A multi-source fusion-based underground parking lot navigation method and system

Abstract

This application relates to a navigation method and system for underground parking lots based on multi-source fusion. The method includes acquiring a 3D building information model of the underground parking lot and constructing a 3D navigation topology map based on the 3D building information model; collecting multi-source data, performing preprocessing operations on the multi-source data to obtain preliminary multi-source positioning results; evaluating the quality indicators of the preliminary multi-source positioning results to obtain multi-source positioning quality evaluation results, and performing dynamic weighted fusion calculations based on the quality evaluation results to obtain 3D positioning results representing the user's real-time location; and performing path planning based on the 3D positioning results and the user's target location, combined with the 3D navigation topology map, to obtain the target navigation path. This application improves the accuracy of navigation in underground parking lots.

Description

Technical Field

[0001] This application relates to the technical field of intelligent parking navigation, and in particular to a navigation method and system for underground parking lots based on multi-source fusion. Background Technology

[0002] Currently, underground parking lots have complex spatial structures and enclosed environments, and traditional navigation methods mostly rely on a single positioning source (such as GPS or Wi-Fi) and two-dimensional map data for route planning. However, the severe signal obstruction and complex three-dimensional structure in underground environments, coupled with factors such as mixed pedestrian and vehicle traffic and dynamic access control, result in significant shortcomings in existing navigation technologies in terms of positioning accuracy, route adaptability, and traffic efficiency.

[0003] Most existing underground parking lot navigation solutions are based on static two-dimensional maps for path planning, lacking structural modeling and application of three-dimensional spatial component information, and failing to effectively express key navigation elements such as floor differences and ramp elevations.

[0004] The existing technical solutions mentioned above have the following drawbacks: existing positioning solutions usually lack a multi-source data fusion mechanism, resulting in significant positioning errors in signal blind spots or high interference areas, which affects the user's navigation experience, and therefore there is room for improvement. Summary of the Invention

[0005] To improve the accuracy of navigation in underground parking lots, this application provides a navigation method and system for underground parking lots based on multi-source fusion.

[0006] The above-mentioned objective of this application is achieved through the following technical solution:

[0007] A navigation method for underground parking lots based on multi-source fusion, the method comprising:

[0008] A three-dimensional building information model of an underground parking lot is obtained, and a three-dimensional navigation topology map is constructed based on the three-dimensional building information model. The three-dimensional navigation topology map includes three-dimensional coordinate information representing spatial nodes and accessibility attributes of connecting paths.

[0009] Collect multi-source data, perform preprocessing operations on the multi-source data to obtain the initial solution of multi-source positioning, wherein the multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot;

[0010] The initial solution of the multi-source positioning is evaluated by quality indicators to obtain the multi-source positioning quality evaluation result. Based on the quality evaluation result, a dynamic weighted fusion operation is performed to obtain the three-dimensional positioning result used to represent the user's real-time location.

[0011] Based on the three-dimensional positioning results and the user's target location, a path is planned using the three-dimensional navigation topology map to obtain the target navigation path.

[0012] By adopting the above technical solutions, and by acquiring a 3D building information model of the underground parking lot and constructing a 3D navigation topology map containing the 3D coordinates and access attributes of spatial nodes, spatial mapping and path modeling of complex parking lot structures can be achieved, thus providing a high-precision structural foundation for navigation path planning. By collecting multi-source data consisting of camera images, user inertial information, and wireless signals, and performing preprocessing operations, the limitations of a single signal source can be overcome, improving data integrity and availability, thereby enhancing the robustness of the positioning algorithm. By performing quality index evaluation on the initial solution of multi-source positioning and performing dynamic weighted fusion calculation based on the quality evaluation results, the data weights can be dynamically adjusted according to the confidence level, improving the accuracy and stability of fusion positioning, thereby achieving high-precision 3D positioning of the user's location. By performing path planning on the 3D navigation topology map based on the 3D positioning results and target location, a navigation path with strong structural adaptability and high positioning continuity can be output, thus significantly improving the navigation experience for users in underground parking lots.

[0013] In one example, this application can be further configured such that: the construction of a three-dimensional navigation topology map based on the three-dimensional building information model specifically includes:

[0014] Structural component information of the underground parking lot is extracted from the three-dimensional building information model. The structural component information includes floor structure, ramp location, entrance and exit space, elevator shaft and passage outline data.

[0015] Based on the structural component information, the building space is partitioned and modeled, and the passage area in the underground parking lot is discretized into multiple spatial nodes. Each spatial node contains three-dimensional coordinates, the floor number it is located on, and the corresponding structural attribute information.

[0016] Establish path connections between identified spatial nodes. These path connections represent the topological connectivity of the travel paths. Mark the corresponding height difference, travel direction, slope information, and structural travel restrictions on each path connection.

[0017] By adopting the above technical solutions, and extracting structural component information such as floor structure, ramps, and entrances from the 3D building information model, a digital model foundation that is highly consistent with the real physical environment can be established, thereby ensuring that the navigation path conforms to the building constraints. By partitioning the building space into models and discretizing the passage area into multiple spatial nodes with structural attributes, complex spatial relationships can be expressed in detail, thereby improving the structural adaptability of path planning. By establishing path connection relationships between spatial nodes and labeling passage attributes such as height difference, slope, and passage restrictions, a comprehensive 3D topological information network can be constructed, thereby laying an accurate connected graph foundation for subsequent path search and dynamic navigation.

[0018] In one example, this application can be further configured such that: the preprocessing operation on the multi-source data to obtain the preliminary solution of multi-source localization specifically includes:

[0019] Image enhancement and target detection operations are performed on video images captured by cameras inside the underground parking lot to identify user or vehicle targets and obtain their two-dimensional image coordinates in the image plane. Based on the two-dimensional image coordinates and camera spatial calibration information, the initial solution for image localization is calculated.

[0020] Resampling and interference filtering operations are performed on the inertial measurement information of the user terminal to extract the acceleration and angular velocity data of the stable segment, and the user displacement trajectory is estimated based on the inertial navigation algorithm to obtain the initial solution of inertial positioning.

[0021] Signal filtering and noise suppression operations are performed on the wireless signal data generated by the wireless communication equipment deployed in the underground parking lot to extract the effective received signal strength index and base station identifier, and the initial solution of wireless signal positioning is calculated based on the signal strength inversion algorithm.

[0022] By adopting the above technical solutions, image enhancement and target detection are performed on camera images to extract the image coordinates of users or vehicles. Combined with camera calibration information, the position is calculated, enabling accurate visual positioning within the visible range and providing important two-dimensional spatial references. By resampling and filtering the inertial information of user terminals and calculating the displacement trajectory based on inertial navigation algorithms, continuous positioning capability can be maintained under visual failure or signal blind spots, ensuring the positioning chain remains unbroken. By filtering and suppressing wireless signal data and using RSSI inversion algorithms to obtain position information, positioning errors caused by inertial drift and visual occlusion can be compensated for, thereby improving the overall robustness and coverage of the positioning system.

[0023] In one example, this application can be further configured such that: the quality index evaluation of the preliminary multi-source localization solution results specifically includes:

[0024] An image recognition quality assessment is performed on the initial image localization solution. The image recognition quality assessment includes image sharpness assessment, target recognition confidence calculation and target bounding box stability analysis to obtain the image recognition quality assessment result.

[0025] An inertial solution quality assessment is performed on the initial inertial positioning solution. The inertial solution quality assessment includes measurements of acceleration variation amplitude, angular velocity curve continuity, and cumulative drift error, to obtain the inertial solution quality assessment result.

