is used to perform three-dimensional spatial modeling on the real-time multi-band positioning signal through a synchronous positioning and mapping algorithm to generate a dynamic spatial distribution map of underground concealed equipment; A virtual overlay unit is used to align the dynamic spatial distribution map of the underground hidden equipment with the preset geographic coordinate data in time and space, and to construct a pixel-level fusion mapping relationship between the virtual equipment outline and the real scene based on a deep learning model; A path planning unit, configured to generate an obstacle avoidance path planning matrix for the construction machinery through a reinforcement learning algorithm based on a fusion mapping relationship between the virtual equipment profile and the real scene; A dynamic calibration unit, configured to iteratively correct the fusion error of the ultra-wideband signal, the lidar point cloud signal, and the inertial navigation data based on a Kalman filter algorithm, and output calibrated construction positioning coordinates; The AR scene construction unit executes the following steps: Extracting a ground feature point set from the laser radar point cloud signal to construct an initial three-dimensional point cloud framework; fusing the phase difference data of the ultra-wideband signal and the acceleration vector of the inertial navigation data to generate dynamic posture correction parameters; According to the dynamic posture correction parameters, the posture of the initial three-dimensional point cloud framework is optimized to generate a dynamic spatial distribution map of the underground concealed equipment.

2. The high-precision positioning system for underground concealed engineering construction sites according to claim 1, characterized in that: The execution steps of the virtual overlay unit include: Converting the dynamic spatial distribution map of underground hidden equipment into a multi-scale feature tensor, and extracting a semantic segmentation mask of the preset geographic coordinate data through a convolutional neural network; Calculating pixel association weights of the multi-scale feature tensor and the semantic segmentation mask based on an attention mechanism to generate a virtual contour fusion coefficient; According to the virtual outline fusion coefficient, nonlinear interpolation rendering is performed on the virtual device outline to achieve sub-pixel alignment with the real scene.

3. The high-precision positioning system for underground concealed engineering construction sites according to claim 1, characterized in that: The execution steps of the path planning unit also include: Construct a kinematic constraint model for construction machinery, including arm joint angle constraints, chassis movement speed constraints, and obstacle avoidance safety distance thresholds; Inputting the fusion mapping relationship between the virtual device profile and the real scene into a Markov decision process model to generate a state transition probability matrix; The state transition probability matrix is ​​subjected to policy iteration through the Q-learning algorithm to output the construction machinery obstacle avoidance path planning matrix.

4. The high-precision positioning system for underground concealed engineering construction sites according to claim 1, characterized in that: The execution steps of the dynamic calibration unit also include: Performing multipath effect compensation on the arrival time difference of the ultra-wideband signal to generate a first calibration factor; Performing environmental noise suppression on the reflection intensity of the laser radar point cloud signal to generate a second calibration factor; The first calibration factor, the second calibration factor, and the angular velocity deviation of the inertial navigation data are jointly optimized by using an extended Kalman filter algorithm to update the covariance matrix of the construction positioning coordinates.

5. The high-precision positioning system for underground concealed engineering construction sites according to claim 1, characterized in that: The execution steps of the positioning signal receiving unit include: The ultra-wideband signal, lidar point cloud signal and inertial navigation data are synchronized in time domain through a real-time multi-band positioning signal fusion algorithm; Extracting carrier phase information of the ultra-wideband signal and constructing a signal propagation path loss model; The attenuation coefficient of the signal propagation path loss model is dynamically adjusted according to the reflection intensity distribution of the lidar point cloud signal.

6. The high-precision positioning system for underground concealed engineering construction sites according to claim 2, characterized in that: The virtual overlay unit further includes: Build a generative adversarial network, where the generator is used to predict the deformation parameters of the virtual device outline, and the discriminator is used to evaluate the lighting consistency between the real scene and the virtual outline; The adversarial loss function of the generator and the discriminator is optimized through a gradient penalty mechanism to improve the robustness of the pixel-level fusion mapping relationship.

7. The high-precision positioning system for underground concealed engineering construction sites as claimed in claim 3, characterized in that: The execution steps of the path planning unit also include: A dynamic obstacle prediction module is introduced to perform spatiotemporal modeling of the trajectories of moving targets within the construction area using a long short-term memory network. Conflict detection is performed on the spatiotemporal modeling result of the trajectory and the construction machinery obstacle avoidance path planning matrix to generate a real-time obstacle avoidance strategy update instruction.

8. The high-precision positioning system for underground concealed engineering construction sites as claimed in claim 4, characterized in that: The execution steps of the dynamic calibration unit also include: Construct multi-sensor fusion residual function: ； in, represents the residual function value, represents the observation value of the i-th sensor, represents the state prediction model, represents the sensor noise variance, Indicates the number of sensor types; The multi-sensor fusion residual function is subjected to nonlinear least squares optimization by the Levenberg-Marquardt algorithm to output optimal calibration parameters.

9. The high-precision positioning system for underground concealed engineering construction sites according to claim 5, characterized in that: The execution steps of the positioning signal receiving unit also include: sparsely representing the frequency hopping sequence of the ultra-wideband signal to construct a compressed sensing observation matrix; reconstructing the signal of the observation matrix through an orthogonal matching pursuit algorithm to extract sub-meter level location features of underground hidden equipment.

10. The high-precision positioning system for underground concealed engineering construction sites according to claim 1, characterized in that: The execution steps of the AR scene construction unit further include: A semantic segmentation network is introduced to perform real-time recognition of underground pipeline categories in the lidar point cloud signal; According to the pipeline category priority, the dynamic spatial distribution map of the underground concealed equipment is rendered in layers to generate a multi-resolution visualization model of the augmented reality scene.

Images