Digital twin park vehicle access control system

Abstract

The application provides a digital twin park vehicle and pedestrian access control system, and relates to the technical field of digital twin, which comprises a collection module for obtaining spatial position data of target objects in a park; a construction module for loading preset static three-dimensional model data of the park and generating corresponding digital twin bodies in combination with the spatial position data; a deduction module for performing virtual motion state deduction on the digital twin bodies based on historical access potential fields in response to interruption of spatial position data transmission, updating the current deduction position, and generating a virtual deduction trajectory; and a correction module for extracting the real reproduced position of the target objects, performing deviation convergence calculation, generating a smooth correction instruction, and controlling the digital twin bodies to move from the current deduction position to the real reproduced position in response to the resumption of the acquisition of the spatial position data. The application solves the problems of unnatural target trajectory prediction, poor logic, and visual mutation during reproduction in the visual blind area, and improves the rationality and authenticity of the blind area deduction result.

Description

Technical Field

[0001] This application relates to the field of digital twin technology, and in particular to a digital twin park pedestrian and vehicle access control system. Background Technology

[0002] The existing park personnel and vehicle control system relies on densely deployed cameras and sensors to capture the real-time location of targets and render it in a digital twin interface.

[0003] In practical applications, park environments are complex, with numerous blind spots, such as building corners, tree obstructions, and behind pillars in underground parking garages. When a target enters a blind spot, traditional control systems suffer from the following technical challenges: First, there's the issue of data transmission interruption. Sensor data loss can easily cause the digital twin to disappear or stop, disrupting the continuity of control. Second, there's the problem of mechanical prediction. Existing completion algorithms are mostly based on simple linear interpolation or Kalman filtering, ignoring common traffic patterns within the park, such as pedestrian shortcuts and vehicle traffic flow lines. This can easily lead to abrupt and illogical predicted trajectories, such as passing through walls or going against traffic. Third, there's the problem of visual abrupt changes. When a target leaves the blind spot and is recaptured, the predicted position often deviates from the actual position. Existing technologies often use forced position synchronization, causing the digital twin to shift or jump on the interface, affecting the operator's visual experience and judgment. Summary of the Invention

[0004] To address the aforementioned shortcomings, this application provides a digital twin park pedestrian and vehicle access control system, which solves the problems of unnatural target trajectory prediction, poor logic, and visual abrupt changes during reproduction in existing technologies under blind spots.

[0005] This application provides a digital twin campus pedestrian and vehicle access control system, including:

[0006] The data acquisition module is used to obtain spatial location data of target objects within the park.

[0007] A construction module is used to load the pre-set static 3D model data of the park and, in combination with the spatial location data, generate a digital twin that maps to the target object;

[0008] The simulation module is used to respond to the detection of the interruption of the spatial location data transmission, to perform virtual motion state simulation of the digital twin based on the historical traffic potential energy field, update the current simulation position of the digital twin, and generate a virtual simulation trajectory; the historical traffic potential energy field is a virtual field constructed based on the historical traffic trajectory data of the park;

[0009] The correction module is used to respond to the detection of the recovery of the spatial location data, extract the true reproduction position of the target object, perform deviation convergence calculation based on the true reproduction position and the current simulation position, generate a smooth correction instruction, and control the digital twin to move from the current simulation position to the true reproduction position.

[0010] Optionally, generating a digital twin mapped to the target object includes:

[0011] Identify the type of the target object, including pedestrians, motor vehicles, and non-motor vehicles;

[0012] Retrieve the corresponding 3D model template from the preset model library according to the type;

[0013] Convert the latitude and longitude coordinates in the spatial location data into local coordinates under the three-dimensional coordinates of the park;

[0014] The local coordinates are assigned to the 3D model template, and then rendered in the preset static 3D model data to generate a digital twin.

[0015] Optionally, the preset static three-dimensional model data includes: the building information model of the park, the topological structure data of the park's road network, and the boundary data of inaccessible areas.

[0016] Optionally, virtual motion state simulation of the digital twin is performed based on the historical circulating potential energy field, including:

[0017] Acquire spatial position data at least two moments before the interruption of the spatial position data transmission, and calculate the first velocity vector of the target object;

[0018] Calculate the second velocity vector of the digital twin at its current position in the historical travel potential energy field;

[0019] The first velocity vector and the second velocity vector are weighted and calculated to obtain the third velocity vector;

[0020] The simulation time step is determined based on the sampling frequency of the spatial location data. The displacement increment of the digital twin is calculated by combining the third velocity vector. The displacement increment is then superimposed on the previous position of the digital twin to obtain the current simulation position.

[0021] Optionally, updating the current projected position of the digital twin includes:

[0022] Call the boundary data of the impassable area to determine whether the current simulated location is located in the impassable area;

[0023] In response to the determination that the location is within the impassable area, the nearest projection point from the current projection location to the boundary data is calculated, and the current projection location is updated to the nearest projection point.

[0024] Optionally, generating the virtual simulation trajectory includes:

[0025] Construct a trajectory point cache queue to store the current estimated position after calculation of the historical passable potential energy field and correction of impassable areas in a time series.

[0026] Extract the coordinate sequence from the trajectory point cache queue, and fit the discrete coordinate sequence into a continuous motion path using an interpolation algorithm;

[0027] Configure the motion path with the deduction state visual attributes, and render the motion path with the deduction state visual attributes into the digital twin scene as the virtual deduction trajectory of the digital twin.

[0028] Optionally, the process of constructing the historical pass potential energy field includes:

[0029] Obtain historical passage trajectory data within a preset time period in the park;

[0030] The historical passage trajectory data is processed into a grid, dividing the park space into multiple discrete spatial units;

[0031] Statistically analyze the trajectory point density and motion direction vector within each spatial unit;

[0032] A potential energy field is generated based on the trajectory point density using a density estimation algorithm, and a guiding gradient field is generated by combining the motion direction vector. The historical passage potential energy field is then constructed by fusing these two methods.

[0033] Optionally, the historical traffic potential energy field includes a pedestrian potential energy field and a vehicle potential energy field;

[0034] According to the type of the target object, the corresponding potential energy field is loaded; in response to the target object being a pedestrian, the pedestrian potential energy field is loaded; in response to the target object being a vehicle, the vehicle potential energy field is loaded.

[0035] Optionally, the step of performing deviation convergence calculation and generating smoothing correction instructions includes:

[0036] Calculate the Euclidean distance deviation between the current simulated position and the actual reproduced position;

[0037] In response to the Euclidean distance deviation being less than the position deviation threshold, the position coordinates of the digital twin are updated to the actual reproduced position;

[0038] In response to the Euclidean distance deviation being greater than or equal to the position deviation threshold, a smoothing correction command is generated.

[0039] Optionally, the smoothing correction instruction includes:

[0040] Set a transition time window, and construct a smooth curve interpolation path based on the current simulation position as the starting point and the actual reproduction position as the ending point;

[0041] Based on the transition time window, the smooth curve interpolation path is discretized into multiple transition points, and the digital twin is controlled to move to the real reproduction position frame by frame.

[0042] Compared with existing technologies, this application introduces a historical traffic potential energy field constructed based on historical traffic trajectory data for motion extrapolation, effectively solving the problems of stiff trajectories and logical errors caused by traditional solutions relying solely on linear interpolation or Kalman filtering when signals are interrupted. By utilizing the group traffic habits and spatial guidance information included in the potential energy field, the system can generate virtual trajectories that conform to the actual road network structure and traffic logic of the park within blind spots, improving the rationality and realism of blind spot extrapolation results.

[0043] To address the sudden position changes during signal recovery, this application abandons the existing forced coordinate synchronization method. By performing deviation convergence calculation and generating smooth correction instructions, it controls the digital twin to move from the projected position to the actual reproduced position in a natural transition, eliminating visual instantaneous shifts and jumps, and improving the smoothness and visual stability of the personnel and vehicle management system on the monitoring screen. Attached Figure Description

[0044] Figure 1 A schematic diagram of a digital twin park pedestrian and vehicle access control system provided in this application embodiment;

[0045] Figure 2 A flowchart for constructing a historical traffic potential field is provided for embodiments of this application;

[0046] Figure 3 A flowchart illustrating a virtual motion state simulation provided in this application embodiment;

[0047] Figure 4 A flowchart of the smooth correction instructions provided in the embodiments of this application. Detailed Implementation

[0048] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0049] See Figure 1This is a schematic diagram of a digital twin park pedestrian and vehicle access control system provided in an embodiment of this application; the digital twin park pedestrian and vehicle access control system includes: a data acquisition module 10, a construction module 20, a deduction module 30, and a correction module 40, wherein:

[0050] The data acquisition module 10 is used to acquire spatial location data of target objects within the park.

[0051] Module 20 is used to load the pre-set static 3D model data of the park and, in combination with the spatial location data, generate a digital twin that maps to the target object;

[0052] The simulation module 30 is used to respond to the detection of the interruption of the spatial location data transmission, perform virtual motion state simulation on the digital twin based on the historical traffic potential energy field, update the current simulation position of the digital twin, and generate a virtual simulation trajectory; the historical traffic potential energy field is a virtual field constructed based on the historical traffic trajectory data of the park;

[0053] The correction module 40 is used to respond to the detection of the recovery of the spatial location data, extract the true reproduction position of the target object, perform deviation convergence calculation based on the true reproduction position and the current simulation position, generate a smoothing correction instruction, and control the digital twin to move from the current simulation position to the true reproduction position.

[0054] Regarding the aforementioned data acquisition module 10:

[0055] In the scenario of pedestrian and vehicle access control in digital twin parks, how to map complex and multi-source dynamic targets in the physical world to the virtual three-dimensional digital space in real time and accurately, and overcome the heterogeneity of physical coordinates and virtual coordinates as well as the uniformity of visual presentation, is the core foundation for building digital twins and realizing the mapping between the virtual and physical worlds.