[0026] A signal quality assessment is performed on the initial solution of wireless signal positioning. The signal quality assessment includes the mean amplitude of the received signal strength RSSI, the hopping rate, and the number of signal sources, to obtain the signal quality assessment result.

[0027] By adopting the above technical solutions, the image positioning quality can be quantified by evaluating image sharpness, calculating target recognition confidence, and analyzing bounding box stability of the image positioning results, thus providing dynamic confidence indicators for subsequent weighted fusion. By evaluating the acceleration amplitude, angular velocity continuity, and drift error of the inertial positioning results, the solution quality and availability can be accurately reflected, thereby avoiding the impact of low-quality inertial navigation data on the final fusion positioning accuracy. By evaluating the mean RSSI, hopping rate, and number of signal sources of the wireless signal results, the stability of the signal environment and the reliability of positioning can be judged in real time, thereby enhancing the system's adaptability to complex channel conditions.

[0028] In one example, this application can be further configured as follows: the dynamic weighted fusion operation performed based on the quality evaluation result to obtain a three-dimensional positioning result representing the user's real-time location specifically includes:

[0029] Based on the image recognition quality assessment results, the inertial calculation quality assessment results, and the signal quality assessment results, a location source confidence index is constructed, and the location source confidence index is normalized and used as a weighting coefficient.

[0030] The initial solution results of image localization, inertial localization, and wireless signal localization are input into a pre-trained multi-source fusion algorithm model, and a weighted calculation operation is performed based on the weighting coefficients to obtain the three-dimensional localization result.

[0031] By adopting the above technical solution, and constructing a positioning reliability index based on the quality assessment results of each source and normalizing it into a weighted coefficient, the contribution weight of each data source can be dynamically measured before fusion calculation, thereby improving the adaptability and interpretability of the fusion process. By inputting various preliminary solution results into the trained multi-source fusion model and performing weighted calculation, the complementary advantages of each information source can be fully utilized to achieve more accurate, stable, and continuous three-dimensional positioning results, thereby improving the core performance of the entire navigation system.

[0032] In one example, this application can be further configured as follows: based on the three-dimensional positioning result and the user's target location, and combined with the three-dimensional navigation topology map, path planning is performed to obtain the target navigation path, specifically including:

[0033] The navigation starting node is determined based on the three-dimensional positioning results, and the navigation target node is determined based on the user's target location. Both the starting node and the target node correspond to spatial nodes in the three-dimensional navigation topology map.

[0034] A path search operation is performed in the three-dimensional navigation topology map to obtain path search results. The path search operation evaluates the path between nodes based on a preset cost function, which includes path length, height difference, passage restrictions and structural adaptability factors.

[0035] Based on the path search results, a node sequence is generated as the spatial structure sequence of the navigation path. The node sequence is then subjected to curvature smoothing and spatial coherence correction to form the target navigation path. The target navigation path includes continuous three-dimensional spatial segment information and labels the spatial coordinates, height attributes, and travel direction information of the path nodes.

[0036] By adopting the above technical solutions, the navigation start node and target node are determined based on the 3D positioning results and the user's target location, enabling accurate mapping of the navigation path on the 3D topology map, thereby avoiding floor errors or path breaks. By performing path search based on cost factors such as path length, height difference, and access restrictions, a high-quality path that balances shortest distance, accessibility, and structural adaptability can be output, thereby optimizing access efficiency and user experience. By performing curvature smoothing and spatial continuity correction on the node sequence, the visual aesthetics and navigation continuity of the path can be improved, resulting in a more natural navigation guidance effect in the 3D display interface.

[0037] In one example, this application can be further configured as follows: the underground parking lot navigation method based on multi-source fusion also includes:

[0038] During the execution of the navigation route, real-time traffic status information of the target navigation route is continuously collected. The traffic status information includes the degree of vehicle congestion, traffic blockage status and user movement behavior characteristics of the corresponding road segments in the route. The traffic status data is input into a preset route traffic scoring model and the scoring result representing the traffic weight of each route segment is output.

[0039] When the scoring result is detected to be lower than the preset scoring threshold, a path search operation is performed based on the three-dimensional topology map, and the updated navigation path is sent to the navigation terminal as an alternative path to replace the content displayed on the current path.

[0040] In the navigation terminal, the three-dimensional spatial positioning result and the updated navigation path are bound to the three-dimensional building information model, and the rendering process of the three-dimensional navigation interface is executed. The rendering process includes three-dimensional path trajectory drawing and spatial orientation auxiliary display.

[0041] By adopting the above technical solutions, and continuously collecting traffic status information such as vehicle congestion, traffic blockage, and user movement behavior during the route execution process, and inputting it into the route scoring model for evaluation, dynamic perception and judgment of the route status can be achieved, thereby enhancing the real-time performance and environmental adaptability of route planning. By re-searching for a route and replacing the current route when the scoring result is below a threshold, impassable or inefficient route segments can be avoided, thereby ensuring the executability and timeliness of navigation. By binding the 3D positioning results and the updated route to a 3D building model and performing rendering on the navigation terminal, dynamic fusion and display of spatial positioning and navigation information can be achieved, thereby improving the user's understanding of the navigation route and current location and the intuitiveness of operation.

[0042] The second objective of this invention is achieved through the following technical solution:

[0043] A multi-source fusion-based underground parking lot navigation system, comprising:

[0044] A 3D modeling module is used to acquire a 3D building information model of an underground parking lot and construct a 3D navigation topology map based on the 3D building information model. The 3D navigation topology map includes 3D coordinate information representing spatial nodes and accessibility attributes of connecting paths.

[0045] The multi-source data acquisition and preprocessing module is used to acquire multi-source data, perform preprocessing operations on the multi-source data, and obtain the initial solution of multi-source positioning. The multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot.

[0046] The quality assessment and fusion module is used to perform quality index assessment on the initial solution of the multi-source positioning to obtain the multi-source positioning quality evaluation result, and to perform dynamic weighted fusion operation based on the quality evaluation result to obtain the three-dimensional positioning result representing the user's real-time location.

[0047] The path planning module is used to plan a path based on the three-dimensional positioning results and the user's target location, combined with the three-dimensional navigation topology map, to obtain the target navigation path.

[0048] By adopting the above technical solutions, and by acquiring a 3D building information model of the underground parking lot and constructing a 3D navigation topology map containing the 3D coordinates and access attributes of spatial nodes, spatial mapping and path modeling of complex parking lot structures can be achieved, thus providing a high-precision structural foundation for navigation path planning. By collecting multi-source data consisting of camera images, user inertial information, and wireless signals, and performing preprocessing operations, the limitations of a single signal source can be overcome, improving data integrity and availability, thereby enhancing the robustness of the positioning algorithm. By performing quality index evaluation on the initial solution of multi-source positioning and performing dynamic weighted fusion calculation based on the quality evaluation results, the data weights can be dynamically adjusted according to the confidence level, improving the accuracy and stability of fusion positioning, thereby achieving high-precision 3D positioning of the user's location. By performing path planning on the 3D navigation topology map based on the 3D positioning results and target location, a navigation path with strong structural adaptability and high positioning continuity can be output, thus significantly improving the navigation experience for users in underground parking lots.

