Multi-source remote sensing collaborative identification method and system for road height restriction obstacles

Abstract

This invention provides a multi-source remote sensing collaborative identification method and system for road height restriction obstacles, relating to the field of road height restriction identification technology. It involves locating M height restriction obstacles using pixel-level elevation statistics through a moving sliding window within elevation information blocks, calculating and correcting the clearance height, and projecting M baseline height restriction heights onto the road network. Based on the vehicle type and real-time load information of freight vehicles, negative height restriction compensation is performed to output dynamic passage heights, traversing the road network to locate R feasible routes, which are then displayed on the freight vehicle's onboard display. This solves the technical problems of existing technologies that rely on manual inspection for height restriction obstacle status monitoring, leading to untimely height restriction status updates, and using static height thresholds for vehicle height restriction judgment, affecting the path planning of large vehicles. It achieves the technical effect of dynamically tracking changes in the status of height restriction obstacles and vehicle passage heights, enabling accurate path planning and visual navigation for large vehicles.

Description

Technical Field

[0001] This invention relates to the field of road height restriction recognition technology, and in particular to a multi-source remote sensing collaborative recognition method and system for road height restriction obstacles. Background Technology

[0002] Existing technologies for identifying road height restriction obstacles face three major technical bottlenecks: First, relying on manual inspections or low-frequency laser scanning methods makes it difficult to achieve large-scale dynamic monitoring, resulting in the inability to capture changes in the status of key obstacles such as newly built overpasses and temporary height restriction frames in a timely manner, leading to delayed traffic risk warnings. Second, existing remote sensing identification schemes rely excessively on single DSM elevation data and lack an effective mechanism to distinguish between interfering objects with similar elevation characteristics, such as utility poles and trees, resulting in a high misjudgment rate of height restriction obstacles. Third, the height calculation model uses static thresholds, failing to incorporate changes in vehicle body sinking caused by fluctuations in the load status of large vehicles, and also failing to consider the impact of longitudinal road undulations on the actual traffic space, resulting in a serious disconnect between theoretical height values ​​and actual safe traffic space.

[0003] In summary, existing technologies rely on manual inspections to monitor the status of height-restricted obstacles, resulting in untimely updates to the height restriction status. Furthermore, the use of static height thresholds to determine vehicle height restrictions affects the planning of routes for large vehicles. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a multi-source remote sensing collaborative identification method and system for road height restriction obstacles. This method solves the technical problems in existing technologies, which rely on manual inspections to monitor the status of height restriction obstacles, resulting in untimely updates of height restriction status, and use static height thresholds to determine vehicle height restriction passage, thus affecting the planning of large vehicle passage routes.

[0005] To achieve the above objectives, the present invention provides a multi-source remote sensing collaborative identification method and system for road height restriction obstacles.

[0006] A first aspect of the present invention provides a multi-source remote sensing collaborative identification method for road height restriction obstacles, the method comprising:

[0007] Based on the origin and destination points uploaded by the freight vehicles, feasible route planning is performed to obtain a traffic network; elevation information blocks and optical image blocks within the spatial range of the traffic network are cropped; according to the segment connection direction of the traffic network, pixel-level elevation statistics are performed in the elevation information block's moving sliding window to locate multiple potential height-restricted obstacle candidate points; the multiple potential height-restricted obstacle candidate points are projected onto the optical image block, and multi-temporal obstacle multi-source feature collaborative recognition is performed to filter and locate M height-restricted obstacles; the clearance height of the M height-restricted obstacles is calculated and corrected in the elevation information block, and M baseline height restriction heights are output; negative height restriction compensation is performed based on the vehicle type and real-time load information of the freight vehicles, and dynamic passage height is output; after projecting the M baseline height restriction heights onto the traffic network, the dynamic passage height is used to traverse the traffic network to locate R feasible routes; the R feasible routes are displayed graphically on the freight vehicle's onboard display screen.

[0008] In one implementation, the process of cropping elevation information blocks and optical image blocks within the spatial range of the road network also includes the following steps:

[0009] Retrieve a digital surface model and high-resolution optical satellite imagery covering the area envelope of the road network; retrieve road vector surface data of the road network from the road GIS database; after performing pixel-level spatial alignment of the digital surface model and high-resolution optical satellite imagery, use the road vector surface data as a mask to crop the elevation information block and optical image block from the digital surface model and high-resolution optical satellite imagery respectively.

[0010] In one implementation, the plurality of potential height-restricted obstacle candidate points are projected onto the optical image block, and multi-temporal obstacle multi-source feature collaborative recognition is performed to filter and locate M height-restricted obstacles. The following processing is also performed:

[0011] Multiple potential height-restricted obstacle candidate points are projected onto the optical image block to locate multiple regions of interest; multiple multi-temporal optical images of the multiple regions of interest are retrieved; multi-source feature collaborative recognition of obstacles is performed on the multiple multi-temporal optical images to output multiple obstacle types; with height-restricted obstacle types as constraints, the multiple obstacle types are traversed to filter M height-restricted obstacles, wherein the height-restricted obstacles are marked with geographic coordinates.

[0012] In one implementation, multi-source feature collaborative recognition of obstacles is performed on the plurality of multi-temporal optical images to output multiple obstacle types, and the following processing is also performed:

[0013] The first multi-temporal optical image of the first region of interest is temporally decomposed to obtain multiple single-temporal images; multi-dimensional visual features are extracted from the multiple single-temporal images to obtain multiple multi-source visual features, wherein each multi-source visual feature includes single-temporal spatial morphological features, single-temporal texture spectral features, single-temporal spatial relationship features, and single-temporal shadow features; obstacle collaborative decision-making is performed on the multiple multi-source visual features to output multiple temporal obstacle labels; the multiple temporal obstacle labels are fused according to the temporal reproduction attributes to output a first obstacle type.

[0014] In one implementation, obstacle collaborative decision-making is performed on the plurality of multi-source visual features to output multiple temporal obstacle labels. Prior to this, the following processing is also performed:

[0015] Interactively obtain N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets of N types of height-restricted obstacles; use the N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets as training data to map and construct N sets of height-restricted obstacle recognition models, wherein each set of height-restricted obstacle recognition models includes a spatial morphology recognition model, a texture spectral recognition model, a spatial relationship recognition model, and a shadow feature recognition model; map and isolate the N sets of height-restricted obstacle recognition models in parallel, and add N obstacle consistency verification channels at the parallel output end to complete the localization of the N height-restricted obstacle recognition networks.

[0016] In one implementation, obstacle collaborative decision-making is performed on the plurality of multi-source visual features to output multiple temporal obstacle labels, and the following processing is also performed:

[0017] The first multi-source visual features are input into the first height restriction obstacle recognition network, and multi-threaded obstacle collaborative decision-making is performed through the first set of height restriction obstacle recognition models to output the first set of height restriction obstacle recognition results. The first set of height restriction obstacle recognition results are input into the first obstacle consistency verification channel. The first obstacle consistency verification channel outputs the first candidate recognition result if and only if all the first set of height restriction obstacle recognition results are in a valid state. Similarly, the first multi-source visual features are input into the N height restriction obstacle recognition networks to perform obstacle collaborative decision-making and output N candidate recognition results. The N candidate recognition results are aggregated as the first temporal obstacle label.