[0056] The acquisition module 10 is based on the principle of multi-source sensing technology fusion. It integrates various types of sensing devices, such as visual, satellite positioning, near-range positioning and inertial navigation. It utilizes the positioning technology characteristics of various devices and selects appropriate acquisition methods for different target objects and scenarios. Through data preprocessing, it converts spatial location data from different sources and in different formats into a unified and standardized format, providing high-quality and consistent input data for the construction module 20.

[0057] The acquisition module 10 compensates for the positioning deficiencies of a single device in complex environments by complementing multiple devices. Through data acquisition processes and coordinate transformation mechanisms, it achieves real-time, accurate, and continuous acquisition of spatial location data of the target object, and processes the acquired raw data into a data format that meets the requirements of subsequent modules.

[0058] In the specific implementation, the first step is to select and deploy sensor equipment. Based on the actual spatial layout of the park, the key control areas, and the movement characteristics of different target objects, suitable multi-source sensing equipment is deployed at key locations such as park entrances and exits, main roads, intersections, parking lots, and pedestrian passages.

[0059] Multi-source sensing devices refer to a collection of various hardware devices used to collect spatial location data of target objects. These include visual devices such as high-definition cameras and 3D cameras; satellite positioning devices such as GPS modules; short-range positioning devices such as Bluetooth beacons and RFID readers; and inertial navigation devices such as IMU (Inertial Measurement Unit). Different types of sensing devices have different positioning accuracies, applicable scenarios, and data output formats. By fusing multiple sources, the positioning deficiencies of a single device in complex park environments can be compensated for.

[0060] For example, in open areas of the park, high-definition cameras, 3D cameras, and other visual devices can be deployed to capture the pixel coordinates of targets in the image; in unobstructed outdoor areas of the park, GPS modules can be equipped on vehicles to directly obtain latitude and longitude coordinates; in areas where satellite signals cannot cover, such as underground parking garages and buildings, short-range positioning devices such as Bluetooth beacons and RFID readers can be deployed to locate pedestrians and indoor vehicles; for some scenarios that require high-precision positioning, UWB ultra-wideband base stations, radar, and other devices can be deployed to improve positioning accuracy; for fast-moving or frequently obstructed targets, inertial navigation devices such as IMU inertial measurement units can be used to ensure the continuity of positioning.

[0061] In practical implementation, the acquisition module 10 serves as the system's data sensing entry point, continuously operating to acquire spatial location data of target objects within the park. Spatial location data refers to data that uniquely identifies the target's geographical location in the physical world, including location information in different formats such as longitude, latitude, and pixel coordinates.

[0062] For example, the acquisition module 10 can obtain the spatial location data of the target object by connecting to the API interface of intelligent sensing devices deployed in the park, such as GPS locators, UWB ultra-wideband base stations, radar, or surveillance cameras with visual recognition functions, and parse and extract the above data stream in real time.

[0063] Then, the spatial location data is preprocessed. First, the interference caused by equipment noise is eliminated by filtering algorithm and abnormal data points are removed. Then, the location data in different formats are converted into a unified latitude and longitude coordinate format to keep the data format consistent.

[0064] Specifically, filtering algorithms can be configured differently based on the noise physical characteristics of different sensing devices in the park environment. For example, for pixel coordinate data collected by visual devices such as high-definition cameras and 3D cameras, since the noise is mainly manifested as high-frequency random jitter caused by changes in illumination or sensor thermal noise, a one-dimensional Gaussian smoothing filtering algorithm can be used. Specifically, first, more than 1000 sets of background pixel coordinate data without target interference collected by the device in history are statistically analyzed, and their noise variance is calculated. Then, the standard deviation of the Gaussian filter is determined using the following formula:

[0065]

[0066] in, The adaptation factor can be adjusted according to the installation height and shooting distance of the visual equipment in the park, and is usually between 1.2 and 1.8.

[0067] Then, the original pixel coordinate data is sorted by time series. Input Gaussian filter, where For the time variable, noise smoothing is achieved through convolution operations, and the filter output formula is:

[0068]

[0069] in, for The filtered pixel coordinates at each time step. The above formula effectively suppresses random noise while preserving the temporal continuity of the target location by performing a Gaussian-distributed weighted average on the original data.

[0070] For devices that output latitude, longitude, or distance data, such as GPS modules and UWB ultra-wideband base stations, the noise is mostly random error with a Gaussian distribution. The Kalman filter algorithm can be used to filter the noise by constructing state equations and observation equations.

[0071] Specifically, define the state vector. ,in, Let k be the two-dimensional planar coordinates of the target object at time k. This represents the corresponding velocity component.

[0072] At the same time, establish state prediction equations ,in, The state transition matrix represents the prior state estimate obtained by predicting the state at time k based on the known information at time k-1. Based on a uniform motion model, it can be set as ,in, The time interval between two positioning data points; This represents the optimal state estimate at time k-1 obtained based on the observation information at time k-1.

[0073] The observation equation is set as follows ,in, Let H be the observation vector at time k, and H be the observation matrix, which can be set as follows: , Let be the observation noise vector at time k, used to characterize the observation error output by the positioning device. The observation noise can be approximated as zero-mean random noise.

[0074] During algorithm execution, the prior state at the current time step is first predicted based on the optimal estimated state at the previous time step, and then the observation data at the current time step is used. Calculate Kalman gain This allows us to correct the prior state and obtain the optimal posterior state estimate for the current time step. Its output position component This is the optimal location data after filtering out Gaussian noise.

[0075] For discrete location data collected by short-range positioning devices such as Bluetooth beacons and RFID readers, outliers are prone to sudden changes due to signal multipath effects. A median filtering algorithm can be used. Specifically, a time sliding window of length W is set, and the original location data sequence collected at the current time t and the preceding W-1 times is used. As the input set. This represents the raw location data collected at the current time t. This indicates the raw location data collected at the start of the time sliding window.

[0076] The algorithm sorts the positional data in the set in ascending order to obtain an ordered sequence, and selects the value at the middle position of the ordered sequence as the filtering result at the current time.

[0077] The algorithm sorts the positional data within the set in ascending order to obtain an ordered sequence, and selects the value at the middle position of this ordered sequence as the filtering result for the current time moment. This method utilizes the robustness of the median to directly eliminate maxima or minima caused by signal jumps within the window, achieving smooth processing of discrete data.

[0078] After filtering, a boundary range verification mechanism is used to remove abnormal data points. Specifically, the latitude and longitude boundary range of the park and the effective acquisition range of each sensing device are stored in advance, such as the latitude and longitude range of the camera's shooting coverage and the latitude and longitude range corresponding to the signal coverage radius of the Bluetooth beacon. For the filtered location data, it is determined whether the data exceeds the corresponding boundary range or the effective acquisition range of the corresponding sensing device. If it exceeds the boundary range, it is directly determined as an abnormal data point and removed.

[0079] For outlier data points that have been removed, linear interpolation can be used to supplement them. Based on the coordinates and time of two valid data points before and after the outlier data point, the interpolated coordinates are calculated through linear fitting. Specifically, the time corresponding to the outlier data point can be set as follows: The time of its previous valid data point is The coordinates are The time of the last valid data point is The coordinates are Then the interpolated coordinates The following formula can be used to maintain the continuity of the data sequence:

[0080]

[0081]

[0082] After data denoising and anomaly removal, a unified conversion is performed on location data of different formats. For pixel coordinates obtained by vision devices, device calibration is performed first. Multiple feature points with known latitude and longitude are selected in the park, and their pixel coordinates and corresponding latitude and longitude coordinates are recorded. A perspective projection transformation model can be used to construct the transformation relationship. After solving the calibration parameters by the least squares method, the filtered and anomaly-removed pixel coordinates are substituted into the model to calculate the latitude and longitude coordinates.

[0083] The perspective projection transformation model establishes a mapping relationship between pixel coordinates on a two-dimensional image plane and geographic coordinates in three-dimensional real space. Because perspective distortion occurs when visual devices such as cameras capture images, the target position in real space and the pixel position in the image are not simply linearly correlated. This model corrects this non-linear distortion by introducing specific transformation parameters, thereby achieving an accurate conversion of pixel coordinates to latitude and longitude coordinates.

[0084] Specifically, to establish the conversion relationship, the first step is to prepare for equipment calibration. After the visual equipment is installed and fixed and the shooting angle is adjusted, select multiple feature points with clear physical locations within the area covered by the equipment. These feature points are accurately measured using GPS to obtain their true latitude and longitude coordinates, and they should be evenly distributed in the shooting frame to avoid being concentrated in one area.

[0085] For example, fixed and immovable points such as corners of signs, center points of street light bases, and intersections of ground markings within the park can be selected. At the same time, the horizontal and vertical pixel values ​​of each feature point in the device's captured image, i.e., pixel coordinates, as well as the corresponding real latitude and longitude coordinates, can be recorded one by one to form multiple sets of complete calibration sample data.

[0086] Next, an appropriate transformation equation is established. Based on the logic of the perspective projection transformation model, a mapping relationship from pixel coordinates to latitude and longitude coordinates is constructed. Considering that the height change of the target object in the park scene is relatively small, a simplified two-dimensional perspective projection transformation logic can be adopted. Multiple coefficients are introduced to characterize the nonlinear relationship between pixel coordinates and latitude and longitude coordinates, and to correct the error caused by perspective distortion.