[0049] In summary, this application includes the following beneficial technical effects:

[0050] 1. By acquiring a 3D building information model of an underground parking lot and constructing a 3D navigation topology map containing the 3D coordinates and access attributes of spatial nodes, spatial mapping and path modeling of complex parking lot structures can be achieved, thus providing a high-precision structural foundation for navigation path planning; by collecting multi-source data consisting of camera images, user inertial information, and wireless signals, and performing preprocessing operations, the limitations of a single signal source can be overcome, data integrity and availability can be improved, thereby enhancing the robustness of the positioning algorithm;

[0051] 2. By performing quality index evaluation on the initial solution of multi-source positioning and performing dynamic weighted fusion calculation based on the quality evaluation results, the data weights can be dynamically adjusted according to the confidence level, thereby improving the accuracy and stability of fusion positioning and achieving high-precision three-dimensional positioning of the user's location. By performing path planning on the three-dimensional navigation topology map based on the three-dimensional positioning results and target location, a navigation path with strong structural adaptability and high positioning continuity can be output, thereby significantly improving the navigation experience of users in underground parking lots. Attached Figure Description

[0052] Figure 1 This is a flowchart of a navigation method for underground parking lots based on multi-source fusion in one embodiment of this application;

[0053] Figure 2 This is a flowchart illustrating the implementation of step S10 in a multi-source fusion-based underground parking lot navigation method according to an embodiment of this application.

[0054] Figure 3This is a flowchart illustrating the implementation of step S20 in a multi-source fusion-based underground parking lot navigation method according to an embodiment of this application.

[0055] Figure 4 This is a flowchart illustrating the implementation of step S30 in a multi-source fusion-based underground parking lot navigation method according to an embodiment of this application.

[0056] Figure 5 This is another implementation flowchart of step S30 in an underground parking lot navigation method based on multi-source fusion in one embodiment of this application;

[0057] Figure 6 This is a flowchart illustrating the implementation of step S40 in a multi-source fusion-based underground parking lot navigation method according to an embodiment of this application.

[0058] Figure 7 This is another implementation flowchart of a navigation method for underground parking lots based on multi-source fusion in one embodiment of this application;

[0059] Figure 8 This is a principle block diagram of an underground parking lot navigation system based on multi-source fusion in one embodiment of this application. Detailed Implementation

[0060] The present application will be further described in detail below with reference to the accompanying drawings.

[0061] In one embodiment, such as Figure 1 As shown, this application discloses a navigation method for underground parking lots based on multi-source fusion, which specifically includes the following steps:

[0062] S10: Obtain the 3D building information model of the underground parking lot, and construct a 3D navigation topology map based on the 3D building information model. The 3D navigation topology map includes the 3D coordinate information of spatial nodes and the accessibility attributes of connecting paths.

[0063] Specifically, the drawing-to-model interface is called to import BIM model data files. The model contains the structural layout, dimensional attributes, and spatial relationships of building elements such as parking spaces, driveways, columns, elevators, exits, and ramps. Based on this 3D model, a spatial analysis operation is performed to extract the passable area and divide it into several 3D spatial nodes. Each spatial node has its 3D coordinates determined by its geometric center or marker point. At the same time, the physical connection relationships between nodes are traversed to generate connection paths. The connection paths record the passage direction, minimum width, height limit, or slope information between adjacent nodes to construct a 3D navigation topology map for navigation calculation. This topology map has both navigation feasibility analysis and spatial positioning adaptation capabilities, and can support subsequent path search and 3D rendering operations. For example, in a B1 level parking area, the path from node N1 to N2 records its floor number as -1, direction as north, slope as 5%, height limit as 2.2 meters, and node spacing as 7 meters.

[0064] S20: Collect multi-source data, perform preprocessing operations on the multi-source data, and obtain the initial solution of multi-source positioning. The multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot.

[0065] Specifically, multiple data acquisition channels are accessed to process the video streams transmitted from the video surveillance equipment inside the underground parking lot into timed screenshots and upload image data. Acceleration and angular velocity data collected in real time by the IMU sensor built into the user terminal are encapsulated into data packets with fixed time intervals and uploaded. At the same time, the signal strength index RSSI and device ID of wireless communication devices deployed in the site, such as Wi-Fi hotspots or Bluetooth beacons, are obtained. During the preprocessing operation, the image data is subjected to resolution adaptation, image cropping and enhancement processing, the inertial measurement data is subjected to deduplication, time alignment and interference segment removal, and the wireless signal data is subjected to weak signal filtering, device outlier identification and intensity normalization. Finally, a set of fused input data for positioning calculation is formed, which includes the preliminary solution of multi-source positioning consisting of image recognition coordinates, inertial trajectory segments and wireless signal field strength. For example, if the image target collected in a certain area is identified as "license plate BX001", its corresponding inertial data is the velocity curve segment within 0.5 seconds, and the RSSI of the three nearest base stations for the wireless signal is -68, -72 and -75 respectively.

[0066] S30: Perform quality index evaluation on the initial solution of multi-source positioning to obtain the multi-source positioning quality evaluation result, and perform dynamic weighted fusion operation based on the quality evaluation result to obtain the three-dimensional positioning result used to represent the user's real-time location.

[0067] Specifically, the built-in quality evaluation function is invoked to assess the clarity and confidence of the identified target in the image positioning data, the continuity and stability of the inertial data, and the intensity fluctuation range of the wireless signal. Through standardization processing, the quality indicators of the three signal sources are converted into confidence values ​​of a unified scale. Then, multiple initial solution results are fused based on the confidence values ​​and weights. Finally, a three-dimensional positioning point representing the user's current location is output, which includes the X, Y, and Z axis coordinate values ​​and the positioning confidence level. For example, if the confidence of the image signal decreases due to occlusion at a certain moment, the fusion process will automatically increase the weighting ratio of the inertial and wireless signals to ensure that the output three-dimensional positioning result remains relatively stable and accurate in the current environment.

[0068] S40: Based on the 3D positioning results and the user's target location, and combined with the 3D navigation topology map, path planning is performed to obtain the target navigation path.

[0069] Specifically, based on the obtained 3D positioning results, the starting node number is extracted, and the target location selected by the user is mapped to the corresponding target node number in the 3D navigation topology map. Path search algorithms such as A* and Dijkstra are called. During the search process, cost factors such as path length between nodes, floor height difference, travel direction, and structural obstacles are comprehensively considered to calculate the candidate sequence of the lowest cost path from the starting point to the end point. After the path is generated, path smoothing processing is also required, such as optimizing the continuity of the turning angles between nodes using B-splines or Bezier curves to avoid the path being jagged or having spatial jumps. At the same time, the 3D spatial coordinates, turning angle, and slope of each path segment are labeled to ensure that the subsequent navigation rendering interface is synchronized with the user's orientation. For example, when driving from the southeast corner of B2 to the middle exit of B1, the selected path includes floor changes and ramp sections, ultimately forming a navigation trajectory composed of 8 continuous path segments.

[0070] In one embodiment, such as Figure 2 As shown, in step S10, which involves constructing a 3D navigation topology map based on the 3D building information model, the specific steps include:

[0071] S11: Extract structural component information of the underground parking lot from the 3D building information model. The structural component information includes floor structure, ramp location, entrance and exit space, elevator shaft and passage outline data.

[0072] Specifically, the structure of the BIM model or IFC file is analyzed to locate and extract component categories related to spatial access, including floor slab components, entrance and exit label components, spatial enclosure structures of stairwells and elevator shafts, as well as ramps, passageways, and partition wall components closely related to access. During the extraction of component information, component classification and coordinate boundary extraction operations are performed. The actual distribution range and spatial orientation of a component in three-dimensional space are determined by its type and its spatial geometric attributes in the model. For example, when extracting a ramp component in B2, its bottom starting point Z-axis height is recorded as -3.2 meters, the ending point Z-axis height is -2.6 meters, the horizontal width is 3.2 meters, the component type is a unidirectional ramp, and the floor labels and coordinate areas connected to its two ends are marked, thus providing a structural basis for subsequent path modeling.