[0018] In one implementation, negative height limit compensation is performed based on the vehicle type and real-time load information of the freight vehicle to output a dynamic passage height, and the following processing is also performed:

[0019] The empty reference height is queried from the reference height mapping library according to the vehicle type; after retrieving the load deformation model according to the vehicle type, the real-time load information is loaded into the load deformation model to calculate and output the real-time sinking amount; after subtracting the real-time sinking amount from the empty reference height, the vehicle longitudinal undulation tolerance is introduced to correct the passage height, and the dynamic passage height is obtained.

[0020] In one implementation, the clearance height of the M height-restricted obstacles is calculated and corrected in the elevation information block to output M reference height restriction heights, and the following processing is also performed:

[0021] Extract the i-th top elevation value of the i-th height-restricted obstacle and the i-th road reference elevation value from the elevation information block; calculate the i-th initial elevation difference between the i-th top elevation value and the i-th road reference elevation value; retrieve the i-th multi-temporal optical image from the i-th region of interest, and perform systematic error correction on the i-th initial elevation difference based on the imaging geometry of the i-th multi-temporal optical image, outputting the i-th corrected elevation difference; perform transverse structure height correction on the i-th corrected elevation difference according to the height-restricted obstacle type of the i-th height-restricted obstacle, outputting the i-th reference height restriction height; similarly, perform clearance height calculation and correction on the M height-restricted obstacles in the elevation information block, outputting the M reference height restriction heights.

[0022] In one implementation, the following processing is also performed:

[0023] Based on the vehicle type and real-time load information, energy consumption is predicted for the R feasible routes, and R energy consumption estimates are output. Travel time is predicted based on traffic flow constraints for the R feasible routes, and R travel time estimates are output. Suspension system wear is predicted based on road condition characteristics for the R feasible routes, and R wear levels are output. A multi-objective comprehensive evaluation is performed based on the R energy consumption estimates, R travel time estimates, and R wear levels, and R comprehensive consumption scores are output. The R feasible routes are prioritized based on the R comprehensive consumption scores, and a feasible route priority sequence is output. The feasible route priority sequence is displayed on the in-vehicle display screen for the driver to select via navigation touch control.

[0024] A second aspect of the present invention provides a multi-source remote sensing collaborative identification system for road height restriction obstacles, the system comprising:

[0025] The route planning unit is used to plan feasible routes based on the origin and destination points of freight vehicles uploaded by the freight vehicles, thereby obtaining a road network. The information acquisition and cropping unit is used to crop elevation information blocks and optical image blocks within the spatial range of the road network. The height restriction obstacle preliminary selection unit is used to perform pixel-level elevation statistics in the elevation information block moving sliding window according to the road segment connection direction of the road network, and locate multiple potential height restriction obstacle candidate points. The height restriction obstacle identification unit is used to project the multiple potential height restriction obstacle candidate points onto the optical image block, and perform multi-temporal obstacle multi-source feature collaborative identification to filter and locate M height restriction obstacles. Obstacles; a reference height calculation unit, used to calculate and correct the clearance height of the M height-restricted obstacles in the elevation information block, and output M reference height restriction heights; a passage height compensation update unit, used to perform negative height restriction compensation based on the vehicle type and real-time load information of the freight vehicle, and output dynamic passage height; a feasible route filtering and output unit, used to project the M reference height restriction heights onto the passage road network, and then use the dynamic passage height to traverse the passage road network to locate R feasible routes; a route display and sending unit, used to display the R feasible routes on the vehicle-mounted display screen of the freight vehicle in a graphical manner.

[0026] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0027] The method provided in this embodiment of the invention obtains a road network by planning feasible routes based on the origin and destination points of freight vehicles uploaded by the freight vehicles; it then trims elevation information blocks and optical image blocks within the spatial range of the road network; based on the segment connection direction of the road network, it performs pixel-level elevation statistics in the elevation information block by moving a sliding window to locate multiple potential height-restricted obstacle candidate points; it projects the multiple potential height-restricted obstacle candidate points onto the optical image block and performs multi-temporal obstacle multi-source feature collaborative recognition to filter and locate M height-restricted obstacles; it calculates and corrects the clearance height of the M height-restricted obstacles in the elevation information block to output M reference height restriction heights; it performs negative height restriction compensation based on the vehicle type and real-time load information of the freight vehicles to output dynamic passage heights; after projecting the M reference height restriction heights onto the road network, it uses the dynamic passage heights to traverse the road network to locate R feasible routes; and it displays the R feasible routes on the in-vehicle display screen of the freight vehicle in a graphical manner. It achieves the technical effect of dynamically tracking changes in the status of height-restricted obstacles and the vehicle passage height, enabling precise path planning and visual navigation for large vehicles. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This invention illustrates a flowchart of a multi-source remote sensing collaborative identification method for road height restriction obstacles provided by the present invention.

[0030] Figure 2 A schematic diagram of the multi-source remote sensing collaborative identification system for road height restriction obstacles provided by the present invention is shown.

[0031] Figure labeling: Route planning unit 1, information acquisition and trimming unit 2, height restriction obstacle initial selection unit 3, height restriction obstacle identification unit 4, reference height calculation unit 5, passage height compensation and update unit 6, feasible route screening and output unit 7, route display and transmission unit 8. Detailed Implementation

[0032] This invention provides a multi-source remote sensing collaborative identification method and system for road height restriction obstacles, which addresses the technical problems in the prior art that rely on manual inspection for height restriction obstacle status monitoring, resulting in untimely updates of height restriction status, and the use of static height thresholds for vehicle height restriction judgment, which affects the planning of large vehicle passage routes.

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0034] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0035] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0036] Example 1: A flowchart of the multi-source remote sensing collaborative identification method for road height restriction obstacles provided in this embodiment of the invention. (See attached diagram) Figure 1 The method includes:

[0037] Step P100: Based on the origin and destination points of the freight vehicles uploaded by the freight vehicles, feasible route planning is carried out to obtain the traffic road network.

[0038] Specifically, a global search is conducted within a large area of ​​remote sensing data. Based on the actual start and end locations of the freight mission, all potential routes that can safely pass and deliver the freight are calculated and a road network is constructed.

[0039] The road network precisely limits the subsequent complex remote sensing data processing and analysis tasks to this limited road space that is highly relevant to traffic demand, laying a precise spatial foundation for subsequent collaborative identification.

[0040] Step P200: Crop the elevation information block and optical image block within the spatial range of the road network.

