Methods and systems for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections

Abstract

This disclosure provides a method and system for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections, relating to the field of road status perception technology. The method includes: acquiring infrared and hyperspectral images of the road; employing a target detection followed by local fusion strategy; performing hyperspectral registration and cross-modal fusion on structurally abnormal areas detected by infrared detection; performing fine matching with standard targets in a database; and combining frame-to-frame verification and anomaly detection mechanisms to achieve high-precision and robust tunnel section positioning. Based on the segment number corresponding to the positioning result and the current timestamp, historical traffic status data of the current segment and several previous set segments, as well as real-time vehicle speed, are obtained. The historical traffic status data of the current segment and the real-time vehicle speed are jointly input into a status judgment model, and the status judgment result is output. This disclosure combines frame-to-frame spatiotemporal consistency and multimodal interference removal mechanisms to achieve stable, accurate, and deployable segment positioning capabilities in complex tunnel environments.

Description

Technical Field

[0001] This disclosure relates to the field of road condition perception technology, specifically to a method and system for vehicle positioning and traffic condition alerts in ultra-long undersea tunnel sections. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the development of intelligent transportation and vehicle-mounted sensing technologies, the accuracy requirements for vehicle positioning and traffic condition perception are constantly increasing. In conventional road environments, GNSS-based positioning systems (such as GPS and BeiDou), inertial navigation systems (INS), and roadside cameras are relatively mature sensing methods. However, in scenarios such as ultra-long undersea tunnels, enclosed tunnels, and spaces under viaducts, due to factors such as satellite signal obstruction, complex electromagnetic environments, and poor lighting conditions, the aforementioned traditional methods often face problems such as low positioning accuracy, signal interruption, and visual blurring failure, making it difficult to meet the needs of safe passage and real-time traffic information prompts.

[0004] Currently, some studies have attempted to introduce vehicle-mounted vision-assisted positioning methods, such as using RGB cameras to acquire images of the road ahead and matching road segments through image recognition. However, due to the uniformity of tunnel pavement materials, unstable lighting, and slight differences in road segments, RGB images have significant limitations in fine road segment division and accurate recognition, especially at night or in low visibility environments, where the stability and robustness of recognition are poor.

[0005] Furthermore, existing image recognition systems generally neglect potential interference factors in the road environment (such as fallen objects, oil stains, construction obstructions, etc.). Such interference directly affects the accuracy of image feature extraction and positioning, increases the risk of false positives and false negatives, and reduces the system's engineering applicability. While some positioning enhancement solutions (such as laying markers and deploying sensors) can improve recognition accuracy to some extent, they usually involve modifications to the road structure or the addition of devices, resulting in high deployment costs, poor versatility, and unsuitability for most existing tunnels and complex environments.

[0006] With the development of deep learning, existing methods and devices for traffic prediction have emerged. These methods train a prediction model based on historical traffic data of a road segment, first-time traffic data, and first-time traffic data of adjacent road segments to predict the traffic status of the road segment. However, this method is implemented under the condition that the location of the road segment is known, and does not involve real-time positioning of the road segment. Secondly, one existing method provides a road vehicle-assisted positioning method based on road surface undulation markings. This method locates vehicles by laying undulation markings on the road surface and utilizing the regular up-and-down undulations generated by vehicles during travel. However, this method achieves vehicle positioning by laying undulation markings on the road segment, which cannot achieve non-destructive testing and does not have the function of predicting traffic flow. Another method is based on the fusion of hyperspectral and infrared images using a graph Laplacian model. It utilizes local kernel ridge regression to construct a nonlinear mapping relationship between images and generates a fused image through manifold regularization and energy minimization. However, this method only focuses on the image fusion algorithm itself and does not involve key steps such as image alignment and registration; furthermore, this method is largely theoretical and not geared towards specific application scenarios. Summary of the Invention