[0073] S12: Based on structural component information, the building space is partitioned and modeled, and the passage area in the underground parking lot is discretized into multiple spatial nodes. Each spatial node contains three-dimensional coordinates, the floor number it is located on, and the corresponding structural attribute information.

[0074] Specifically, spatial projection and connectivity analysis are performed on the passable areas. First, the driveway area and pedestrian passage on each floor are projected from the 3D building information model to a 2D plane to extract the passable contours. Then, based on the contours, mesh discretization and node division are performed. Each passable area is divided into several regular or irregular shaped spatial units. The geometric center of each spatial unit is set as a spatial node. This spatial node records its floor number, 3D spatial coordinates (X,Y,Z), and structural attribute information such as ramp entrance, nearby exit, or whether it is near a wall. For example, on floor B1, the node numbered N_101 is located 3 meters north of exit E2, which is a gentle slope area. Its Z-axis is marked as -2.8 meters, and its structural attributes include "height limit 2.1 meters, direction east". This operation realizes the transformation from structural components to navigation data structure.

[0075] S13: Establish path connection relationships between the identified spatial nodes. The path connection relationship represents the topological connectivity of the passage path. Mark the corresponding height difference, passage direction, slope information and structural passage restrictions on each path connection.

[0076] Specifically, based on the dual constraints of spatial distance and structural accessibility, the Euclidean distance of each pair of spatial nodes is calculated and it is determined whether they are within the adjacent access area. Then, combined with the information of structural components, it is determined whether the path is blocked by factors such as walls, railings, or slopes exceeding the safety threshold. If the connectivity conditions are met, a directed or undirected path is established between the nodes, and structural attribute annotations are added to the path, including the height difference between the starting and ending nodes, the slope of the path, the direction of passage (one-way / two-way), height restrictions, obstacle types, and other information. For example, when establishing a path between nodes N_205 and N_206 in B2, the system records that the path length is 5.5 meters, the slope is 4.6%, the height difference at the starting point is 0.3 meters, the direction is a one-way north-south line, and an "adjacent column" restriction label is added for feasibility and access cost assessment during subsequent path planning.

[0077] In one embodiment, such as Figure 3 As shown, in step S20, preprocessing is performed on the multi-source data to obtain the initial solution for multi-source localization, specifically including:

[0078] S21: Perform image enhancement and target detection operations on the video images captured by the cameras inside the underground parking lot, identify user or vehicle targets and obtain their two-dimensional image coordinates in the image plane, and calculate the initial solution of image localization based on the two-dimensional image coordinates and camera spatial calibration information.

[0079] Specifically, firstly, feature recognition is used to determine the video image corresponding to the user or vehicle. Assuming a user enters the "B2 South Entrance" with a mobile phone, this area is covered by "Camera_08". The camera reports images to the backend recognition system every second, while the user terminal also reports inertial navigation data and RSSI values. The backend matches a moving target A in the image, whose displacement direction and speed are consistent with the inertial navigation estimation, and whose vehicle color matches the user's pre-registered information. Therefore, this image is considered the user's current location. For vehicles, the video image containing the vehicle with the corresponding license plate information is the user's current location. Subsequently, during the image enhancement and target detection operations performed on the video images captured by the cameras inside the underground parking lot, brightness equalization, edge sharpening, and noise reduction are first applied to the video frames to improve image clarity and target distinguishability. Then, a pre-trained deep learning model is used. For example, YOLOv5 automatically detects and labels vehicles or pedestrians in video images. It extracts the centroid coordinates of the target from the detection box as two-dimensional image coordinates, and projects the two-dimensional image coordinates into corresponding three-dimensional spatial coordinates according to the spatial calibration parameter matrix of the camera to obtain the initial solution of image localization. When performing the conversion, the intrinsic and extrinsic parameter matrices of the camera are referenced, and the error is corrected by combining the viewing angle parameters and installation position. The corresponding timestamp is recorded. For example, for a red car, its image coordinates are at pixel (480, 620), which corresponds to a position of (X=8.2, Y=4.3, Z=-2.1) meters in physical three-dimensional space. This image localization result can be used for subsequent fusion with other data to improve the localization stability and robustness.

[0080] S22: Perform resampling and interference filtering operations on the inertial measurement information of the user terminal, extract the acceleration and angular velocity data of the stable segment, and estimate the user's displacement trajectory based on the inertial navigation algorithm to obtain the initial solution of inertial positioning.

[0081] Specifically, during the resampling and interference filtering process of the inertial measurement information of the user terminal, the sampling frequency of different terminal models is first unified to a fixed standard. Then, the sudden interference data caused by hand tremors, environmental vibrations, etc., during the acquisition process is removed by means of sliding window mid-range filtering, bandpass filtering, etc. During the extraction of stable segments, the rate of change of acceleration and the continuity of angular velocity are monitored to determine the period of stable attitude change. Then, inertial navigation algorithms such as SINS (Straight-through inertial navigation) or PDR (pedestrian inertial navigation) are used to estimate the user's velocity and displacement trajectory, calculate the motion trajectory after the initial position offset, and use the time window sliding algorithm to find the end position coordinates of the trajectory segment and timestamp it. After the acquired end segment data is solved by inertial navigation, the three-dimensional coordinates of the current terminal relative to the preset entry point can be obtained, for example (X=+6.4m, Y=−3.2m, Z=−2.8m). This coordinate is used as the end positioning result of the initial solution of inertial positioning.

[0082] S23: Perform signal filtering and noise suppression operations on the wireless signal data generated by the wireless communication equipment deployed in the underground parking lot, extract the effective received signal strength index and base station identifier, and calculate the initial solution of wireless signal positioning based on the signal strength inversion algorithm.

[0083] Specifically, in the process of performing signal filtering and noise suppression operations on the wireless signal data generated by the wireless communication equipment deployed in the underground parking lot, the collected RSSI signal sequence is first subjected to moving average and outlier removal operations to eliminate abrupt data jumps caused by reflection, multipath, or transient interference. Then, the mean RSSI intensity, signal source ID (such as Wi-Fi MAC address or Bluetooth Beacon UUID), and its reception timestamp are extracted from the effective signal for each time period. A signal strength-spatial correspondence matrix is ​​constructed, and the distance information of the current receiver relative to the base station is calculated based on the pre-constructed signal inversion model. In the case of simultaneous reception of signals from multiple base stations, geometric inversion or KNN matching is used to generate spatial coordinate estimates to obtain the initial solution of wireless signal positioning. The solution is then timestamped. For example, in the parking area on the north side of B3 level, the user terminal receives signal strengths of three Beacon nodes of -68dBm, -72dBm, and -64dBm, respectively. The model inversion results show that the current user location is estimated as X=21.6, Y=8.4, Z=-6.0, with a confidence level of 83%.

[0084] In one embodiment, such as Figure 4 As shown, in step S30, the quality index evaluation is performed on the initial solution of the multi-source localization, which specifically includes:

[0085] S31: Perform image recognition quality assessment on the initial solution of image localization. The image recognition quality assessment includes image sharpness assessment, target recognition confidence calculation and target bounding box stability analysis to obtain the image recognition quality assessment result.