[0087] Then, the calibration parameters are solved by the least squares method. The previously obtained multiple sets of calibration sample data are substituted into the constructed two-dimensional perspective projection transformation logic to form an overdetermined variance set. By minimizing the sum of squared errors of coordinate transformation, the optimal solution of all unknown coefficients is calculated. In this way, the perspective projection transformation model is constructed and the specific conversion rules between pixel coordinates and latitude and longitude coordinates are clarified through the above process.

[0088] To more clearly illustrate the construction and parameter solving process of the perspective projection transformation model in this embodiment, the following detailed explanation is provided in conjunction with specific data.

[0089] For example, in this embodiment, the system uses a homography matrix to represent the nonlinear mapping relationship from the pixel coordinate system to the local geographic coordinate system of the park, and the constructed transformation equation is shown in the following formula:

[0090]

[0091] in, Represents the coordinates of pixels in an image captured by a visual sensing device. Represents the corresponding Cartesian coordinates in the physical world of the park; matrix The 3×3 homography matrix to be determined includes... to There are 8 independent parameters, Normalized to 1.

[0092] During the system deployment phase, the main intersection of the park's main road was selected as the calibration area, and four sets of landmark feature points, such as zebra crossing corners, were collected as calibration samples. Specifically, the pixel coordinates of the first set of samples (150, 200) correspond to the physical coordinates (10.5, 5.0); the pixel coordinates of the second set of samples (850, 210) correspond to the physical coordinates (25.5, 5.2); the pixel coordinates of the third set of samples (120, 900) correspond to the physical coordinates (10.2, 35.0); and the pixel coordinates of the fourth set of samples (880, 910) correspond to the physical coordinates (26.0, 35.5).

[0093] Substituting the above four sets of sample data into the transformation equations respectively, an overdetermined linear equation system is constructed, and the matrix is ​​solved using the Singular Value Decomposition (SVD) algorithm. The optimal solution. In one actual calculation of this embodiment, the parameter matrix obtained is... for:

[0094]

[0095] To verify the transformation accuracy of the model, the system selected a non-calibrated point, such as the bottom of a street lamp pole, for testing. The pixel coordinates of this test point are (500, 500), which were substituted into the matrix obtained above. In the model, the calculated physical coordinates are (17.85, 17.90).

[0096] Subsequently, the actual physical coordinates of the bottom of the streetlight pole were measured using an RTK-GPS device as (17.80, 17.88). The system further calculated the Euclidean distance error between the model's calculated value and the actual measured value, which was approximately 0.053 meters. Since this error value is much smaller than the preset threshold required for the park's vehicle control scenario, the model calibration was deemed successful, and the system immediately transferred the matrix... The parameters are fixed and used for coordinate transformation calculations of the target position in subsequent real-time video streams.

[0097] Finally, the conversion verification is performed. Several feature points that were not calibrated are reselected, and their pixel coordinates are substituted into the constructed conversion model to calculate the corresponding latitude and longitude coordinates. These coordinates are then compared with the actual latitude and longitude coordinates of the feature point measured by GPS. If the error between the two is within the preset threshold for the park management scenario, it indicates that the conversion model is effective. If the error exceeds the threshold, feature points are reselected or the number of feature points is increased, and the calibration process is repeated until the model conversion accuracy meets the requirements.

[0098] The preset thresholds are determined by combining the park's management and control needs, the hardware performance of the vision-based devices, and the park's spatial scale. For example, if the management and control needs focus on accurate trajectory tracking and target positioning, the thresholds can be adapted to high-precision positioning standards; if only the traffic situation needs to be statistically analyzed and the approximate location needs to be identified, the thresholds can be appropriately relaxed. At the same time, a reasonable margin is reserved with reference to the inherent positioning error of the devices, and the final thresholds are determined after verification through multi-area field testing.

[0099] Finally, the standardized latitude and longitude coordinate data is transmitted to the construction module 20 in real time to realize the continuous and accurate collection of spatial location data of target objects such as pedestrians, motor vehicles, and non-motor vehicles in the park.

[0100] Regarding the aforementioned building module 20:

[0101] In existing digital twin park management systems, digital twin scenarios lack effective spatial constraints and semantic logic. The general model cannot distinguish the attributes of different types of target objects, and the latitude and longitude coordinates output by the acquisition module 10 cannot be directly used for rendering in the park's 3D virtual scene. This results in an inaccurate mapping relationship between the generated digital twin and the physical target object, failing to truly reflect the position and status of the target object within the park, and affecting the accurate mapping between the physical world and the virtual world.

[0102] To address the lack of spatial constraints and semantic logic in digital twin scenarios, module 20 constructs a virtual park environment with geometric constraints and physical attributes based on pre-set static 3D model data. It achieves accurate matching of model templates through target object type recognition, adapts physical latitude and longitude coordinates to virtual 3D local coordinates using coordinate transformation technology, and finally integrates the model templates with location information through 3D rendering technology to generate a digital twin that maps one-to-one with the physical target.

[0103] In practice, the first step is to load the pre-set static 3D model data of the park. This pre-set static 3D model data refers to the digital baseboard of the park, which is pre-made using techniques such as oblique photography, laser scanning, or manual modeling before the system runs. It is used to build the spatial environment of the virtual park and includes the park's building information model, the topological structure data of the park's road network, and the boundary data of inaccessible areas.

[0104] Among them, the building information model is used to accurately represent the appearance and geometry of buildings; the topological data of the park's road network is used to define the connection relationship and traffic direction of roads, such as the connection method between main roads and branch roads, one-way or two-way traffic rules, etc., providing a logical basis for path deduction; the boundary data of impassable areas is used to identify the electronic fence range of physical obstacles such as lawns, lakes, and walls, such as the boundary outline of the landscape lake in the park and the direction of the wall, providing spatial motion constraints for the digital twin.

[0105] By loading the aforementioned static data, the system constructs a virtual scene with geometric constraints and physical properties for the activities of dynamic targets.

[0106] Next, target object type identification is performed. The construction module 20 receives standardized latitude and longitude coordinate data transmitted by the acquisition module 10, and simultaneously interfaces with the additional data interface of the sensing device. This additional data interface refers to the communication channel used by the sensing device to transmit structured metadata in addition to outputting location data, supporting data connection with the construction module 20 via standard protocols such as HTTP / REST and MQTT. Through this interface, the construction module 20 can obtain depth attribute information preprocessed and generated by the edge computing unit of the sensing device.

[0107] Specifically, for visual perception devices, they are equipped with pre-trained target detection neural network models, such as YOLO or SSD series models based on convolutional neural network architecture.

[0108] For example, this embodiment preferably uses the YOLOv8n model. The model architecture includes four core modules: an input layer, a backbone feature extraction network, a neck feature fusion network, and a head detection output layer. Specifically, the input layer adopts a mosaic data augmentation strategy to scale the input video frames to 640×640 pixels. At the same time, random cropping, flipping, and color gamut distortion are used to expand sample diversity and adapt to different lighting conditions in the park, such as strong sunlight on sunny days, weak light at night, and occluded scenes. The backbone feature extraction network uses a C2f module combined with a spatial pyramid pooling module. Pooling-Fast (SPPF) enables multi-scale feature extraction. The C2f module enhances the preservation of low-level positional information and high-level semantic features through branched parallel convolution and residual connections. The neck feature fusion network adopts a PAN-FPN structure, which performs cross-scale fusion of the 8x, 16x, and 32x downsampled feature maps output by the backbone feature extraction network through upsampling and downsampling operations, outputting fused feature maps at three scales of 80×80, 40×40, and 20×20. The head detection output layer adopts a decoupled head structure, which outputs the target class probability and detection box coordinate offset through classification and regression branches, respectively. Finally, the non-maximum suppression algorithm is used to filter out the effective detection results.

[0109] During the model training phase, fine-tuning and optimization are performed for park scenarios. First, a park-specific dataset is constructed, including pedestrian images, motor vehicle images, and non-motor vehicle images, covering typical park scenarios such as daytime, nighttime, sunny days, rainy days, building corners, and open squares. The annotation format adopts the VOC standard, including object detection boxes and category labels. The training set, validation set, and test set can be divided in an 8:1:1 ratio.

[0110] The training batch size can be set to 8, the initial learning rate can be set to 0.001, and a cosine annealing decay strategy is used, decaying to 0.01 after 50 iterations. The optimizer can be AdamW, and the loss function can be CIoU loss or Focal loss, with the modulation factor set to 2.0 and the class balance factor set to 0.25. The training process iterates for 100 rounds, and the mean accuracy (mAP)@0.5 is used as the evaluation metric for the model on the validation set.

[0111] When applying this model, the real-time video frame sequence collected by the visual sensor is first extracted frame by frame as the input data of the model. Through the inference of the model after training and optimization, the detection box coordinate data of the target object in the image pixel coordinate system is output, including the coordinates of the center point of the box and the pixel width and height. Then, combined with the pre-calibrated camera intrinsic and extrinsic parameter matrices in the scene, the mapping from pixel coordinates to the world physical coordinate system is completed.

[0112] The camera intrinsic parameter matrix can be obtained using the Zhang Zhengyou calibration method. This matrix describes the camera's optical characteristics and imaging geometry. Its focal length along the x-axis and y-axis matches the imaging requirements of the park's visual perception equipment. The x-coordinate of the principal point corresponds to the midpoint of the image width, and the y-coordinate corresponds to the midpoint of the image height, ensuring alignment between the imaging reference and the image's geometric center. Lens distortion correction parameters include radial distortion coefficients and tangential distortion coefficients. Both coefficients are small correction values ​​calculated through calibration, used to accurately compensate for radial and tangential distortions generated during lens imaging, ensuring the accuracy of pixel coordinate transformation.