[0041] In one implementation, the elevation information blocks and optical image blocks within the spatial range of the road network are cropped. Step P200 of the method provided by this invention includes:

[0042] Step P210: Retrieve the digital surface model and high-resolution optical satellite imagery covering the envelope of the road network area.

[0043] Step P220: Retrieve the road vector surface data of the traffic network from the road GIS database.

[0044] Step P230: After performing pixel-level spatial alignment of the digital surface model and the high-resolution optical satellite image, using the road vector surface data as a mask, crop the elevation information block and the optical image block from the digital surface model and the high-resolution optical satellite image, respectively.

[0045] This embodiment performs a fine-grained cropping operation on remote sensing data to identify road height restriction obstacles. This operation aims to accurately separate the two core data layers required for subsequent analysis—elevation information and optical imagery—from the original remote sensing dataset covering a wide area, based on the specific road network space range that freight vehicles plan to pass through, thereby realizing the analysis from macro-regional data to micro-road space.

[0046] Specifically, this embodiment obtains a digital surface model dataset and high spatial resolution optical satellite image products from a satellite data center that meets data service standards or an authorized remote sensing cloud service platform. These datasets fully cover the entire outer contour envelope area of ​​the road network determined by the origin and destination of freight transport.

[0047] The digital surface model must be a surface model generated by densely matched point clouds containing information on the vertical structure of the ground surface and the features above it, rather than a traditional digital elevation model. Meanwhile, the optical satellite imagery must have a spatial resolution better than a specified threshold (e.g., within 0.5 meters) to meet the requirements for obstacle detail identification. Furthermore, the two types of data must maintain a unified standard in terms of coordinate system, projection parameters, and data format.

[0048] The road vector surface data is obtained by extracting a set of road boundary vector elements that completely correspond to the current road network from a professional road geographic information database or a standardized data product provided by a commercial map service provider with corresponding surveying and mapping qualifications.

[0049] The road vector surface data not only contains accurate road planar geometry information, but also has complete topological relationship attributes to ensure the connectivity and spatial correctness of the road network, providing an authoritative geometric benchmark for implementing precise spatial mask clipping operations.

[0050] Feature point matching algorithms based on scale-invariant feature transformation or directional fast rotation Brief descriptors enable pixel-level spatial positioning calibration of digital surface models and high-resolution optical satellite imagery. This ensures that the same geographic coordinate point is completely overlapped in grid cells of two heterogeneous data sources, eliminating systematic offsets caused by sensor differences or inconsistent processing procedures.

[0051] Based on this, the preferred scan line filling rasterizes the road vector surface data into a binary mask layer that matches the spatial resolution of the original data. In this layer, the road area is assigned as the effective pixels for target analysis, while the non-road area is marked as background pixels to be excluded.

[0052] Subsequently, this mask is used as a spatial filter to perform pixel-by-pixel logical AND operations on the digital surface model and optical satellite imagery, thereby retaining only the effective information inside the road space and completely eliminating interference signals from the surrounding environment. This ultimately generates elevation information blocks and optical image blocks that strictly match the road network in terms of spatial range and meet the requirements of subsequent complex algorithm processing in terms of data quality.

[0053] This embodiment provides a standardized input dataset that has undergone spatial regularization and quality control for subsequent height restriction obstacle recognition, avoiding the problems of low computational efficiency and impaired recognition accuracy caused by blurred data boundaries or excessive redundant information.

[0054] Step P300: Based on the road segment connection direction of the road network, perform pixel-level elevation statistics in the elevation information block moving sliding window to locate multiple potential height restriction obstacle candidate points.

[0055] Specifically, this embodiment first needs to determine the geometric connection direction and spatial extension trend of each road segment in the traffic network based on road vector data; on this basis, a rectangular sliding window adapted to the road width is used to perform pixel-by-pixel translation and coverage along the road axis; at each window's stopping position, all elevation values ​​within its range are statistically processed, including calculating the local elevation mean, variance, and detecting whether there are elevation abrupt change points exceeding a set threshold. By recording the planar coordinate positions of these elevation anomalies, a candidate point set containing multiple potential height-restricted obstacle candidate points is finally formed, providing target-oriented spatial location guidance for subsequent fine identification and verification based on optical images.

[0056] Step P400: Project the multiple potential height-restricted obstacle candidate points onto the optical image block, and perform multi-temporal obstacle multi-source feature collaborative recognition to screen and locate M height-restricted obstacles.

[0057] In one implementation, the plurality of potential height-restricted obstacle candidate points are projected onto the optical image block, and multi-temporal obstacle multi-source feature collaborative recognition is performed to screen and locate M height-restricted obstacles. The method step P400 provided by the present invention includes:

[0058] Step P410: Locate multiple regions of interest by projecting the multiple potential height-limiting obstacle candidate points onto the optical image block.

[0059] Step P420: Retrieve multiple multi-temporal optical images of the multiple regions of interest.

[0060] Step P430: Perform multi-source feature collaborative recognition of obstacles on the multiple multi-temporal optical images and output multiple obstacle types.

[0061] Step P440: Using the height restriction obstacle type as a constraint, traverse the multiple obstacle types and filter out M height restriction obstacles, wherein the height restriction obstacles are marked with geographical coordinates.

[0062] In one implementation, multi-source feature collaborative recognition of obstacles is performed on the multiple multi-temporal optical images to output multiple obstacle types. The method step P430 provided by this invention includes:

[0063] Step P431: Decompose the first multi-temporal optical image of the first region of interest in time sequence to obtain multiple single-temporal images.

[0064] Step P432: Perform multi-dimensional visual feature extraction on the multiple single-temporal images to obtain multiple multi-source visual features, wherein each multi-source visual feature includes single-temporal spatial morphological features, single-temporal texture spectral features, single-temporal spatial relationship features, and single-temporal shadow features.

[0065] Step P433: Perform obstacle collaborative decision-making on the multiple multi-source visual features and output multiple temporal obstacle labels.

[0066] Step P434: Based on the temporal reproduction attributes, fuse the multiple temporal obstacle tags and output the first obstacle type.

[0067] In one implementation, obstacle collaborative decision-making is performed on the multiple multi-source visual features to output multiple temporal obstacle labels. Previously, the method step P433 provided by this invention included:

[0068] Step P433-10: Interact to obtain N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets for N types of height-limited obstacles.

[0069] Step P433-20: Use the N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets as training data to map and construct N sets of height restriction obstacle recognition models. Each set of height restriction obstacle recognition models includes a spatial morphological recognition model, a texture spectral recognition model, a spatial relationship recognition model, and a shadow feature recognition model.

[0070] Steps P433-30: Map, isolate, and connect the N sets of height restriction obstacle recognition models in parallel, and add N obstacle consistency verification channels at the parallel output end to complete the localization of the N height restriction obstacle recognition networks.

[0071] In one implementation, obstacle collaborative decision-making is performed on the multiple multi-source visual features to output multiple temporal obstacle labels. The method step P433 provided by this invention includes:

[0072] Step P4331: Input the first multi-source visual features into the first height restriction obstacle recognition network, perform multi-threaded obstacle collaborative decision-making through the first set of height restriction obstacle recognition models, and output the first set of height restriction obstacle recognition results.