[0086] Specifically, the image recognition quality assessment first uses the Laplacian operator to calculate the edge sharpness of video frame images, obtaining a sharpness score to evaluate whether the image has effective resolution. Based on this, YOLOv5 is used to perform target detection and output the target recognition confidence score. Frames with a confidence score lower than a preset threshold (e.g., 0.5) are considered unreliable. At the same time, Kalman filtering is combined to perform cross-frame target tracking to extract the spatial stability features of the target bounding box. The stability fluctuation is evaluated by analyzing the rate of change of the coordinates of the bounding box center point in consecutive frames. If the bounding box shows significant drift or jump within a certain period, the localization result of that segment of the image is judged to be unstable. If any of the image quality score, recognition confidence score, or boundary stability index fails to meet the set threshold, the image localization result is marked as a low-quality result and is excluded from the subsequent fusion process. For example, when the camera is continuously detecting the area at the exit of a ramp, if the vehicle edges in the image are blurred and the bounding box jitters severely, it is judged as a low-quality image frame and is not included in the subsequent fusion process.

[0087] S32: Perform an inertial solution quality assessment on the initial inertial positioning solution. The inertial solution quality assessment includes the measurement of acceleration change amplitude, angular velocity curve continuity and cumulative drift error, and obtain the inertial solution quality assessment result.

[0088] Specifically, the inertial calculation quality assessment analyzes the user terminal's acceleration and angular velocity data using a sliding window approach. It calculates the acceleration variation amplitude and the continuity of the angular velocity curve using standard deviation and rate of change functions. Simultaneously, ZUPT is used to assist in detecting stationary segments. By comparing the cumulative displacement and acceleration integral results of non-stationary segments with the stationary determination results, the cumulative drift error is estimated. If the angular velocity curve within a certain time window exhibits frequent inflection points or abrupt changes, and the drift error exceeds a specific distance threshold (e.g., 0.8m), then that segment of inertial data is considered to have a high risk of deviation, and its confidence level will be lowered or it will be removed. For example, if a continuous movement trajectory still appears in a stationary vehicle through inertial calculation, it indicates that the cumulative drift error is too large, and its weight should be reduced or the trajectory segment removed. This assessment process effectively eliminates misleading inputs caused by sensor errors or environmental interference, improving the overall reliability of the fusion process.

[0089] S33: Perform a signal quality assessment on the initial solution of wireless signal positioning. The signal quality assessment includes the mean amplitude of the received signal strength RSSI, the hopping rate, and the number of signal sources, to obtain the signal quality assessment result.

[0090] Specifically, the wireless signal quality assessment obtains the mean amplitude of RSSI (Received Signal Strength Index) as a basic stability index by performing FFT filtering and moving average processing on RSSI data. Then, the amplitude and frequency of RSSI changes over continuous time are calculated using differential statistics to obtain the hopping rate index. Based on this, the number of identifiable signal sources, such as Wi-Fi AP or Bluetooth Beacon, is statistically analyzed at the current positioning time. If the number of signal sources is less than three or the hopping rate is greater than a set threshold (e.g., 30%), the signal data is marked as a low-confidence sample. The entire signal assessment process adopts a time window analysis strategy combined with an RSSI confidence interval distribution model to make the assessment process real-time and fault-tolerant, thereby improving the data adaptability of the positioning fusion module. For example, in blind areas with concrete walls on both sides underground, there are usually few signal sources and large RSSI fluctuations. Such data will be marked as low-quality data to avoid introducing fusion errors.

[0091] In one embodiment, such as Figure 5 As shown, in step S30, a dynamic weighted fusion operation is performed based on the quality evaluation results to obtain a three-dimensional positioning result representing the user's real-time location, specifically including:

[0092] S34: Construct a location source confidence index based on the image recognition quality assessment results, inertial calculation quality assessment results, and signal quality assessment results, and use the normalized location source confidence index as a weighting coefficient.

[0093] Specifically, key indicator parameters from the image recognition quality assessment results, inertial calculation quality assessment results, and signal quality assessment results are read and mapped to confidence values ​​between 0 and 1. The target recognition confidence and bounding box stability from the image recognition quality assessment results serve as components of the image source confidence. The acceleration stability and angular velocity continuity indicators from the inertial calculation quality assessment results are mapped to the inertial source confidence. The mean RSSI and sag rate from the signal quality assessment results are used to determine the wireless source confidence. After constructing the confidence indicators for the three sources, normalization is performed on each indicator using a maximum-minimum normalization strategy to ensure a uniform numerical distribution range for the confidence indicators of different sources. This is used for subsequent weighting factor calculation in fusion. A confidence threshold can be set to eliminate low-quality source inputs; for example, when the image source confidence is below 0.4, its corresponding weighting coefficient is set to zero. This achieves a dynamic filtering and fusion weight adjustment mechanism for data quality, ensuring that the fused output positioning results are more stable and reliable.

[0094] S35: Input the initial solution results of image localization, inertial localization, and wireless signal localization into the pre-trained multi-source fusion algorithm model, and perform weighted calculation based on the weighting coefficients to obtain the three-dimensional localization result.

[0095] Specifically, the initial solutions for image localization, inertial localization, and wireless signal localization are processed uniformly under the same time stamp. First, sample groups with the same or closest timestamps are extracted from the three types of localization data. Time synchronization alignment is then performed on each sample group to ensure temporal consistency during fusion. Subsequently, the three types of initial solutions and their corresponding weighting coefficients are input into a pre-trained multi-source fusion algorithm model. This multi-source fusion algorithm model is constructed based on the weighted center estimation method. At each time step, the fused coordinates are calculated using confidence weighting, i.e., the final three-dimensional coordinate result is output by multiplying the weights by the corresponding source coordinates and then summing them. This process is then applied to the three-dimensional coordinate vectors in the image localization results and the endpoints of the inertial localization trajectory. The coordinates obtained from the inversion of the wireless signal are vector-weighted and superimposed, and multi-dimensional feature fusion is completed by combining the preset model parameters. The output is a fused three-dimensional coordinate result to represent the user's real-time spatial position. For example, at a specific time t, the initial solution for image positioning is (x1, y1, z1), the initial solution for inertial positioning is (x2, y2, z2), and the initial solution for wireless signal positioning is (x3, y3, z3). The corresponding confidence weighting coefficients are α1, α2, and α3, respectively. Then the formula for calculating the fused coordinates is X=α1x1+α2x2+α3x3, Y=α1y1+α2y2+α3y3, Z=α1z1+α2z2+α3z3. Finally, the fused three-dimensional coordinates are used as the positioning result at the current time for subsequent navigation path matching and update processes.

[0096] In one embodiment, such as Figure 6 As shown, in step S40, based on the 3D positioning results and the user's target location, path planning is performed using the 3D navigation topology map to obtain the target navigation path, specifically including:

[0097] S41: Determine the navigation starting node based on the 3D positioning results, and determine the navigation target node based on the user's target location. Both the starting node and the target node correspond to spatial nodes in the 3D navigation topology map.

[0098] Specifically, spatial matching is performed based on the 3D coordinates contained in the 3D positioning results and the spatial node information in the navigation topology map. Nearest neighbor search is used to determine the node closest to the current positioning point as the navigation starting node. Then, the target area is parsed out through the target location text, QR code, or point selection operation input by the user terminal and projected into 3D space. The topology map node that best matches the target location is matched as the navigation target node through spatial reverse lookup or semantic space mapping. For example, when the user is positioned in the northwest corner of B2 floor near the elevator area, with positioning coordinates (23.5, 47.8, -4.2), the node numbered N132 is matched as the starting node. The target location is "B1 floor exit direction". Then, through natural language parsing and spatial semantic matching, the node numbered N05 closest to the exit is selected as the target node. The starting node and target node are assigned to the start and end parameters of the path search engine for subsequent path calculation operations.