[0113] The camera extrinsic parameter matrix consists of a rotation matrix and a translation vector, which is used to characterize the spatial attitude and positional relationship of the camera coordinate system relative to the world coordinate system. The world coordinate system is defined as follows: with the park's reference point as the origin, the X-axis extends along the main road of the park, the Y-axis is perpendicular to the main road and parallel to the ground, and the Z-axis is perpendicular to the ground and upward.

[0114] The rotation matrix corresponds to three rotation angles: roll, pitch, and yaw. These are determined by the camera's installation orientation within the park. This rotation matrix enables precise alignment of the camera coordinate system with the world coordinate system in terms of attitude. The translation vector corresponds to three translation distances, which are the values ​​representing the actual deployment position of the camera relative to the park's reference point, accurately reflecting the camera's installation orientation within the park space.

[0115] When using perspective transformation algorithm for coordinate mapping, the original pixel coordinates are first corrected by distortion correction formula. This formula is based on principal point coordinates and distortion correction coefficient, and combines the deviation between the original pixel coordinates and principal point coordinates and the square term of the deviation to construct multi-dimensional correction logic, thereby achieving accurate compensation for lens distortion.

[0116] The corrected pixel coordinates are then converted into homogeneous coordinates in the camera coordinate system. During the conversion process, the ground height parameter of the target object is introduced. The appropriate value for this parameter is determined based on the conventional physical height characteristics of pedestrians, motor vehicles, and non-motor vehicles.

[0117] Finally, the coordinate transformation from the camera coordinate system to the world coordinate system is completed through the extrinsic parameter matrix, and the three-dimensional coordinates of the target object in the physical space of the park are obtained.

[0118] Based on the target position in two consecutive frames in the world coordinate system, and the frame interval time derived from the frame rate of the visual sensor, the actual moving velocity vector of the target object is calculated. The velocity in the Z-axis direction is negligible due to its small change. At the same time, based on the calibration ratio of the pixel width and height of the detection box to the actual width and height of the target in the world coordinate system, which can be obtained by calibration using reference objects of known size in the park, and combined with the type characteristics of the target object to correct the aspect ratio, the geometric dimensions (length, width, and height) of the target object in physical space are calculated. The height parameter can be determined by the difference in world coordinates from the ground to the top edge of the detection box. Finally, the physical dimensions and velocity vector data of the target object are output, providing accurate input for subsequent type recognition and twin construction.

[0119] For positioning devices such as GPS / BeiDou locators and UWB base stations, the registration ID of the bound target, device type identifier, and instantaneous speed are directly output through protocol messages; for RFID readers, the static attribute data pre-stored in the electronic tag is read, such as the number of wheels and vehicle type classification code.

[0120] Through the aforementioned interface, module 20 acquires multi-dimensional attribute information, including the physical dimensions, movement speed, and appearance features of the target object. Based on multi-feature fusion recognition rules, it classifies the target object into three categories: pedestrians, motor vehicles, and non-motor vehicles. The fusion recognition rules are a comprehensive judgment logic set according to the distribution range of the core features of different types of targets in physical space. Utilizing the geometric and motion constraints of the physical world, the initial recognition results from the sensing device are further cleaned and calibrated to eliminate false detections or drift that may occur with a single data source.

[0121] For example, the determination logic is as follows: if the initial screening result of the sensing device is a non-motorized target, and the system detects that the target's physical size is small, its moving speed is stable within the range of normal human walking speed, and it does not have fixed wheel structure features, then it is determined to be a pedestrian; if the initial screening result of the sensing device is a vehicle, and the target's physical size is large, its moving speed meets the park's speed limit features, it has four or more wheel structure features, or meets the length-to-width ratio standard of a motor vehicle, then it is determined to be a motor vehicle; if the target's physical size is between that of a pedestrian and a motor vehicle, and the feature data shows through analysis that it has two wheel structure features or meets riding posture features, then it is determined to be a non-motorized vehicle.

[0122] Next, the 3D model template is retrieved. The system has a built-in pre-set template library, which stores 3D model templates adapted to different types and sizes of target objects, including various sub-scene templates for pedestrians, motor vehicles, and non-motor vehicles.

[0123] Among them, 3D model templates refer to standardized 3D models stored in the template library that correspond to different types of target objects. They have geometric structures and basic attributes that match the physical target objects and are used as virtual carriers for quickly generating digital twins, such as human body shape models for pedestrians, various vehicle models for motor vehicles, and bicycle or electric vehicle models for non-motorized vehicles.

[0124] Based on the identified target object type, the corresponding 3D model template is retrieved from the template library. At the same time, combined with the specific characteristics of the target object, such as the vehicle type characteristics of motor vehicles and the morphological characteristics of pedestrians, the optimal template is further matched to maximize the match between the model template and the attributes of the physical target object.

[0125] Next, perform coordinate transformation. First, establish a unified three-dimensional coordinate system for the park, select fixed reference points within the park, such as the center point of the base of the park's landmark building or the intersection of the park's boundaries, as the origin of the coordinate system, set the X-axis along a fixed direction within the park, the Y-axis perpendicular to the X-axis and parallel to the ground, and the Z-axis perpendicular to the ground and upward, and clarify the scale of the coordinate system.

[0126] Next, a coordinate transformation algorithm is used for conversion. Specifically, the latitude and longitude coordinates output by the acquisition module 10 are converted into geodetic rectangular coordinates using coordinate transformation methods relevant to the field of geodesy. Geodesy provides standardized, high-precision mathematical models and calculation methods for conversions between different coordinate systems. Latitude and longitude coordinates belong to geographic coordinate system data, while geodetic rectangular coordinates are a three-dimensional rectangular coordinate system established with the Earth's center of mass as the origin, describing the spatial location of a point through X, Y, and Z coordinates.

[0127] In this embodiment, the WGS-84 ellipsoid model can be used as the conversion reference, setting the Earth's semi-major axis a to 6,378,137 meters, the oblateness f to 1 / 298.257223563, and the first eccentricity squared... For any target latitude and longitude point collected. Where B is latitude, L is longitude, and h is altitude, first calculate the radius of curvature N of the circumpolar region at that latitude using the following formula:

[0128]

[0129] Then, its geodetic rectangular coordinates are calculated using a set of formulas. ,in:

[0130]

[0131]

[0132]

[0133] Subsequently, based on the park's geographical location parameters, a mapping relationship between geodetic rectangular coordinates and the park's three-dimensional local coordinates was established. The park's geographical location parameters include the overall latitude and longitude range of the park, the latitude and longitude coordinates of the reference point, and the geodetic rectangular coordinates corresponding to the reference point. This mapping relationship was constructed through spatial geometric operations.

[0134] Specifically, the spatial geometric calculations include: the system using the geodetic rectangular coordinates of the park's reference point as a three-dimensional translation vector; simultaneously calculating the sine and cosine values ​​of the reference point's longitude and latitude; and combining these with the pre-defined coordinate axis orientations of the park's three-dimensional scene to construct a rotation matrix. This rotation matrix represents the angular deflection relationship between the axes of the Earth's geocentric coordinate system and the axes of the park's local coordinate system. Finally, the aforementioned three-dimensional translation vector and rotation matrix are combined to generate an integrated coordinate transformation model.

[0135] For example, in order to eliminate the influence of the Earth's curvature on the small-scale planar mapping of the park, the system selects the center of the park or the base of a landmark building as the local coordinate origin. Its geodetic rectangular coordinates are determined as follows: The corresponding latitude and longitude are .

[0136] A rotation matrix R is constructed using spatial geometric operations. This matrix is ​​used to rotate the geocentric coordinate system to a position relative to the origin. The rotation matrix for the East-North-Up (ENU) coordinate system with the tangent point is constructed as follows:

[0137]

[0138] Finally, an integrated coordinate transformation model is generated, and the calculation formula is as follows:

[0139]

[0140] in, The target point is at the origin. The coordinates in the local northeast-sky coordinate system established for the tangent point. Indicates the eastward coordinate components. Indicates the northward coordinate components. Represents the celestial coordinate components; Represents the geodetic rectangular coordinates of the target point. Represents the local coordinate origin The geodetic rectangular coordinates.

[0141] Next, the coordinate transformation model described above is used to perform specific coordinate transformation calculations on the collected points. Specifically, a translation operation is first performed. By subtracting the translation vector, the origin of the geodetic rectangular coordinate system is translated and normalized to the park's reference point, eliminating the large numerical positional deviation between the park and the Earth's center of mass. Then, a rotation operation is performed. The rotation matrix is ​​used to rotate the translated coordinates axially, so that the transformed Z-axis is perpendicular to the park's ground, and the X and Y axes are aligned with the spatial direction of the park's virtual scene. Finally, local coordinates adapted to the park's static 3D model data are obtained.

[0142] It should be noted that the vertical Z-coordinate can be set as a reference value according to the target object type and scene characteristics. For example, the Z-coordinate of a ground target is based on the height of the park ground and can be corrected by combining the height data collected by the sensing device, so that the local coordinate can reflect the actual position of the target object in the three-dimensional space of the park.

[0143] To verify the accuracy and effectiveness of the above coordinate transformation model in a park setting, this embodiment selects the center point of the park gate as the origin, with its GPS measured coordinates of (32.05°N, 118.78°E, 20m). The corresponding rectangular coordinates of this origin are calculated using geodetic formulas. It is approximately (−2606863.31,4776472.24,3365320.15).

[0144] Assume the system collects the real-time coordinates of a moving vehicle as (32.0501°N, 118.7801°E, 20m), meaning the latitude and longitude each deviate by 0.0001 degrees. First, based on the WGS-84 ellipsoid model parameters, the vehicle's latitude and longitude coordinates are converted to geodetic rectangular coordinates, and the vehicle's three-dimensional spatial rectangular coordinate values ​​are calculated. It is approximately (-2606871.58, 4776466.85, 3365320.15).