[0073] Step P4332: Input the first group of height restriction obstacle recognition results into the first obstacle consistency verification channel. If and only if all the first group of height restriction obstacle recognition results are in a valid state, the first obstacle consistency verification channel outputs the first alternative recognition result.

[0074] Step P4333: By analogy, input the first multi-source visual feature into the N height restriction obstacle recognition networks to perform obstacle collaborative decision-making and output N alternative recognition results.

[0075] Step P4334: Aggregate the N candidate identification results as the first temporal obstacle label.

[0076] Specifically, in this embodiment, an affine transformation matrix is ​​established between the digital surface model coordinate system and the optical image block coordinate system. Then, a bilinear interpolation algorithm is applied to resample and calculate the coordinates of all candidate points to eliminate the positioning deviation caused by the difference in data resolution. Finally, based on the preset size parameters, a regular grid covering each projection point is generated to form a set of regions of interest with strictly aligned spatial positions. The preset size parameters are preferably 30×30 pixels.

[0077] Based on this, the multiple potential height-limiting obstacle candidate points are projected onto the optical image block to locate multiple regions of interest. Each region of interest is a square or rectangular area of ​​regular size centered on each candidate point. The boundary of the region must cover the obstacle body and its surrounding related ground features to ensure the completeness of subsequent feature extraction.

[0078] Spatial range query conditions are generated based on the geographic coordinates of each region of interest. The spatiotemporal retrieval interface of the distributed image storage system is called based on the spatial range query conditions to filter the original image products that meet the preset time series requirements and have cloud cover below a specified threshold. Then, radiometric normalization and atmospheric correction preprocessing are performed on the obtained original images to eliminate the influence of illumination differences. Finally, the multi-temporal image data stack corresponding to each region of interest is organized and output according to the time dimension to form a complete time series multi-temporal optical image.

[0079] This embodiment performs multi-source feature collaborative recognition of obstacles using multiple multi-temporal optical images of multiple regions of interest, and outputs multiple obstacle types. The specific technical implementation is as follows:

[0080] The technical implementation of multi-source feature collaborative recognition of obstacles for each multi-temporal optical image is basically the same. Therefore, this embodiment takes the analysis of multi-temporal optical images of any region of interest to identify obstacle types as an example to elaborate on the technical solution in detail.

[0081] The first multi-temporal optical image is essentially a data cube containing multiple temporal observations. In this embodiment, a time dimension deconstruction operation is performed on the dataset of the first multi-temporal optical image corresponding to the first region of interest. That is, according to the order of the timestamps of the image acquisition, the first multi-temporal optical image is decomposed into a sequence of single images with independent temporal phases. This sequence includes multiple single-temporal images that are strictly arranged and discrete along the time axis, providing basic data units for subsequent single-temporal feature extraction.

[0082] Multi-dimensional visual features are extracted from the multiple single-temporal images to obtain multiple multi-source visual features. Each multi-source visual feature includes four types of heterogeneous features: single-temporal spatial morphology features, single-temporal texture spectral features, single-temporal spatial relationship features, and single-temporal shadow features.

[0083] Among them, spatial morphology features obtain the geometric structural parameters of obstacles, such as aspect ratio, area, and perimeter, through edge detection and contour fitting algorithms; texture spectral features capture surface material properties using local binary mode operators and spectral reflectance calculation models; spatial relationship features calculate the relative orientation and distance relationship between obstacles and surrounding fixed features (such as streetlights and traffic signs) based on spatial topology analysis algorithms; and shadow features extract shadow length, direction, and area ratio using illumination direction models and projection geometry principles.

[0084] The system performs collaborative obstacle decision-making on the multiple multi-source visual features and outputs multiple temporal obstacle labels. It relies on N pre-constructed height restriction obstacle recognition networks. These N height restriction obstacle recognition networks can perform collaborative decision-making on the multi-source visual features through a heterogeneous feature model parallel verification mechanism and output height restriction obstacle type identifiers with confidence.

[0085] The method for constructing the N height-restricted obstacle recognition networks is as follows:

[0086] The N types of height-restricted obstacles are N typical height-restricted obstacles, such as bridges, gantry cranes, and tunnel entrances. Through an interactive annotation platform, four-dimensional feature training sample sets of the N types of height-restricted obstacles under different time phases are collected to obtain N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets.

[0087] The first sample multi-temporal spatial morphological feature set, the first sample multi-temporal texture spectral feature set, the first sample multi-temporal spatial relationship feature set, and the first sample multi-temporal shadow feature set of the first type of sample height-limited obstacle are extracted by mapping from the N sample multi-temporal spatial morphological feature set, the N sample multi-temporal texture spectral feature set, the N sample multi-temporal spatial relationship feature set, and the N sample multi-temporal shadow feature set.

[0088] The first sample multi-temporal spatial morphological feature set is input into a support vector machine classifier, and a first spatial morphological recognition model is constructed through radial basis kernel function mapping. The first sample multi-temporal texture spectral feature set is imported into a random forest framework, and the first texture spectral recognition model is trained using the Gini impurity criterion. The first sample multi-temporal spatial relationship feature set is modeled with topological associations of ground features through graph convolutional neural networks to generate a spatial relationship recognition model. The first sample multi-temporal shadow feature set is used to establish a shadow-height mapping relationship through a multilayer perceptron regression network to form a first shadow feature recognition model.

[0089] The first spatial morphology recognition model, the first texture spectrum recognition model, the first spatial relationship recognition model, and the first shadow feature recognition model are trained according to the obstacle type of the first type of height-restricted obstacle. After each model is equipped with the skills to perform feature recognition and confidence labeling of the first type of height-restricted obstacle, the first spatial morphology recognition model, the first texture spectrum recognition model, the first spatial relationship recognition model, and the first shadow feature recognition model are isolated in parallel to ensure that the model inference process does not interfere with each other.

[0090] Then, a dedicated consistency verification logic unit is connected to the output of the four models. This unit has built-in Boolean operation rules. Specifically, it only triggers a valid output when the spatial morphology recognition confidence is >0.9, the texture spectrum classification probability is >0.85, the spatial relationship matching degree meets the standard, and the shadow inversion height error is <0.5 meters. The valid output is the comprehensive judgment result of the height restriction obstacle recognition network on the current sample, which includes obstacle type identification and fusion confidence weight.

[0091] By analogy, N sets of height restriction obstacle recognition models are constructed for the N types of sample height restriction obstacles. Then, the N sets of height restriction obstacle recognition models are mapped, isolated, and connected in parallel. At the parallel output end, N obstacle consistency verification channels are added to complete the localization configuration of the N height restriction obstacle recognition networks.