[0099] S42: Perform a path search operation in the 3D navigation topology map to obtain the path search results. The path search operation evaluates the path between nodes based on a preset cost function, which includes path length, height difference, passage restrictions, and structural adaptability factor.

[0100] Specifically, in the constructed 3D navigation topology map, a path search operation is performed to obtain a sequence of passable paths from the starting node to the target node. The path search process uses the A* algorithm, and the cost of each connecting edge is used as the basis for path selection. The cost function comprehensively considers the spatial distance between nodes as the path length factor, the vertical height difference between high and low floors as the vertical movement cost, and the obstacles, restricted time, or closed passage markers in the path passage records as passage restriction factors. The passage cost coefficient of the edge is dynamically adjusted in combination with the structural adaptability requirements such as whether the slope of the area traversed by the path exceeds the equipment driving limit and whether the space width supports vehicle passage. Thus, under the premise of ensuring passage feasibility, smooth, low-cost, and structurally reasonable path segments are prioritized as path search results. During the path search process, to improve the compatibility between the navigation path and the actual structure of the underground parking lot, a "structural adaptability factor" is designed to quantitatively evaluate the passage path. This factor is used to measure the structural coupling degree between the candidate path and the building components, specifically including the matching degree between the path and the ramp inclination angle, the matching degree between the path and the passage width, the consistency between the path and the entrance / exit direction, and the structural continuity between path nodes. Among them, the matching degree of the ramp inclination angle can be calculated by normalizing the absolute value of the difference between the path segment inclination angle and the ramp structural design value; the matching degree of the passage width can be constructed based on the ratio of the path segment's lateral width to the vehicle width to build an adaptation score; the consistency of direction can be calculated using the cosine value of the angle between the path node's direction vector and the structural guidance vector; and the structural continuity can be determined by binary judgment based on whether the path segment has structural breaks or crosses non-passage areas. Finally, the various indicators are weighted and integrated to form a structural adaptability score, which is used as a quantitative indicator in the path cost function to participate in the path optimization process, thereby achieving full adaptation to the building's physical environment and ensuring the rationality of navigation.

[0101] S43: Generate a node sequence based on the path search results as the spatial structure sequence of the navigation path, and perform curvature smoothing and spatial coherence correction on the node sequence to form the target navigation path. The target navigation path includes continuous three-dimensional spatial segment information and labels the spatial coordinates, height attributes and travel direction information of the path nodes.

[0102] Specifically, based on the path segment results obtained from the path search operation, an ordered node sequence is generated according to the connection order of the path nodes. This node sequence is used as the spatial skeleton structure of the navigation path. Further curvature smoothing processing is performed to correct the problem of excessively large polygonal angles caused by the discrete distribution of nodes. Spline curve fitting methods can be used to make the path have more natural curve continuity in space. At the same time, spatial coherence correction operations are performed on possible height jumps or abrupt changes in direction between nodes to ensure that there are no abrupt changes in the connection of the path between consecutive floors, slopes or elevators. Finally, a complete target navigation path is constructed, which includes a series of continuous three-dimensional spatial segments. Each segment is accompanied by its start and end coordinates, height attributes and recommended travel direction information for vehicles or people in three-dimensional space to support subsequent navigation trajectory drawing and real-time update operations.

[0103] In one embodiment, such as Figure 7 As shown, this multi-source fusion-based underground parking lot navigation method also includes:

[0104] S50: During the execution of the navigation route, continuously collect real-time traffic status information of the target navigation route. The traffic status information includes the degree of vehicle congestion, traffic blockage status and user movement behavior characteristics of the corresponding road segments in the route. Input the traffic status data into the preset route traffic scoring model and output the scoring result representing the traffic weight of each route segment.

[0105] Specifically, by continuously monitoring the vehicle traffic status information of the user's road segment during the navigation path execution, the system sequentially acquires traffic density from video images, obstruction signs or area closure prompts detected by cameras, and speed change patterns reflected in inertial measurement data. This allows the system to determine whether the traffic status of each segment of the target navigation path is abnormal or congested. For each segment, the system extracts indicators such as average vehicle speed, percentage of stationary time, probability of traffic obstruction, and user dwell characteristics to construct a traffic status vector. This traffic status vector is then input into a preset path traffic scoring model to perform a comprehensive evaluation calculation, outputting a traffic weight score reflecting the current path's smoothness. If a user exhibits behavioral characteristics such as continuous stationary movement, short-distance back-and-forth movement, or prolonged failure to enter the next node in a certain road segment, this can be combined with high-slope or intersection areas in the path attributes to further confirm the traffic obstruction situation in that segment. This allows the scoring model to be adaptable to different time periods and different types of paths. The pre-defined route access scoring model is obtained through offline construction and training. The training process first collects sample data from multiple typical road sections in the underground parking lot at different times and under different traffic conditions. This includes multi-source feature vectors such as traffic density images extracted from camera videos, user movement patterns reflected in inertial measurement data, wireless signal strength stability, and route access history. These feature vectors are standardized and used as model inputs. The route access levels generated by manual annotation or historical experience data are combined as supervision labels to construct a route access scoring sample set. A lightweight regression model, such as an XGBoost regressor or an MLP neural network, is used as the training framework. The model is trained and optimized by minimizing the mean square error between the predicted score and the actual access level. The model output is a access score value in the range of 0 to 1, where a value closer to 1 indicates smooth traffic and a value closer to 0 indicates congestion or blockage. Before deployment, the trained model needs to undergo robustness evaluation and parameter adjustment through cross-validation and actual route test data to improve its generalization ability and real-time scoring accuracy in complex environments.

[0106] S60: When the score result is detected to be lower than the preset score threshold, a path search operation is performed based on the three-dimensional topology map, and the updated navigation path is sent to the navigation terminal as an alternative path to replace the content displayed on the current path.

[0107] Specifically, in the route traffic status monitoring and reconstruction mechanism, the scoring threshold is set based on modeling the historical distribution of traffic scores for navigation route segments and the expected tolerance of navigation users. During the construction phase, the scoring model is trained using traffic status data collected in real-world scenarios. It combines features such as different congestion levels, obstacle frequency, user dwell behavior, and route breakpoint risk to generate a statistical matching relationship between traffic scores and route segments. Through weighted clustering and empirical threshold setting methods, traffic scores are divided into multiple level intervals. The scoring threshold is the dividing point between "lower acceptability" and "inaccessible," and its number... The value can be dynamically adjusted based on the target application scenario. For example, in high-risk warning states such as fires and police emergencies, the threshold can be raised to trigger reconstruction in advance. In normal navigation states, it can be kept near the back 1 / 4 quantile of the empirical distribution to balance stability and sensitivity. When the score of any path during navigation is lower than the preset score threshold, a path search operation is automatically triggered to find alternative paths in the 3D navigation topology map. The new optimal path is dynamically generated by combining the current user positioning result and the target location. The alternative path will update the content displayed on the navigation terminal interface to improve traffic efficiency and enhance the ability to respond to emergencies.

[0108] S70: In the navigation terminal, the three-dimensional spatial positioning results and the updated navigation path are bound to the three-dimensional building information model, and the rendering process of the three-dimensional navigation interface is executed. The rendering process includes three-dimensional path trajectory drawing and spatial orientation auxiliary display.