[0145] Subsequently, the value is substituted into the transformation model to perform a translation operation. The large numerical deviation is eliminated by calculating the difference vector between the vehicle coordinates and the origin coordinates, and the intermediate vector after translation is obtained as ∆P≈(-8.27,-5.39,0).

[0146] Next, a rotation matrix R is constructed. Based on the origin longitude L0 of 118.78° and latitude B0 of 32.05°, the trigonometric function values ​​sin(118.78°)≈0.8765, cos(118.78°)≈-0.4814, sin(32.05°)≈0.5307, and cos(32.05°)≈0.8476 are calculated to generate specific transformation matrix parameters.

[0147] The system performs matrix multiplication, that is:

[0148]

[0149] in, ; This represents the three-dimensional position vector of the target point in the local coordinate system of the park after coordinate transformation, with its components corresponding to the X, Y, and Z coordinates of the local coordinate system.

[0150] The translated intermediate vector ∆P is mapped to the local coordinate system of the park, and the position of the vehicle in the local coordinate system of the park is calculated. The distance is (9.43, 11.09, 0) meters, indicating that the vehicle is located 9.43 meters east and 11.09 meters north of the park gate, with a Z-axis height difference of 0. The calculated result has an error of less than 5 centimeters from the actual physical distance measurement, which verifies that the coordinate transformation algorithm can accurately map latitude and longitude to local coordinates that are compatible with the static 3D model data of the park, so that the positioning of the digital twin in the virtual scene is highly consistent with the physical world.

[0151] After coordinate transformation, a distinguishing identifier is assigned to the 3D model template, and rendering is completed. The transformed local coordinates are then assigned to the retrieved 3D model template, allowing it to be positioned within the virtual scene constructed from static 3D model data, corresponding to the physical target. Furthermore, different visual features, such as varying shades of color, are set for different target objects of the same type. Simultaneously, based on the instantaneous velocity and direction of motion information of the target objects acquired by sensing devices, the dynamic attributes of the 3D model template are matched to render dynamic effects in real-time that correspond to the current motion state of the physical target, such as wheel rotation matching vehicle speed or gait frequency matching pedestrian speed, enhancing the realism and simulation of the digital twin.

[0152] After rendering is complete, a digital twin corresponding to the physical target object is generated. The system stores the location coordinates, type identifier and other information of the digital twin in the system database and displays the digital twin in the static 3D model scene of the park in real time. At the same time, the digital twin is updated synchronously with the position change of the physical target object, realizing the real-time mapping between the physical world and the virtual world.

[0153] Regarding the aforementioned deduction module 30:

[0154] Digital twin parks contain numerous blind spots, such as building corners, tree obstructions, and underground parking garage pillars. When a target object enters these blind spots, spatial location data transmission is interrupted. Existing technologies, such as traditional control systems, may cause the digital twin to disappear or become stationary, disrupting control continuity. Alternatively, they may use simple linear interpolation or Kalman filtering for trajectory prediction, ignoring pedestrian shortcuts and vehicular traffic flow within the park. This results in unnatural and illogical predicted trajectories, sometimes even including instances of objects passing through walls or moving in the wrong direction. These systems fail to accurately recreate the target object's movement within the blind spots, impacting the accuracy and effectiveness of control.

[0155] The simulation module 30 uses historical traffic potential energy fields combined with the target object's motion state before the interruption to simulate virtual motion states within blind zones. The historical potential energy field integrates group traffic habits and spatial guidance information from a large amount of historical traffic trajectory data within the park, providing motion constraints that conform to the actual park environment. By calculating the first velocity vector of the target object before the interruption and the second velocity vector from the historical traffic potential energy field, a weighted fusion is performed to obtain a realistic third velocity vector. This vector is then combined with the simulation time step to calculate the displacement increment, determining the current simulation position. Simultaneously, positional deviations are corrected using boundary data of impassable areas. Finally, based on continuous simulation positions, a coherent virtual simulation trajectory is generated.

[0156] See Figure 2 The flowchart for constructing a historical traffic potential energy field provided in this application embodiment includes steps S201 to S204, wherein:

[0157] S201, Obtain historical passage trajectory data within a preset time period in the park;

[0158] S202, The historical passage trajectory data is processed into a grid, dividing the park space into several discrete spatial units;

[0159] S203, Calculate the trajectory point density and motion direction vector within each spatial unit;

[0160] S204, using a density estimation algorithm to generate a potential energy field based on the trajectory point density, and combining it with the motion direction vector to generate a guiding gradient field, and then fusing them to construct the historical passage potential energy field.

[0161] In the specific implementation, the first step is to construct the historical traffic potential field. Specifically, this involves acquiring historical traffic trajectory data within a preset time period in the park, including various traffic trajectory point sets for pedestrians, motor vehicles, and non-motor vehicles. .in, This represents the nth trajectory point in the set of travel trajectory points, where n is the trajectory point index. Let n represent the S-th trajectory point in the set of travel trajectory points, where S is the total number of trajectory points, and n and S are positive integers.

[0162] These historical traffic trajectory data are processed into a grid, and the grid resolution parameter r is set to divide the entire park space into several discrete spatial units. Where r represents the unit size of the spatial grid division in the park; This represents the spatial cell in the i-th row and j-th column, where i and j represent the row index and column index of this spatial cell in the grid, respectively.

[0163] Subsequently, the trajectory point density and motion direction vector within each spatial unit are statistically analyzed. The trajectory point density represents the frequency of passage in the area, and the motion direction vector represents the main motion direction of the target object in the area.

[0164] The potential energy field can be generated based on the density of trajectory points using density estimation algorithms. Kernel density estimation algorithm can be used, with the center coordinates of each spatial cell as the reference point, a bandwidth coefficient as the smoothing window radius, and a kernel function, such as a Gaussian kernel function, set. The distances of all historical trajectory points falling within the window relative to the reference point are substituted into the standard Gaussian kernel function for weighted summation to obtain the probability density value of the spatial cell.

[0165] For example, the input to the kernel density estimation algorithm is the number of cells falling within the spatial cell center. Center and bandwidth Given the set of trajectory points within the radius's neighborhood, the output is the probability density value of that cell. In this embodiment, the bandwidth coefficient can be set. Given a height of 1.5 meters, set the standard Gaussian kernel function K(q):

[0166]

[0167] Wherein, variable q represents the normalized distance between historical trajectory points and the baseline point, i.e. , As the center of the space unit With the nth trajectory point The Euclidean distance between them.

[0168] The specific formula for calculating the probability density value is as follows:

[0169]

[0170] To verify the effectiveness of the algorithm, assume that there are two historical trajectory points A (100.5, 100.2) and B (99.8, 100.1) within a 1.5-meter radius around the center of the spatial cell with coordinates (100, 100). Calculate the distance from point A to the center. The distance from point B to the center is approximately 0.54 meters. Approximately 0.22 meters. Substituting into the Gaussian kernel formula, calculate the weighted contribution value; the contribution of point A... Approximately 0.37, contribution of point B Approximately 0.39. The density value of this cell is obtained by normalized summation. It is approximately equal to 0.25.

[0171] Subsequently, a negative correlation mapping relationship between density and potential energy is established, and the probability density value is transformed using an exponential decay function or an inverse proportional function. For example, an exponential decay mapping formula can be used:

[0172]

[0173] in, This represents the potential energy value of the corresponding spatial unit; The maximum potential energy reference value can be set to 100; The adjustment coefficient can be set to 2.

[0174] The above probability density values Substituting the values, the potential energy value of the space unit is calculated. It is approximately equal to 60.65.

[0175] Thus, through this algorithm, regions with higher trajectory point density generate lower potential energy values, forming potential energy valleys and representing high passage probability; conversely, regions with lower trajectory point density generate higher potential energy values, forming potential energy barriers and representing low passage probability.

[0176] Simultaneously, a guiding gradient field is generated by combining the motion direction vector of each spatial unit. For each spatial unit, based on the motion direction vector of that unit, the directional difference between adjacent spatial units is calculated. Through gradient calculation methods, such as the finite difference method, a gradient vector is generated. The direction of the gradient vector is the optimal motion direction of that spatial unit. The magnitude of the gradient vector represents the guiding strength of that direction. The more uniform the motion direction of a region, the larger the gradient vector and the stronger the guiding effect, thus forming a guiding gradient field covering the entire park space.

[0177] Specifically, first calculate each spatial unit. The original average motion direction vector within The formula is:

[0178]

[0179] in, The nth trajectory point within the spatial unit The instantaneous velocity vector.

[0180] For example, within cell (10,10), there are two trajectory points with velocity vectors of (0,2) and (0.2,1.8) respectively. After normalization and averaging, we obtain... The value is approximately (0.05, 0.99), representing the predominantly northward movement within this region.

[0181] Next, the directional differences between adjacent spatial units can be calculated using the finite difference method to generate the steering gradient vector. The specific formula is as follows:

[0182]

[0183] in, As a smoothing factor, it can be set to 0.5 in this embodiment; This represents the average motion direction vector of the historical trajectory points within the spatial cell of the i-th row and j-th column; This represents the average motion direction vector of the next adjacent unit in the first index direction of the spatial unit, i.e., the spatial unit in the (i+1)th row and jth column. This represents the average motion direction vector of the spatial cell adjacent to the previous cell in the first index direction, i.e., the spatial cell in the (i-1)th row and jth column. and These represent the average motion direction vectors of the adjacent units of the spatial unit in the second index direction.