[0092] Subsequently, the N height-limited obstacle recognition networks can be used to perform collaborative obstacle decision-making on the multiple multi-source visual features, outputting multiple temporal obstacle labels. The specific data processing procedure is as follows:

[0093] The first multi-source visual features, including spatial morphological feature vectors, texture spectral feature matrices, spatial relationship topology maps, and shadow feature parameter sets, extracted from the first single-phase image, are input into the first height-restricted obstacle recognition network. This triggers the spatial morphological recognition model deployed within the first height-restricted obstacle recognition network to perform contour matching degree calculation, the texture spectral recognition model to perform material classification probability prediction, the spatial relationship recognition model to perform topology rule compliance judgment, and the shadow feature recognition model to perform height inversion verification. The four models run in parallel on isolated computing resources with multi-threaded inference tasks, and finally each outputs the first set of height-restricted obstacle recognition results with confidence scores, such as morphological matching degree 0.92, probability of material being metal 0.88, spatial relationship verification passed, and shadow estimated height 5.2 meters.

[0094] The first set of height restriction obstacle recognition results is input into the first obstacle consistency verification channel. The first obstacle consistency verification channel has a built-in set of preset hard judgment rules: the spatial shape matching degree must exceed the threshold ≥0.85, the texture spectral classification probability must be greater than ≥0.8, the spatial relationship verification result must pass, and the shadow inversion height must be within the effective physical range, such as 2-10 meters and the error tolerance is less than ±0.3 meters. The verification channel outputs the first candidate recognition result containing the obstacle type identifier and the comprehensive confidence level only when all four conditions are met. If any condition is not met, a null value is output and the transmission of the set of results is terminated. It should be understood that the obstacle type identifier in the first candidate recognition result is the first obstacle type corresponding to the first height restriction obstacle recognition network.

[0095] The first obstacle consistency verification channel outputs the first alternative identification result only when all the first group of height restriction obstacle identification results are in a valid state.

[0096] Similarly, the first multi-source visual features are input into the N height-restricted obstacle recognition networks to perform obstacle collaborative decision-making, output N candidate recognition results, and aggregate the N candidate recognition results as the first temporal obstacle label.

[0097] Similarly, the multiple multi-source visual features are input into the N height-limited obstacle recognition networks to perform obstacle collaborative decision-making and output the multiple temporal obstacle labels.

[0098] The continuity of the temporal tag sequence containing the multiple temporal obstacle tags is verified. If there are K consecutive temporal phases, preferably K≥3, the output obstacle type identifiers are consistent and the spatial position overlap exceeds a set threshold. If the center point offset is <2 pixels, the obstacle is determined to exist continuously and the type is output as the final result.

[0099] When a type conflict occurs, such as when phase T1 is identified as a bridge and T2 as a gantry, the judgment conclusion of the latest phase T2 is selected according to the principle of prioritizing the nearest time. For time-series interruption scenarios, such as when two consecutive phases are lost after identification, a backtracking mechanism is activated to check the consistency ratio of historical labels. If more than the proportion threshold (e.g., 70%) of the valid phases are of the same type, the result is adopted, and finally the first obstacle type classification conclusion fused with time-series evidence is generated, and the first obstacle type is output.

[0100] By analogy, the multi-source feature collaborative recognition of obstacles is performed on the multiple multi-temporal optical images, and multiple obstacle types are output. Using height-restricted obstacle types as constraints, the multiple obstacle types are traversed, and M height-restricted obstacles are selected, wherein the height-restricted obstacles are marked with geographic coordinates.

[0101] This embodiment achieves the technical effect of significantly reducing the misjudgment rate of height-restricted obstacles and outputting structured recognition results with accurate geographic coordinates, while providing spatial location of obstacles for subsequent safe passage path planning for height-restricted areas.

[0102] Step P500: Calculate and correct the clearance height of the M height-restricted obstacles in the elevation information block, and output the M reference height restrictions.

[0103] In one implementation, the clearance height of the M height-restricted obstacles is calculated and corrected in the elevation information block, and M reference height restriction heights are output. The method step P500 provided by this invention includes:

[0104] Step P510: Extract the top elevation value of the i-th height-restricted obstacle and the reference elevation value of the i-th road from the elevation information block.

[0105] Step P520: Calculate the initial elevation difference between the i-th top elevation value and the i-th road reference elevation value.

[0106] Step P530: Retrieve the i-th multi-temporal optical image in the i-th region of interest, and perform system error correction on the i-th initial elevation difference based on the imaging geometry of the i-th multi-temporal optical image, and output the i-th corrected elevation difference.

[0107] Step P540: Based on the height restriction obstacle type of the i-th height restriction obstacle, correct the cross-structure height of the i-th corrected height difference, and output the i-th reference height restriction height.

[0108] Step P550: By analogy, calculate and correct the clearance height of the M height-restricted obstacles in the elevation information block, and output the M reference height restrictions.

[0109] Specifically, in this embodiment, M obstacle center points are located in the elevation information block based on the M geographical coordinates of the M height-restricted obstacles. Then, using the M obstacle center points as a reference, a Gaussian weighted average algorithm is used to extract the pixel elevation set covering the top area of ​​the obstacles, and the elevation peak value is extracted from the pixel elevation set as the i-th top elevation value.

[0110] Synchronously, at the road reference plane position directly below the top area of ​​the obstacle, the elevation mean filter of the sliding window with the window size adapted to the road width is used to remove the interference of local road surface undulations, and the smoothed i-th road reference elevation value is output to ensure that the reference plane elevation reflects the true height of the vehicle passage plane.

[0111] Perform an arithmetic subtraction operation between the i-th top elevation value and the i-th road reference elevation value to generate the i-th initial elevation difference without error compensation. This value includes projection distortion caused by DSM data acquisition errors, road longitudinal slope inclination, and elevation pollution caused by attachments at the bottom of obstacles, and needs to be systematically corrected in subsequent steps.

[0112] The system calls the multi-temporal optical image stack corresponding to the i-th region of interest, analyzes the rational polynomial coefficients of each image, and inverts the spatial equations of the imaging rays for each temporal phase. Then, it optimizes the three-dimensional spatial intersection accuracy of the target point by fusing the temporal observation data through multi-temporal bundle adjustment. Finally, it calculates the system offset vector between the elevation projection point of the digital surface model and the image point of the optimized optical image, and applies dynamic compensation to the i-th initial elevation difference accordingly, such as parallax compensation of +0.3 meters for the top view angle, and outputs the i-th corrected elevation difference to eliminate sensor geometric distortion.

[0113] Based on the type identifier of the i-th height-restricted obstacle, the pre-set cross-span structure height correction rule library is invoked. The specific rule library content includes: for bridge-type obstacles, a fixed compensation value Δh needs to be added due to the thickness of the bridge body, such as +0.5 meters; for gantry-type obstacles, cosine projection compensation is calculated based on the tilt angle of the support; for tunnel entrances, the reference plane offset needs to be corrected in combination with the slope gradient. Finally, the i-th reference height restriction height that integrates the structural characteristics of the obstacle is output. This value represents the true vertical clearance of the space where vehicles can pass.