[0109] Specifically, by synchronously mapping the current 3D spatial positioning result and the updated navigation path to the corresponding position coordinates in the original 3D building information model, and triggering rendering logic in the navigation terminal to load the new path display content, the rendering process includes drawing continuous line segments for each 3D path, marking the passage status with color or thickness, and marking the direction information with arrows. At the same time, the current floor number and the spatial relationship between the user's current position and the target position are provided at the edge of the interface. If there are cross-floor connecting segments in the current path, the upward or downward direction can be indicated by animation effects or gradient prompts, thereby realizing complete path guidance and improved spatial orientation perception from the user's perspective.

[0110] Furthermore, the integrated design of visualization capabilities based on the 3D building information model of the underground parking lot and video AI algorithms has expanded to include an AI-BIM fusion-based intelligent alarm function. This function accesses video streams from surveillance cameras inside the underground parking lot and uses an image intelligent analysis model deployed on edge devices or servers to identify specific abnormal events in the video in real time. These events include, but are not limited to, high-risk scenarios such as people falling, smoke spreading, open flames, or vehicles driving in the wrong direction. Once an alarm condition is triggered, the system can locate the location of the abnormal event and simultaneously retrieve facility information associated with that location from the BIM database. This includes the spatial location and status parameters of surrounding fire hydrants, smoke exhaust fans, emergency evacuation indicator lights, and other components. The system also automatically generates an event response decision support information package, which includes the event type, location, a list of related facilities, and their current activation status. This information package is highlighted in the 3D visualization interface on the management side and simultaneously triggers multiple channels of alarms and evacuation guidance mechanisms, such as broadcast systems and mobile app push notifications. This improves response efficiency and the level of intelligence in handling complex emergency scenarios.

[0111] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0112] In one embodiment, a multi-source fusion-based underground parking lot navigation system is provided, which corresponds one-to-one with the multi-source fusion-based underground parking lot navigation method described in the above embodiments. For example... Figure 8 As shown, this multi-source fusion-based underground parking lot navigation system includes modules A, B, C, and D. Detailed descriptions of each functional module are as follows:

[0113] The 3D modeling module is used to acquire the 3D building information model of the underground parking lot and construct a 3D navigation topology map based on the 3D building information model. The 3D navigation topology map includes the 3D coordinate information of spatial nodes and the accessibility attributes of connecting paths.

[0114] The multi-source data acquisition and preprocessing module is used to acquire multi-source data, perform preprocessing operations on the multi-source data, and obtain the initial solution of multi-source positioning. The multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot.

[0115] The quality assessment and fusion module is used to perform quality index assessment on the initial solution of multi-source positioning, obtain the multi-source positioning quality evaluation result, and perform dynamic weighted fusion operation based on the quality evaluation result to obtain the three-dimensional positioning result used to represent the user's real-time location.

[0116] The path planning module is used to plan the target navigation path based on the 3D positioning results and the user's target location, combined with the 3D navigation topology map.

[0117] Optional, the 3D modeling module includes:

[0118] The structural component extraction submodule is used to extract structural component information of the underground parking lot from the 3D building information model. The structural component information includes floor structure, ramp location, entrance and exit space, elevator shaft and passage outline data.

[0119] The spatial node modeling submodule is used to partition and model the building space based on structural component information, discretizing the passage area in the underground parking lot into multiple spatial nodes. Each spatial node contains three-dimensional coordinates, the floor number it is located on, and corresponding structural attribute information.

[0120] The path connection establishment submodule is used to establish path connection relationships between identified spatial nodes. The path connection relationship represents the topological connectivity of the travel path, and the corresponding height difference, travel direction, slope information and structural travel restrictions are marked on each path connection.

[0121] Optional, the multi-source data acquisition and preprocessing module includes:

[0122] The image localization preprocessing submodule is used to perform image enhancement and target detection operations on video images captured by cameras inside the underground parking lot, identify user or vehicle targets and obtain their two-dimensional image coordinates in the image plane, and calculate the initial image localization solution based on the two-dimensional image coordinates and camera spatial calibration information.

[0123] The inertial positioning preprocessing submodule is used to perform resampling and interference filtering operations on the inertial measurement information of the user terminal, extract the acceleration and angular velocity data of the stable segment, and estimate the user's displacement trajectory based on the inertial navigation algorithm to obtain the initial solution of inertial positioning.

[0124] The wireless signal positioning preprocessing submodule is used to perform signal filtering and noise suppression operations on the wireless signal data generated by the wireless communication equipment deployed in the underground parking lot, extract the effective received signal strength index and base station identifier, and calculate the initial solution of wireless signal positioning based on the signal strength inversion algorithm.

[0125] Optional, the quality assessment and integration module includes:

[0126] The image quality assessment submodule is used to perform image recognition quality assessment on the initial image localization solution results. The image recognition quality assessment includes image sharpness assessment, target recognition confidence calculation and target bounding box stability analysis to obtain the image recognition quality assessment results.

[0127] The inertial mass assessment submodule is used to perform inertial solution quality assessment on the initial inertial positioning solution results. The inertial solution quality assessment includes the acceleration change amplitude, the continuity of the angular velocity curve and the cumulative drift error measurement, and obtains the inertial solution quality assessment results.

[0128] The signal quality assessment submodule is used to perform signal quality assessment on the initial solution of wireless signal positioning. The signal quality assessment includes the average amplitude of the received signal strength RSSI, the hopping rate, and the number of signal sources, to obtain the signal quality assessment result.

[0129] The confidence construction submodule is used to construct the location source confidence index based on the image recognition quality assessment result, the inertial calculation quality assessment result, and the signal quality assessment result, and then normalize the location source confidence index and use it as a weighting coefficient.

[0130] The fusion calculation submodule is used to input the initial solution results of image localization, inertial localization, and wireless signal localization into a pre-trained multi-source fusion algorithm model, and perform weighted calculation operations based on weighting coefficients to obtain the three-dimensional localization result.

[0131] Optionally, the route planning module includes:

[0132] The navigation node determination submodule is used to determine the navigation start node based on the 3D positioning results and the navigation target node based on the user's target location. Both the start node and the target node correspond to spatial nodes in the 3D navigation topology map.

[0133] The path search submodule is used to perform path search operations in the 3D navigation topology map and obtain path search results. The path search operation evaluates the path between nodes based on a preset cost function, which includes path length, height difference, passage restrictions and structural adaptability factors.

[0134] The path optimization generation submodule is used to generate a node sequence based on the path search results, which serves as the spatial structure sequence of the navigation path. The node sequence is then subjected to curvature smoothing and spatial coherence correction to form the target navigation path. The target navigation path includes continuous three-dimensional spatial segment information and labels the spatial coordinates, height attributes, and travel direction information of the path nodes.

[0135] Optionally, this multi-source fusion-based underground parking lot navigation system also includes:

[0136] The traffic status collection and scoring module is used to continuously collect real-time traffic status information of the target navigation path during the execution of the navigation path. The traffic status information includes the degree of vehicle congestion, traffic blockage status and user movement behavior characteristics of the corresponding road segments in the path. The traffic status data is input into the preset path traffic scoring model and the scoring result representing the traffic weight of each segment of the path is output.

[0137] The path reconstruction trigger module is used to perform a path search operation based on the 3D topology map when the detected score result is lower than the preset score threshold, and to send the updated navigation path as an alternative path to the navigation terminal to replace the current path display content.