[0184] If there is a sudden change in the motion direction of adjacent spatial units, the difference term will correct the vector of the current unit. The final gradient vector is obtained. Its direction is the optimal motion direction of the spatial unit after neighborhood smoothing, and its magnitude is... Characterizing the guiding consistency strength in this direction, if the surrounding directions are chaotic, vector cancellation leads to a smaller magnitude and weaker guidance, thus forming a guiding gradient field covering the entire park space. .

[0185] The generated potential energy field is fused with the guiding gradient field, since the potential energy field generated in the previous steps... Essentially a scalar field, but with a directional gradient field. Since the potential energy is a vector field, to achieve effective fusion, the scalar potential energy is first converted into vector form. Specifically, the negative gradient vector of the potential energy field of each spatial unit can be calculated using the central difference method. This vector is essentially a potential energy driving force, and its calculation principle is based on the current spatial unit. The X-axis component is determined by the potential energy difference between the left and right adjacent units, and the Y-axis component is determined by the potential energy difference between the upper and lower adjacent units. Finally, the opposite direction is taken, and the formula is:

[0186]

[0187] Numerically, this vector points in the direction where the potential energy decreases the fastest, i.e., the direction with the higher density of historical trajectories. Physically, it manifests as a corrective force that pulls deviations from the target back to the commonly used path.

[0188] Subsequently, the potential energy driving force vector and guided gradient vector Normalize the magnitudes of the potential energy separately to obtain the normalized negative gradient vector. and normalized guided gradient vector The calculation formula is:

[0189]

[0190]

[0191] This unifies the numerical range of both to a unit vector, eliminating the dimensional differences.

[0192] Based on this, the fusion weights are set, where, the fusion weights are set. As the potential energy field weight, To guide the gradient field weights, the potential energy value of each spatial unit is fused with the gradient vector using a weighted vector summation formula to obtain the fused field value for each spatial unit. The formula for weighted vector summation is:

[0193]

[0194] The fusion weight setting can be combined with the core needs of the park's traffic scenarios. If the park management focuses more on traffic frequency, such as prioritizing guiding the target along the commonly used path, then the weight of the potential energy field is higher than that of the guiding gradient field, such as a potential energy field weight of 0.6 and a guiding gradient field weight of 0.4. If the focus is more on movement direction constraints, such as strictly following the lane direction and pedestrian walkway direction, then the weight of the guiding gradient field is higher than that of the potential energy field, such as a potential energy field weight of 0.4 and a guiding gradient field weight of 0.6. For areas where traffic frequency and direction constraints are equally important, the weight of each can be set to 0.5.

[0195] To verify the effectiveness of the fusion algorithm, for example, a spatial unit at a bend in the park... The normalized negative gradient vector of potential energy obtained through calculation The value (0.6, 0.8) represents the trend towards the upper right in the regression path; the normalized guided gradient vector... (0,1) represents the inertia of traveling due north along the road; if the weight combination emphasizing the direction of motion constraint is set as... It is 0.4. If the value is 0.6, then the fusion field value calculated by substituting it into the formula is... The value is (0.24, 0.92), indicating that the synthesized projection direction will maintain the dominant northward movement while being corrected by a 0.24 component to the right. This adjustment verifies that the algorithm can enable the digital twin to maintain its motion inertia during blind zone projection while being automatically constrained by the potential energy field within the historical high-frequency passage area. The fused field values ​​of all spatial units together constitute the historical passage potential energy field that takes into account both passage frequency and motion orientation.

[0196] Meanwhile, based on the type of target object, pedestrian potential energy fields and vehicle potential energy fields are constructed respectively. The pedestrian potential energy field focuses on adapting to the characteristics of pedestrian trajectories, while the vehicle potential energy field focuses on adapting to the characteristics of vehicle trajectories.

[0197] See Figure 3 The flowchart for virtual motion state deduction provided in this application embodiment includes steps S301 to S304, wherein:

[0198] S301, acquire spatial position data at least two moments before the spatial position data transmission was interrupted, and calculate the first velocity vector of the target object;

[0199] S302, Calculate the second velocity vector of the digital twin at its current position in the historical potential energy field;

[0200] S303, perform a weighted calculation on the first velocity vector and the second velocity vector to obtain the third velocity vector;

[0201] S304. Determine the simulation time step based on the sampling frequency of the spatial location data, calculate the displacement increment of the digital twin by combining the third velocity vector, and superimpose the displacement increment on the previous position of the digital twin to obtain the current simulation position.

[0202] When an interruption in spatial location data transmission is detected, the simulation module 30 initiates a virtual motion state simulation process. First, it retrieves valid spatial location data from at least two time points prior to the interruption, denoted as... and Combined with the sampling time interval of the sensing device The time interval between two positioning data points can be used to calculate the velocity vector of the target object before the interruption, i.e., the first velocity vector, using the finite difference formula. Preserve its direction of motion and speed information, and calculate its rate. .

[0203] The finite difference formula is as follows:

[0204]

[0205] The rate formula is:

[0206]

[0207] For example, if the target object's coordinates are (100, 200) before the interruption and (102, 201) at the interruption, and the sampling interval is 1 second, then the calculated first velocity vector is (2, 1) m / s, and the speed is approximately 2.24 m / s.

[0208] Next, based on the type of the target object, the corresponding historical traffic potential energy field is loaded. If it is a pedestrian, a pedestrian potential energy field is loaded; if it is a vehicle, a vehicle potential energy field is loaded, ensuring that the inference constraints match the target type. Then, based on the current position of the digital twin in the historical traffic potential energy field, the spatial cell to which that position belongs is determined, and the motion direction vector of that spatial cell is extracted. and guided gradient vector , which serves as the input data for the algorithm.

[0209] in, The average direction of big data statistics representing the historical travel trajectories in this area; This represents the negative gradient vector in the potential energy field, pointing from the high potential energy region to the low potential energy region. To eliminate dimensional differences between different data sources, and considering the potential energy value at that location in the potential energy field, the motion direction vector is first normalized. The calculation formula is as follows:

[0210]

[0211] in, The normalized direction vector of motion. It represents the magnitude of the vector.

[0212] Then, weights are assigned to both, and the normalized motion direction vector and the normalized steering gradient vector are summed using a weighted summation formula. By performing fusion calculations, the second velocity vector is obtained. The calculation formula is as follows:

[0213]

[0214] in, As a historical inertia weight, Gradient-guided weights, satisfying .

[0215] The historical inertia weight and gradient guidance weight can be set by combining the target object type and the function of the park area, and determined by a lookup method based on spatial unit attributes. Specifically, the system pre-marks the spatial units in the park into two categories: strongly constrained areas and weakly constrained areas. For strongly constrained areas such as driveways and corridors with clear path boundaries and where the target object must strictly follow the road direction, a higher gradient guidance weight is set, with specific values... It can be set to 0.2. It can be set to 0.8 to ensure the projected trajectory does not deviate from the road center. For weakly constrained areas such as squares and wide intersections where free passage is permitted and movement direction relies more on historical traffic statistics, a lower gradient guidance weight should be set, with the specific value to be determined. It can be set to 0.6. It can be set to 0.4.

[0216] For example, for pedestrian targets, in areas with clear traffic directions such as leisure walkways and office building corridors, the proportion of gradient guidance weight is increased to emphasize adherence to the established guidance rules of the area; in areas without clear traffic directions such as open squares, the proportion of historical inertia weight is increased to balance the individual movement trend of the target with the traffic patterns of the area. For vehicle targets, in areas with strict constraints on driving direction such as main roads and parking lot passages, the proportion of gradient guidance weight is significantly increased to ensure that vehicle movement conforms to the lane's prescribed direction; in areas with more flexible driving direction such as temporary parking areas, the proportion of historical inertia weight is appropriately increased to accommodate the needs of flexible vehicle movement.

[0217] It should be noted that, in order to avoid false displacement of the target object in a stationary or parked state due to the influence of the high flow velocity of the historical potential energy field, the system first executes the stationary state determination process before calculating the third velocity vector by weighting the first velocity vector and the second velocity vector. Specifically, the magnitude of the first velocity vector, i.e. the instantaneous velocity of the target object before the signal is interrupted, is calculated and compared with the preset stationary determination threshold.

[0218] The preset static determination threshold can be determined comprehensively based on the noise characteristics of the sensing device and the motion characteristics of the target object. Specifically, considering that GPS, UWB, or vision devices may still generate slight coordinate jumps, i.e., zero-bias noise, when the target is stationary, this threshold can be set to be greater than the maximum noise velocity value generated by the sensing device due to signal drift in a stationary state, effectively filtering out hardware noise interference.

[0219] The maximum noise velocity value is calculated by keeping the sensing device physically stationary during the system deployment phase or device initialization, continuously collecting positioning data for several time periods, calculating the pseudo displacement velocity between adjacent sampling points due to signal drift, and selecting the peak value in the statistical distribution of the pseudo displacement velocity or the upper limit of the high confidence interval based on the Gaussian distribution as the maximum noise velocity value.

[0220] Meanwhile, the static determination threshold is also related to the type of target object. For example, for motor vehicle targets, considering that they have rigid static characteristics after braking, the threshold is set relatively low to improve the sensitivity of starting detection; for pedestrian targets, considering that they often have slight limb movements when stationary, the threshold is set relatively high to accommodate non-displacement micro-disturbances and prevent false displacement of the digital twin.

[0221] If the magnitude of the first velocity vector is less than the preset static determination threshold, the target object is determined to be stationary. In this case, the guiding effect of the historical potential energy field is ignored, and the third velocity vector is directly set to zero, or the weight of the second velocity vector is forcibly adjusted to 0, so that the deduced position remains unchanged. If the magnitude of the first velocity vector is greater than or equal to the preset static determination threshold, the target object is determined to be in motion, and the weighted calculation step continues.