[0114] For the remaining M-1 height-restricted obstacles, perform the pipeline operation of steps P510-P540 repeatedly to output M reference height restrictions.

[0115] This implementation corrects the output of M baseline height limits, providing reliable reference data for subsequent feasibility route determination based on vehicle dynamic height limits.

[0116] Step P600: Perform negative height limit compensation based on the vehicle type and real-time load information of the freight vehicle, and output the dynamic passage height.

[0117] In one implementation, negative height limit compensation is performed based on the vehicle type and real-time load information of the freight vehicle, and a dynamic passage height is output. Step P600 of the method provided by this invention includes:

[0118] Step P610: Query the unloaded reference height in the reference height mapping library according to the vehicle type.

[0119] Step P620: After retrieving the load deformation model according to the vehicle type, load the real-time load information into the load deformation model to calculate and output the real-time subsidence.

[0120] Step P630: Subtract the real-time subsidence from the unloaded reference height, and then introduce the vehicle longitudinal undulation tolerance to correct the passage height, thus obtaining the dynamic passage height.

[0121] Specifically, in this embodiment, based on the vehicle type of the freight vehicle mentioned above, a precise query operation is performed in the pre-built vehicle reference height mapping library. This database integrates the chassis specifications of the vehicle manufacturer and the measured height report of the unloaded state issued by the third-party testing agency, and stores the unloaded reference height values ​​of various vehicle types under standard working conditions in a structured manner.

[0122] In practice, a hash indexing mechanism is used to accelerate the retrieval process, outputting a floating-point value for the unloaded reference height that strictly matches the current vehicle type, ensuring the authority and traceability of the reference data. It should be understood that the chassis specifications include suspension type, tire size, and chassis ground clearance.

[0123] Based on the vehicle type identifier, a pre-set load deformation calculation model is invoked. This model is jointly calibrated through finite element simulation and actual vehicle load experiments to establish a load mass-frame deformation mapping function. After real-time collected load information, such as cargo weight of 32 tons and distribution center of gravity coordinates, is input into the model, the load distribution ratio of each bearing is first calculated based on the vehicle wheelbase parameters. Then, combined with the suspension system stiffness coefficient and tire compression characteristics, the overall vehicle body sag and attitude tilt angle caused by non-uniform deformation are calculated. Finally, the real-time sag quantification value with millimeter-level accuracy is output, providing key deformation parameters for dynamic height correction.

[0124] After subtracting the real-time subsidence from the unloaded reference height, the vehicle's longitudinal undulation tolerance is introduced to correct the passage height, and the dynamic passage height, which reflects the vehicle's real-time passage capability, is output. The dynamic passage height is then used as the basis for passage decisions.

[0125] This embodiment combines vehicle type and load to dynamically predict the passage height, achieving the technical effect of providing a reference for subsequent height restriction decisions and improving the real-time performance and accuracy of height restriction decisions.

[0126] Step P700: After projecting the M reference height limits onto the road network, use the dynamic road height to traverse the road network and locate R feasible routes.

[0127] Specifically, firstly, based on the high-precision spatial reference system transformation rules, the geographic coordinates associated with the M reference height limits are mapped to the topological network structure of the road network. Height limit value binding relationships are established at the road arc node positions, forming an enhanced road network data model with spatial constraint attributes. Then, the dynamic passage height parameters generated in real time by the vehicle are loaded, and all possible passage paths of the road network are systematically scanned through a graph traversal algorithm. For each candidate path's continuous arc sequence, the reference height values ​​bound to each height limit obstacle in the path's spatial corridor are checked segment by segment to see if they are continuously greater than the dynamic passage height values. If and only if all height limit obstacles along the entire path meet the height passability conditions, the path is registered to the set of feasible routes, ultimately obtaining the aforementioned R feasible routes.

[0128] Step P800: Display the R feasible routes graphically on the in-vehicle display screen of the freight vehicle.

[0129] In one implementation, the method provided by the present invention further includes:

[0130] Step P810: Based on the vehicle type and real-time load information, predict the energy consumption of the R feasible routes and output R energy consumption estimates.

[0131] Step P820: Based on the traffic flow constraints of the R feasible routes, predict the travel time and output R estimated travel times.

[0132] Step P830: Based on the road condition characteristics of the R feasible routes, predict the suspension system loss and output R loss levels.

[0133] Step P840: Perform a multi-objective comprehensive evaluation based on the R energy consumption estimates, R travel time estimates, and R loss levels, and output R comprehensive consumption scores.

[0134] Step P850: Based on the R comprehensive consumption scores, prioritize the R feasible routes and output the feasible route priority sequence.

[0135] Step P860: Display the priority sequence of feasible routes on the vehicle display screen for the driver to select via navigation touch control.

[0136] Specifically, based on the vehicle type and real-time load information, energy consumption is predicted for the R feasible routes, outputting R energy consumption estimates. Integrating real-time traffic flow data and road network topology features, a spatiotemporal velocity field model is used to predict travel time between nodes, and the forced deceleration periods caused by height restriction obstacles are superimposed, outputting R travel time estimates. Based on suspension system parameters and road condition characteristics, the dynamic load on key components is calculated using a stress accumulation model, and losses are classified into levels I-IV according to material damage thresholds, outputting R loss levels.

[0137] The R energy consumption estimates, R travel time estimates, and R loss levels are normalized according to preset weights, and R comprehensive consumption scores are output.

[0138] Based on the R comprehensive consumption scores, the R feasible routes are prioritized, and a feasible route priority sequence is output. The feasible route priority sequence is then displayed on the vehicle display screen for the driver to select via navigation touch control.

[0139] This embodiment achieves the technical effect of precise path planning and visual navigation for large vehicles by using multi-source remote sensing to collaboratively identify and dynamically eliminate interference and monitor changes in obstacle status, combined with vehicle load compensation to generate a safe passage height.

[0140] Example 2: Based on the same inventive concept as the multi-source remote sensing collaborative identification method for road height restriction obstacles in the foregoing examples, this invention provides a multi-source remote sensing collaborative identification system for road height restriction obstacles. See [link to example]. Figure 2 As shown, the system includes:

[0141] Route planning unit 1 is used to plan feasible routes based on the origin and destination points of freight vehicles uploaded by the freight vehicles, and to obtain the traffic network.

[0142] Information acquisition and cropping unit 2 is used to crop elevation information blocks and optical image blocks within the spatial range of the road network.

[0143] The height restriction obstacle preliminary selection unit 3 is used to perform pixel-level elevation statistics in the elevation information block sliding window according to the road segment connection direction of the road network, and locate multiple potential height restriction obstacle candidate points.

[0144] The height restriction obstacle identification unit 4 is used to project the multiple potential height restriction obstacle candidate points onto the optical image block and perform multi-temporal obstacle multi-source feature collaborative identification to screen and locate M height restriction obstacles.