[0138] The navigation interface rendering module is used to bind the three-dimensional spatial positioning results and the updated navigation path to the three-dimensional building information model in the navigation terminal, and to execute the rendering process of the three-dimensional navigation interface. The rendering process includes three-dimensional path trajectory drawing and spatial orientation auxiliary display.

[0139] For specific limitations regarding the multi-source fusion-based underground parking lot navigation system, please refer to the limitations of the multi-source fusion-based underground parking lot navigation method described above, which will not be repeated here. Each module in the aforementioned multi-source fusion-based underground parking lot navigation system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0140] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0141] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A navigation method for underground parking lots based on multi-source fusion, characterized in that, The aforementioned underground parking lot navigation method based on multi-source fusion includes: A three-dimensional building information model of an underground parking lot is obtained, and a three-dimensional navigation topology map is constructed based on the three-dimensional building information model. The three-dimensional navigation topology map includes three-dimensional coordinate information representing spatial nodes and accessibility attributes of connecting paths. Collect multi-source data, perform preprocessing operations on the multi-source data to obtain the initial solution of multi-source positioning, wherein the multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot; The initial solution of the multi-source positioning is evaluated by quality indicators to obtain the multi-source positioning quality evaluation result. Based on the quality evaluation result, a dynamic weighted fusion operation is performed to obtain the three-dimensional positioning result used to represent the user's real-time location. Based on the three-dimensional positioning results and the user's target location, a path is planned using the three-dimensional navigation topology map to obtain the target navigation path.

2. The underground parking lot navigation method based on multi-source fusion according to claim 1, characterized in that, The construction of a 3D navigation topology map based on the 3D building information model specifically includes: Structural component information of the underground parking lot is extracted from the three-dimensional building information model. The structural component information includes floor structure, ramp location, entrance and exit space, elevator shaft and passage outline data. Based on the structural component information, the building space is partitioned and modeled, and the passage area in the underground parking lot is discretized into multiple spatial nodes. Each spatial node contains three-dimensional coordinates, the floor number it is located on, and the corresponding structural attribute information. Establish path connections between identified spatial nodes. These path connections represent the topological connectivity of the travel paths. Mark the corresponding height difference, travel direction, slope information, and structural travel restrictions on each path connection.

3. The underground parking lot navigation method based on multi-source fusion according to claim 1, characterized in that, The preprocessing operation performed on the multi-source data to obtain the preliminary solution for multi-source localization specifically includes: Image enhancement and target detection operations are performed on video images captured by cameras inside the underground parking lot to identify user or vehicle targets and obtain their two-dimensional image coordinates in the image plane. Based on the two-dimensional image coordinates and camera spatial calibration information, the initial solution for image localization is calculated. Resampling and interference filtering operations are performed on the inertial measurement information of the user terminal to extract the acceleration and angular velocity data of the stable segment, and the user displacement trajectory is estimated based on the inertial navigation algorithm to obtain the initial solution of inertial positioning. Signal filtering and noise suppression operations are performed on the wireless signal data generated by the wireless communication equipment deployed in the underground parking lot to extract the effective received signal strength index and base station identifier, and the initial solution of wireless signal positioning is calculated based on the signal strength inversion algorithm.

4. The underground parking lot navigation method based on multi-source fusion according to claim 3, characterized in that, The quality index evaluation of the preliminary multi-source localization solution results specifically includes: An image recognition quality assessment is performed on the initial image localization solution. The image recognition quality assessment includes image sharpness assessment, target recognition confidence calculation and target bounding box stability analysis to obtain the image recognition quality assessment result. An inertial solution quality assessment is performed on the initial inertial positioning solution. The inertial solution quality assessment includes measurements of acceleration variation amplitude, angular velocity curve continuity, and cumulative drift error, to obtain the inertial solution quality assessment result. A signal quality assessment is performed on the initial solution of wireless signal positioning. The signal quality assessment includes the mean amplitude of the received signal strength RSSI, the hopping rate, and the number of signal sources, to obtain the signal quality assessment result.

5. The underground parking lot navigation method based on multi-source fusion according to claim 4, characterized in that, The step of performing a dynamic weighted fusion operation based on the quality evaluation results to obtain a three-dimensional positioning result representing the user's real-time location specifically includes: Based on the image recognition quality assessment results, the inertial calculation quality assessment results, and the signal quality assessment results, a location source confidence index is constructed, and the location source confidence index is normalized and used as a weighting coefficient. The initial solution results of image localization, inertial localization, and wireless signal localization are input into a pre-trained multi-source fusion algorithm model, and a weighted calculation operation is performed based on the weighting coefficients to obtain the three-dimensional localization result.

6. The underground parking lot navigation method based on multi-source fusion according to claim 1, characterized in that, The step of calculating the target navigation path based on the 3D positioning results and the user's target location, combined with the 3D navigation topology map, specifically includes: The navigation starting node is determined based on the three-dimensional positioning results, and the navigation target node is determined based on the user's target location. Both the starting node and the target node correspond to spatial nodes in the three-dimensional navigation topology map. A path search operation is performed in the three-dimensional navigation topology map to obtain path search results. The path search operation evaluates the path between nodes based on a preset cost function, which includes path length, height difference, passage restrictions and structural adaptability factors. Based on the path search results, a node sequence is generated as the spatial structure sequence of the navigation path. The node sequence is then subjected to curvature smoothing and spatial coherence correction to form the target navigation path. The target navigation path includes continuous three-dimensional spatial segment information and labels the spatial coordinates, height attributes, and travel direction information of the path nodes.

7. The underground parking lot navigation method based on multi-source fusion according to claim 1, characterized in that, The underground parking lot navigation method based on multi-source fusion also includes: During the execution of the navigation route, real-time traffic status information of the target navigation route is continuously collected. The traffic status information includes the degree of vehicle congestion, traffic blockage status and user movement behavior characteristics of the corresponding road segments in the route. The traffic status data is input into a preset route traffic scoring model and the scoring result representing the traffic weight of each route segment is output. When the scoring result is detected to be lower than the preset scoring threshold, a path search operation is performed based on the three-dimensional topology map, and the updated navigation path is sent to the navigation terminal as an alternative path to replace the content displayed on the current path. In the navigation terminal, the three-dimensional spatial positioning result and the updated navigation path are bound to the three-dimensional building information model, and the rendering process of the three-dimensional navigation interface is executed. The rendering process includes three-dimensional path trajectory drawing and spatial orientation auxiliary display.

8. A navigation system for underground parking lots based on multi-source fusion, characterized in that, The aforementioned underground parking lot navigation system based on multi-source fusion includes: A 3D modeling module is used to acquire a 3D building information model of an underground parking lot and construct a 3D navigation topology map based on the 3D building information model. The 3D navigation topology map includes 3D coordinate information representing spatial nodes and accessibility attributes of connecting paths. The multi-source data acquisition and preprocessing module is used to acquire multi-source data, perform preprocessing operations on the multi-source data, and obtain the initial solution of multi-source positioning. The multi-source data includes camera image information inside the underground parking lot, inertial measurement information of user terminals, and wireless signal data of wireless communication devices in the underground parking lot. The quality assessment and fusion module is used to perform quality index assessment on the initial solution of the multi-source positioning to obtain the multi-source positioning quality evaluation result, and to perform dynamic weighted fusion operation based on the quality evaluation result to obtain the three-dimensional positioning result representing the user's real-time location. The path planning module is used to plan a path based on the three-dimensional positioning results and the user's target location, combined with the three-dimensional navigation topology map, to obtain the target navigation path.

Images