[0222] The third velocity vector is obtained by weighting the first and second velocity vectors. This step aims to resolve the conflict between the instantaneous inertia of the individual and the constraints of the environmental field at the moment of signal interruption, and constructs the following velocity fusion model formula:

[0223]

[0224] in, This means assigning the direction of the second velocity vector to the magnitude of the first velocity vector; To maintain the weight of individual inertia, the weight can be set according to the park's management needs. For example, when focusing on preserving individual movement trends, the first velocity vector has a higher weight; when focusing on conforming to common traffic habits, the second velocity vector has a higher weight. This weighted fusion takes into account both individual characteristics and group patterns. Alternatively, a classification and assignment method based on the target object type can be used. Specifically, when the target object is identified as a motor vehicle, considering the vehicle's large inertia and smooth turning process, the system focuses on maintaining the initial state before the interruption, and the weight can be set as follows: The value is 0.7; when the target object is identified as a pedestrian or non-motorized vehicle, considering that their movement direction is flexible and more easily guided by the environment, the system can set... It is 0.3.

[0225] For example, taking a signal interruption caused by a motor vehicle in a strongly restricted lane area as an example for numerical verification, the target is first identified as a motor vehicle, and then... The value is 0.7; the current location is identified as a strongly constrained region, and the value is set to 0.7. It is 0.2. It is 0.8. Assume the vehicle's first velocity vector before the interruption is... The coordinate (10,0) represents rapid travel in the due east direction; the normalized motion direction vector at this position is... The normalized guided gradient vector is (0.6, 0.8). The value is (0,1). Substitute the above parameters into the formula to calculate the second velocity vector. The value is (0.12, 0.96), indicating that the direction of the environmental field strength correction points due north. Substituting this into... Calculate the final derivation speed, i.e. The value is (7.36, 2.97). The calculation results show that the final obtained third velocity vector is... While maintaining the eastward inertia (reduced from 10 to 7.36), a northward velocity component of 2.97 was introduced. This resulted in the vehicle's projected trajectory within the blind spot exhibiting a natural and smooth turning curve, without directly hitting the wall or making a right-angle turn, demonstrating the effectiveness and feasibility of the weight setting method.

[0226] Next, based on the sampling frequency of the spatial location data in the acquisition module 10, the reciprocal of the sampling frequency is used as the basic time step, which is then fine-tuned according to the accuracy requirements of park management to determine the final projection time step. For example, if higher projection accuracy is required, the basic time step can be reduced to a fraction of its original value; if it is necessary to reduce system computing power consumption, the time step can be appropriately increased. The final projection time step is adapted to the data acquisition rhythm to avoid excessive deviation between the projection speed and the actual movement speed.

[0227] Then, the displacement increment is calculated to obtain the current simulated position. Combining the third velocity vector and the simulated time step, the displacement increment of the digital twin within that time step is calculated. The position of the digital twin at the previous moment is superimposed with the displacement increment to obtain the current simulated position.

[0228] The system calls the impassable boundary data from the preset static 3D model data in the construction module 20 to determine whether the initially obtained current simulation position is located within an impassable area. The ray method can be used for this determination. Specifically, a ray is drawn from the current simulation position in any direction, such as horizontally to the right. The number of intersections between the ray and the boundary of the impassable area is counted. If the number of intersections is odd, the position is determined to be within an impassable area; if it is even or zero, it is determined to be within a passable area.

[0229] If it is determined that the location is in an impassable area, the nearest projection point from the current simulation position to the boundary data of the impassable area is calculated, and the current simulation position is updated to the nearest projection point to avoid logical errors such as passing through walls or entering obstacles; if it is determined that the location is not in an impassable area, the initially obtained current simulation position remains unchanged.

[0230] The calculation of the nearest projection point involves first traversing all edges or faces of the inaccessible area boundary, calculating the perpendicular distance from the current projection position to each edge or face for each edge or face, finding the edge or face with the smallest distance, and then calculating the foot of the perpendicular from the current position to that edge or face through vector projection. This foot of the perpendicular is the nearest projection point.

[0231] The system continuously performs simulations at each time step to obtain consecutive simulation positions. To ensure the continuity of subsequent trajectory generation and data integrity, the system first constructs a trajectory point cache queue to store all valid simulation positions in chronological order. After completing the simulation for each time step and obtaining the corrected current simulation position, the position is immediately stored in the queue in chronological order to maintain the temporal continuity of the trajectory points.

[0232] Next, the virtual trajectory generation process is initiated. First, all stored simulation positions are extracted from the trajectory point cache queue and arranged chronologically to form a coordinate sequence. This coordinate sequence is a discrete set of spatial coordinates; directly displaying it would present a discontinuous trajectory, failing to reflect the continuous motion process. Using interpolation algorithms, such as cubic spline interpolation, the discrete coordinate sequence is fitted. Based on the spatial relationship and time interval between adjacent coordinate points, the coordinates of transition points between coordinate points are calculated, thus completing the discrete coordinate sequence into a continuous and smooth motion path, bridging the gaps between discrete points.

[0233] For example, in time The coordinates of the calculated position at that time are At the next sampling time The position coordinates at that time are In order to obtain The smooth transition points in the middle of time can be used to construct information about time. cubic spline function The formula is:

[0234]

[0235] Where c0, c1, c2, and c3 are the coefficients of the interpolation polynomial, which can be solved based on the continuity of the second derivative.

[0236] The time points that need to be filled in Substituting into the specific model formula above, calculate the components of the transition point position coordinates. Based on this result, the system calculates the values ​​at the original discrete points. and Specific complete coordinates are generated between them. By performing the above operation on all coordinate components, continuous smooth trajectory data including high-density intermediate interpolation points is output.

[0237] Then, configure the simulated state visual attributes for the motion path, including line type and color, to distinguish it from the visual attributes of the trajectory when the target object is traveling normally. For example, the line type of the simulated trajectory can be set to a dashed line or a dotted line, to distinguish it from the continuous solid line of the normal trajectory; the color can be set to a highly recognizable warning color, such as yellow or orange, to distinguish it from the normal color, such as green or blue, of the normal trajectory.

[0238] Finally, the motion path with the visual attributes of the deduced state is rendered into the digital twin scene as the virtual deduced trajectory of the digital twin, until the spatial location data is recovered or the target object leaves the blind zone.

[0239] Regarding the aforementioned correction module 40:

[0240] When the target object leaves the visual blind spot and the acquisition module 10 resumes acquiring spatial location data, there is often a discrepancy between the current projected position of the digital twin and the actual reconstructed position of the target object. Existing technologies often use forced position synchronization to directly update the position of the digital twin to the actual reconstructed position, causing visual abrupt changes such as instantaneous movement and jumps in the digital twin scene. This affects the operator's visual experience and judgment accuracy, and disrupts the continuity and smoothness of personnel and vehicle management.

[0241] The correction module 40 performs deviation convergence calculations and adaptively selects the correction method based on the magnitude of the distance deviation between the current simulated position and the actual reproduced position. When the deviation is small, the position is directly updated to ensure control efficiency; when the deviation is large, a smooth curve interpolation path is constructed and discretized into transition points, generating a smooth correction command to control the digital twin to move from the current simulated position to the actual reproduced position frame by frame.

[0242] In practical implementation, the first step is to extract the true reconstructed location. The correction module 40 monitors the data transmission status of the acquisition module 10 in real time. When it detects that spatial location data has been recovered, it immediately extracts the true geographical coordinates of the target object from the standardized data output by the acquisition module 10, uses this as the true reconstructed location, and records the timestamp corresponding to this location to ensure consistency with the time dimension of the projected location. Simultaneously, it acquires the current projected location and corresponding trajectory information of the digital twin in the projection module 30, providing basic data for deviation calculation.

[0243] Next, deviation convergence calculation is performed. Based on the coordinates of the current simulated position and the actual reproduced position, and using the unified local projection coordinate system of the park, the coordinates of the current simulated position are set as follows: The actual reconstructed location coordinates are The quantified positional deviation value can be obtained by calculating the straight-line distance between the two using the Euclidean distance formula. The calculation formula is:

[0244]

[0245] Then, the preset position deviation threshold is called. The threshold setting can be combined with the requirements of park management accuracy and visual smoothness. If the park management emphasizes accuracy and has high requirements for visual experience, the threshold setting should be more stringent; if it is a general monitoring scenario, the threshold can be appropriately relaxed. Based on the relationship between the Euclidean distance deviation and the position deviation threshold, deviation classification is performed. In specific verification, for motor vehicle targets on the main road of the park, considering the large size of the vehicles and the lane width of approximately 3.5 meters, deviations within the lane are acceptable, but deviations beyond the lane will cause visual collision misjudgments. Therefore, a threshold can be set... The distance is 3.0 meters; for pedestrian targets on sidewalks, due to their higher sensitivity to location, the distance is set to... It is 0.8 meters.

[0246] If the Euclidean distance deviation is determined to be less than the position deviation threshold, it means that the position difference between the two is small. A position update instruction is then directly generated to synchronize the position coordinates of the digital twin with the actual position.

[0247] If the Euclidean distance deviation is determined to be greater than or equal to the position deviation threshold, it indicates that the position difference between the two is large, and a smoothing correction instruction is generated to start the smooth transition correction process.

[0248] For example, after a vehicle is detected passing through a blind spot obstructed by a building, data recovery occurs. At this point, the system records the current projected location coordinates as (158.5, 204.2), while the extracted actual reconstructed location coordinates are (162.1, 206.8). Substituting these coordinate values ​​into the formula yields the location deviation value. Approximately 4.44 meters. The system automatically compares the calculated result of 4.44 meters with the preset vehicle deviation threshold of 3.0 meters to determine... > If the deviation is too large, a smoothing correction command is triggered to avoid the visual abrupt change caused by directly teleporting the vehicle from (158.5,204.2) to (162.1,206.8).