[0145] The reference height calculation unit 5 is used to calculate and correct the clearance height of the M height-restricted obstacles in the elevation information block, and output the M reference height restrictions.

[0146] The passage height compensation update unit 6 is used to perform negative height limit compensation based on the vehicle type and real-time load information of the freight vehicle, and output dynamic passage height.

[0147] The feasible route filtering output unit 7 is used to project the M reference height limits onto the road network and then use the dynamic road height to traverse the road network to locate R feasible routes.

[0148] The route display and transmission unit 8 is used to display the R feasible routes graphically on the vehicle-mounted display screen of the freight vehicle.

[0149] In one implementation, the information acquisition and clipping unit 2 is further used for:

[0150] Retrieve a digital surface model and high-resolution optical satellite imagery covering the area envelope of the road network; retrieve road vector surface data of the road network from the road GIS database; after performing pixel-level spatial alignment of the digital surface model and high-resolution optical satellite imagery, use the road vector surface data as a mask to crop the elevation information block and optical image block from the digital surface model and high-resolution optical satellite imagery respectively.

[0151] In one implementation, the height-limiting obstacle recognition unit 4 is further used for:

[0152] Multiple potential height-restricted obstacle candidate points are projected onto the optical image block to locate multiple regions of interest; multiple multi-temporal optical images of the multiple regions of interest are retrieved; multi-source feature collaborative recognition of obstacles is performed on the multiple multi-temporal optical images to output multiple obstacle types; with height-restricted obstacle types as constraints, the multiple obstacle types are traversed to filter M height-restricted obstacles, wherein the height-restricted obstacles are marked with geographic coordinates.

[0153] In one implementation, the height-limiting obstacle recognition unit 4 is further used for:

[0154] The first multi-temporal optical image of the first region of interest is temporally decomposed to obtain multiple single-temporal images; multi-dimensional visual features are extracted from the multiple single-temporal images to obtain multiple multi-source visual features, wherein each multi-source visual feature includes single-temporal spatial morphological features, single-temporal texture spectral features, single-temporal spatial relationship features, and single-temporal shadow features; obstacle collaborative decision-making is performed on the multiple multi-source visual features to output multiple temporal obstacle labels; the multiple temporal obstacle labels are fused according to the temporal reproduction attributes to output a first obstacle type.

[0155] In one implementation, the height-limiting obstacle recognition unit 4 is further used for:

[0156] Interactively obtain N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets of N types of height-restricted obstacles; use the N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets as training data to map and construct N sets of height-restricted obstacle recognition models, wherein each set of height-restricted obstacle recognition models includes a spatial morphology recognition model, a texture spectral recognition model, a spatial relationship recognition model, and a shadow feature recognition model; map and isolate the N sets of height-restricted obstacle recognition models in parallel, and add N obstacle consistency verification channels at the parallel output end to complete the localization of the N height-restricted obstacle recognition networks.

[0157] In one implementation, the height-limiting obstacle recognition unit 4 is further used for:

[0158] The first multi-source visual features are input into the first height restriction obstacle recognition network, and multi-threaded obstacle collaborative decision-making is performed through the first set of height restriction obstacle recognition models to output the first set of height restriction obstacle recognition results. The first set of height restriction obstacle recognition results are input into the first obstacle consistency verification channel. The first obstacle consistency verification channel outputs the first candidate recognition result if and only if all the first set of height restriction obstacle recognition results are in a valid state. Similarly, the first multi-source visual features are input into the N height restriction obstacle recognition networks to perform obstacle collaborative decision-making and output N candidate recognition results. The N candidate recognition results are aggregated as the first temporal obstacle label.

[0159] In one implementation, the passage height compensation update unit 6 is further configured to:

[0160] The empty reference height is queried from the reference height mapping library according to the vehicle type; after retrieving the load deformation model according to the vehicle type, the real-time load information is loaded into the load deformation model to calculate and output the real-time sinking amount; after subtracting the real-time sinking amount from the empty reference height, the vehicle longitudinal undulation tolerance is introduced to correct the passage height, and the dynamic passage height is obtained.

[0161] In one implementation, the reference height solving unit 5 is further configured to:

[0162] Extract the i-th top elevation value of the i-th height-restricted obstacle and the i-th road reference elevation value from the elevation information block; calculate the i-th initial elevation difference between the i-th top elevation value and the i-th road reference elevation value; retrieve the i-th multi-temporal optical image from the i-th region of interest, and perform systematic error correction on the i-th initial elevation difference based on the imaging geometry of the i-th multi-temporal optical image, outputting the i-th corrected elevation difference; perform transverse structure height correction on the i-th corrected elevation difference according to the height-restricted obstacle type of the i-th height-restricted obstacle, outputting the i-th reference height restriction height; similarly, perform clearance height calculation and correction on the M height-restricted obstacles in the elevation information block, outputting the M reference height restriction heights.

[0163] In one implementation, the route display sending unit 8 is further configured to:

[0164] Based on the vehicle type and real-time load information, energy consumption is predicted for the R feasible routes, and R energy consumption estimates are output. Travel time is predicted based on traffic flow constraints for the R feasible routes, and R travel time estimates are output. Suspension system wear is predicted based on road condition characteristics for the R feasible routes, and R wear levels are output. A multi-objective comprehensive evaluation is performed based on the R energy consumption estimates, R travel time estimates, and R wear levels, and R comprehensive consumption scores are output. The R feasible routes are prioritized based on the R comprehensive consumption scores, and a feasible route priority sequence is output. The feasible route priority sequence is displayed on the in-vehicle display screen for the driver to select via navigation touch control.

[0165] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-source remote sensing collaborative identification method for road height restriction obstacles, characterized in that, include: Based on the origin and destination points uploaded by freight vehicles, feasible routes are planned to obtain the traffic network; Cut the elevation information blocks and optical image blocks within the spatial range of the road network; Based on the road segment connection direction of the road network, pixel-level elevation statistics are performed in the moving sliding window of the elevation information block to locate multiple potential height restriction obstacle candidate points; The multiple potential height-restricted obstacle candidate points are projected onto the optical image block to perform multi-temporal obstacle multi-source feature collaborative recognition, so as to screen and locate M height-restricted obstacles; In the elevation information block, the clearance height of the M height-restricted obstacles is calculated and corrected, and M reference height restriction heights are output. Based on the vehicle type and real-time load information of the freight vehicle, negative height limit compensation is performed, and dynamic passage height is output. After projecting the M reference height limits onto the road network, the dynamic road height is used to traverse the road network to locate R feasible routes; The R feasible routes are displayed graphically on the in-vehicle display screen of the freight vehicle.

2. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 1, characterized in that, Cropping elevation information blocks and optical image blocks within the spatial range of the road network includes: Retrieve digital surface models and high-resolution optical satellite images covering the envelope of the road network area; Retrieve the road vector surface data of the traffic network from the road GIS database; After performing pixel-level spatial alignment between the digital surface model and the high-resolution optical satellite imagery, the elevation information block and the optical imagery block are cropped from the digital surface model and the high-resolution optical satellite imagery respectively, using the road vector surface data as a mask.

3. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 1, characterized in that, The multiple potential height-restricted obstacle candidate points are projected onto the optical image block, and multi-temporal obstacle multi-source feature collaborative recognition is performed to filter and locate M height-restricted obstacles, including: Multiple regions of interest are located by projecting the multiple potential height-limiting obstacle candidate points onto the optical image block; Retrieve multiple multi-temporal optical images of the multiple regions of interest; Perform multi-source feature collaborative recognition of obstacles on the multiple multi-temporal optical images and output multiple obstacle types; Using the height restriction obstacle type as a constraint, traverse the multiple obstacle types and filter out M height restriction obstacles, wherein the height restriction obstacles are marked with geographical coordinates.

4. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 3, characterized in that, Perform multi-source feature collaborative recognition of obstacles on the multiple multi-temporal optical images, and output multiple obstacle types, including: The first multi-temporal optical image of the first region of interest is temporally decomposed to obtain multiple single-temporal images; Multi-dimensional visual features are extracted from the multiple single-temporal images to obtain multiple multi-source visual features, wherein each multi-source visual feature includes single-temporal spatial morphological features, single-temporal texture spectral features, single-temporal spatial relationship features, and single-temporal shadow features; Perform obstacle collaborative decision-making on the multiple multi-source visual features and output multiple temporal obstacle labels; Based on the temporal reproduction attributes, the multiple temporal obstacle tags are fused together to output the first obstacle type.

5. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 4, characterized in that, Perform obstacle collaborative decision-making on the multiple multi-source visual features and output multiple temporal obstacle labels, including: Interactively obtain N samples of multi-temporal spatial morphological feature sets, N samples of multi-temporal texture spectral feature sets, N samples of multi-temporal spatial relationship feature sets, and N samples of multi-temporal shadow feature sets for N types of height-limited obstacles; The N sample multi-temporal spatial morphological feature sets, N sample multi-temporal texture spectral feature sets, N sample multi-temporal spatial relationship feature sets, and N sample multi-temporal shadow feature sets are used as training data to map and construct N sets of height restriction obstacle recognition models. Each set of height restriction obstacle recognition models includes a spatial morphological recognition model, a texture spectral recognition model, a spatial relationship recognition model, and a shadow feature recognition model. The N sets of height restriction obstacle recognition models are mapped, isolated, and connected in parallel. N obstacle consistency verification channels are added at the parallel output end to complete the localization of the N height restriction obstacle recognition networks.

6. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 5, characterized in that, Perform obstacle collaborative decision-making on the multiple multi-source visual features and output multiple temporal obstacle labels, including: The first multi-source visual features are input into the first height restriction obstacle recognition network, and multi-threaded obstacle collaborative decision-making is performed through the first set of height restriction obstacle recognition models to output the first set of height restriction obstacle recognition results. The first set of height restriction obstacle recognition results are input into the first obstacle consistency verification channel. The first obstacle consistency verification channel outputs the first alternative recognition result if and only if all the first set of height restriction obstacle recognition results are in a valid state. Similarly, the first multi-source visual feature is input into the N height restriction obstacle recognition networks to perform obstacle collaborative decision-making and output N alternative recognition results; The N candidate identification results are aggregated and used as the first temporal obstacle label.

7. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 1, characterized in that, Based on the vehicle type and real-time load information of the freight vehicle, negative height restriction compensation is performed, and dynamic passage height is output, including: Query the unloaded reference height in the reference height mapping library according to the vehicle type; After retrieving the load deformation model according to the vehicle type, the real-time load information is loaded into the load deformation model to calculate and output the real-time subsidence. The dynamic passage height is obtained by subtracting the real-time subsidence from the unloaded reference height and then introducing the vehicle longitudinal undulation tolerance for passage height correction.

8. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 1, characterized in that, The clearance height of the M height-restricted obstacles is calculated and corrected within the elevation information block, outputting M baseline height restrictions, including: Extract the top elevation value of the i-th height-restricted obstacle and the reference elevation value of the i-th road from the elevation information block; Calculate the initial elevation difference between the i-th top elevation value and the i-th road reference elevation value; Retrieve the i-th multi-temporal optical image in the i-th region of interest, and perform systematic error correction on the i-th initial elevation difference based on the imaging geometry of the i-th multi-temporal optical image, and output the i-th corrected elevation difference; Based on the height restriction obstacle type of the i-th height restriction obstacle, the cross-structure height of the i-th corrected height difference is corrected, and the i-th reference height restriction height is output; By analogy, the clearance height of the M height-restricted obstacles is calculated and corrected in the elevation information block, and the M reference height restrictions are output.

9. The multi-source remote sensing collaborative identification method for road height restriction obstacles as described in claim 1, characterized in that, Also includes: Based on the vehicle type and real-time load information, predict the energy consumption of the R feasible routes and output R energy consumption estimates. Based on the traffic flow constraints of the R feasible routes, predict the travel time and output R estimated travel times. Based on the road condition characteristics of the R feasible routes, the suspension system loss is predicted, and R loss levels are output. Based on the R energy consumption estimates, R travel time estimates, and R loss levels, a multi-objective comprehensive evaluation is performed, and R comprehensive consumption scores are output. Based on the R comprehensive consumption scores, prioritize the R feasible routes and output the feasible route priority sequence; The priority sequence of feasible routes is displayed on the vehicle display screen for the driver to select via navigation touch control.

10. A multi-source remote sensing collaborative identification system for road height restriction obstacles, characterized in that, The steps for implementing the method according to any one of claims 1 to 9 include: The route planning unit is used to plan feasible routes based on the origin and destination points of freight vehicles uploaded by them, and to obtain the traffic road network. The information acquisition and cropping unit is used to crop elevation information blocks and optical image blocks within the spatial range of the road network. The height restriction obstacle preliminary selection unit is used to perform pixel-level elevation statistics in the elevation information block sliding window according to the road segment connection direction of the road network, and locate multiple potential height restriction obstacle candidate points. The height restriction obstacle identification unit is used to project the multiple potential height restriction obstacle candidate points onto the optical image block, and perform multi-temporal obstacle multi-source feature collaborative identification to screen and locate M height restriction obstacles; The reference height calculation unit is used to calculate and correct the clearance height of the M height-restricted obstacles in the elevation information block, and output the M reference height restrictions. The passage height compensation update unit is used to perform negative height limit compensation based on the vehicle type and real-time load information of the freight vehicle, and output dynamic passage height. The feasible route filtering and output unit is used to project the M reference height limits onto the road network and then use the dynamic road height to traverse the road network to locate R feasible routes; The route display and transmission unit is used to display the R feasible routes graphically on the vehicle-mounted display screen of the freight vehicle.

Images