[0249] See Figure 4 The flowchart of the smooth correction instruction provided in the embodiments of this application includes steps S401 to S402, wherein:

[0250] S401, set a transition time window, and construct a smooth curve interpolation path based on the current simulation position as the starting point and the actual reproduction position as the ending point;

[0251] S402, based on the transition time window, the smooth curve interpolation path is discretized into several transition points, and the digital twin is controlled to move to the real reproduction position frame by frame.

[0252] In practice, a transition time window is first set, and the window duration is adjusted according to the needs of the park scenario. For example, in densely populated areas, the duration must be set to ensure that the movement of the digital twin does not affect the recognition of other targets. In open areas, the duration can be appropriately shortened to improve the correction efficiency. At the same time, the window duration must be sufficient to support smooth movement and avoid visual blurring caused by excessively fast movement.

[0253] Next, based on the current simulation position Starting point, accurately reproducing the location As the endpoint, select appropriate interpolation algorithms such as Bézier curve interpolation and cubic spline interpolation, and combine them with the type and motion characteristics of the target object, such as the flexibility of pedestrian movement and the straightness of vehicle movement, to construct a smooth curve interpolation path, avoid impassable areas, and conform to the actual road network and traffic logic of the park.

[0254] To ensure that the corrected path conforms to the motion characteristics of the target object, such as the tangential inertia of a vehicle, this embodiment can use a third-order Bézier curve as the interpolation algorithm. This algorithm introduces two control points. and This constrains the tangential direction of the path, ensuring that the generated trajectory remains tangentially continuous with the projected velocity direction and the actual velocity direction at the start and end points, respectively, thus conforming to vehicle dynamics.

[0255] The mathematical expression for a third-order Bézier curve is as follows:

[0256]

[0257] in, The normalized time parameter changes from 0 to 1 as the transition time window progresses; The starting coordinates; The coordinates of the endpoint; and For control points, the calculation method is as follows:

[0258]

[0259]

[0260] in, To deduce the terminal velocity vector, To accurately reproduce the initial velocity vector, This is a smoothing coefficient that is adaptively adjusted based on distance, and is usually taken as 1 / 3 of the distance.

[0261] Subsequently, based on the transition time window and the rendering frame rate of the digital twin scene, the smooth curve interpolation path is discretized into several consecutive transition points. The coordinates of each transition point are arranged in chronological order to form a sequence of transition points.

[0262] The correction module 40 controls the digital twin to move frame by frame according to the transition point sequence based on the generated smoothing correction instructions. During the movement, the correction module 40 monitors the position status of the digital twin in real time, ensuring that it moves strictly along the smooth curve interpolation path. At the same time, the correction module 40 continuously receives the latest real-time position data uploaded by the acquisition module 10.

[0263] To prevent the smoothing correction process from being ineffectively interrupted due to signal noise from the positioning device, such as GPS drift, the system sets a repositioning deviation threshold. This repositioning deviation threshold is set based on the physical dimensions of the target object and the error range of the current positioning device, employing a dynamic adjustment mechanism. Specifically, firstly, the physical characteristic dimensions of the target object, such as vehicle length or pedestrian width, are obtained as basic reference values; secondly, the type of positioning device used to collect the current location information is determined, along with the corresponding error range; finally, the final repositioning deviation threshold is determined by combining the basic reference value and the error range.

[0264] For example, in outdoor areas with good signal, when using GPS positioning and the target is a motor vehicle, considering its large physical size and the meter-level error of the GPS device itself, the threshold can be set to 2.0 meters to 3.0 meters; while in indoor areas where high-precision UWB base stations are deployed, when using UWB positioning and the target is a pedestrian, considering the small size of the pedestrian and the centimeter-level accuracy of the UWB device, the threshold can be set to 0.5 meters to 0.8 meters.

[0265] The specific judgment logic is as follows: calculate the distance difference between the latest acquired real-time position and the target endpoint position of the currently executed smoothing correction instruction. If the distance difference is less than the repositioning deviation threshold, it is determined that the difference is within the allowable fluctuation range, the correction process is not interrupted, and the digital twin continues to move towards the original target endpoint to ensure visual smoothness; if the distance difference is greater than or equal to the repositioning deviation threshold, it is determined that the target object has undergone a substantial change in motion state, such as a sudden change in motion direction or acceleration. At this time, the current correction process is immediately interrupted, the latest acquired real-time position is taken as the new true reproduction position, and the deviation convergence calculation is re-executed to generate a new smoothing correction instruction.

[0266] When the digital twin moves to the last transition point, which is the actual reproduction position, the correction process ends, the digital twin returns to the normal real-time mapping state, and is subsequently updated synchronously with the actual position acquired by the acquisition module 10.

[0267] Those skilled in the art will understand that, in the methods described above in the specific embodiments, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic. It should be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0268] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0269] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A digital twin-based park pedestrian and vehicle access control system, characterized in that: include: The data acquisition module is used to obtain spatial location data of target objects within the park. A construction module is used to load the pre-set static 3D model data of the park and, in combination with the spatial location data, generate a digital twin that maps to the target object; The simulation module is used to respond to the detection of the interruption of the spatial location data transmission, to perform virtual motion state simulation of the digital twin based on the historical traffic potential energy field, update the current simulation position of the digital twin, and generate a virtual simulation trajectory; the historical traffic potential energy field is a virtual field constructed based on the historical traffic trajectory data of the park; Based on the historical prevailing potential energy field, the virtual motion state of the digital twin is simulated, including: Acquire spatial position data at least two moments before the interruption of the spatial position data transmission, and calculate the first velocity vector of the target object; Calculate the second velocity vector of the digital twin at its current position in the historical travel potential energy field; The first velocity vector and the second velocity vector are weighted and calculated to obtain the third velocity vector; The simulation time step is determined based on the sampling frequency of the spatial location data. The displacement increment of the digital twin is calculated by combining the third velocity vector. The displacement increment is then superimposed on the previous position of the digital twin to obtain the current simulation position. The generation of the virtual simulation trajectory includes: Construct a trajectory point cache queue to store the current estimated position after calculation of the historical passable potential energy field and correction of impassable areas in a time series. Extract the coordinate sequence from the trajectory point cache queue, and fit the discrete coordinate sequence into a continuous motion path using an interpolation algorithm; Configure the motion path with the deduction state visual attributes, and render the motion path with the deduction state visual attributes into the digital twin scene as the virtual deduction trajectory of the digital twin. The correction module is used to respond to the detection of the recovery of the spatial location data, extract the true reproduction position of the target object, perform deviation convergence calculation based on the true reproduction position and the current simulation position, generate a smooth correction instruction, and control the digital twin to move from the current simulation position to the true reproduction position.

2. The digital twin park pedestrian and vehicle access control system according to claim 1, characterized in that, Generating a digital twin mapped to the target object includes: Identify the type of the target object, including pedestrians, motor vehicles, and non-motor vehicles; Retrieve the corresponding 3D model template from the preset model library according to the type; Convert the latitude and longitude coordinates in the spatial location data into local coordinates under the three-dimensional coordinates of the park; The local coordinates are assigned to the 3D model template, and then rendered in the preset static 3D model data to generate a digital twin.

3. The digital twin park pedestrian and vehicle access control system according to claim 2, characterized in that, The pre-set static 3D model data includes: the building information model of the park, the topological structure data of the park's road network, and the boundary data of inaccessible areas.

4. The digital twin park pedestrian and vehicle access control system according to claim 3, characterized in that, Updating the current projected position of the digital twin includes: Call the boundary data of the impassable area to determine whether the current simulated location is located in the impassable area; In response to the determination that the location is within the impassable area, the nearest projection point from the current projection location to the boundary data is calculated, and the current projection location is updated to the nearest projection point.

5. The digital twin park pedestrian and vehicle access control system according to claim 1, characterized in that, The construction process of the historical prevailing potential energy field includes: Obtain historical passage trajectory data within a preset time period in the park; The historical passage trajectory data is processed into a grid, dividing the park space into multiple discrete spatial units; Statistically analyze the trajectory point density and motion direction vector within each spatial unit; A potential energy field is generated based on the trajectory point density using a density estimation algorithm, and a guiding gradient field is generated by combining the motion direction vector. The historical passage potential energy field is then constructed by fusing these two methods.

6. The digital twin park pedestrian and vehicle access control system according to claim 5, characterized in that, The historical potential energy field includes the pedestrian potential energy field and the vehicle potential energy field; According to the type of the target object, the corresponding potential energy field is loaded; in response to the target object being a pedestrian, the pedestrian potential energy field is loaded; in response to the target object being a vehicle, the vehicle potential energy field is loaded.

7. The digital twin park pedestrian and vehicle access control system according to claim 1, characterized in that, The execution deviation convergence calculation and generation of smoothing correction instructions include: Calculate the Euclidean distance deviation between the current simulated position and the actual reproduced position; In response to the Euclidean distance deviation being less than the position deviation threshold, the position coordinates of the digital twin are updated to the actual reproduced position; In response to the Euclidean distance deviation being greater than or equal to the position deviation threshold, a smoothing correction command is generated.

8. The digital twin park pedestrian and vehicle access control system according to claim 7, characterized in that, The smoothing correction instruction includes: Set a transition time window, and construct a smooth curve interpolation path based on the current simulation position as the starting point and the actual reproduction position as the ending point; Based on the transition time window, the smooth curve interpolation path is discretized into multiple transition points, and the digital twin is controlled to move to the real reproduction position frame by frame.

Images