[0007] To address the aforementioned issues, this disclosure proposes a method and system for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections. It constructs a standard road section database, filters candidate regions with structural differences based on echo response characteristics, and then performs target detection on the continuous infrared images of the candidate regions using a target detection algorithm based on structural saliency rules. By leveraging the thermal radiation continuity and local gradient changes in the infrared images, it extracts ROI regions with road structural characteristics and matches them with hyperspectral image features to achieve high-precision current road section identification and positioning. Furthermore, it integrates vehicle wheel speed information and historical traffic data, and uses a spatiotemporal analysis model to comprehensively assess the traffic status of the current and forward road sections. Finally, it provides driver assistance feedback through image prompts and voice broadcasts.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] Methods for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections include:

[0010] Pre-build a pre-defined image-traffic status data index library for specific road segments;

[0011] Real-time road infrared and hyperspectral images are acquired and preprocessed. The preprocessed infrared texture map and spectral feature image are stitched together, and the overlapping area is registered using the feature point-based stitching RANSAC algorithm to obtain continuous infrared and continuous spectral images.

[0012] Candidate regions with structural differences in the road surface structure are screened based on the echo response mechanism. The target detection algorithm based on the structural saliency rule is used to detect targets in the continuous infrared image of the candidate regions. By using the thermal radiation continuity and local gradient changes of the infrared image, ROI regions with road structure characteristics are extracted.

[0013] Based on the ROI region and the spatial mapping relationship of the infrared-hyperspectral camera, registration is performed with the corresponding region of the hyperspectral image as the reference to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on the infrared-hyperspectral target region pairs to obtain local target feature vectors. The fusion value is evaluated by the cross-modal feature complementarity quantification index to obtain key fusion features.

[0014] The key fusion features are matched with the pre-generated local target feature vectors in the data index at the target level, and a comprehensive matching score is calculated. The database road segment number with the highest score is selected as the preliminary positioning result.

[0015] Based on the road segment number corresponding to the location result and the current timestamp, obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time vehicle speed. Input the historical traffic status data of the current road segment and the real-time vehicle speed into the status judgment model, output the status judgment result, and provide status prompts.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] The vehicle positioning and traffic status alert system for ultra-long undersea tunnel sections includes:

[0018] The initialization module is used to pre-build a pre-defined road segment image-traffic status data index library;

[0019] The image acquisition and preprocessing module is used to acquire real-time road infrared images and hyperspectral images, preprocess them, stitch the preprocessed infrared texture map and spectral feature image together, and use the feature point-based stitching RANSAC algorithm to register overlapping regions to obtain continuous infrared images and continuous spectral images.

[0020] The localization module is used to filter candidate regions with structural differences in the road surface structure based on the echo response mechanism. A target detection algorithm based on structural saliency rules performs target detection on the continuous infrared images of the candidate regions. By leveraging the thermal radiation continuity and local gradient changes of the infrared images, Regions of Interest (ROIs) with road structural characteristics are extracted. Based on the ROI regions and the spatial mapping relationship between the infrared and hyperspectral cameras, registration is performed using the corresponding regions in the hyperspectral images as a benchmark to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on these pairs to obtain local target feature vectors. The fusion value is evaluated using a cross-modal feature complementarity quantification index to obtain key fusion features. The key fusion features are then matched with pre-generated local target feature vectors in the data index at the target level, and a comprehensive matching score is calculated. The database road segment number with the highest score is selected as the preliminary localization result.

[0021] The status judgment module is used to obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time speed of the vehicle, based on the road segment number corresponding to the positioning result and the current timestamp. The historical traffic status data of the current road segment and the real-time speed of the vehicle are jointly input into the status judgment model, output the status judgment result, and provide status prompts.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] A computer program product includes a computer program that, when executed by a processor, implements the method for vehicle positioning and traffic status indication in the ultra-long undersea tunnel section.

[0024] According to some embodiments, the present disclosure adopts the following technical solutions:

[0025] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the method for vehicle positioning and traffic status indication in an ultra-long undersea tunnel section.

[0026] According to some embodiments, the present disclosure adopts the following technical solutions:

[0027] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the vehicle positioning and traffic status prompting method for the ultra-long undersea tunnel section.

[0028] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0029] This disclosed method for vehicle positioning and traffic status alerts in ultra-long undersea tunnels integrates ultrasonic, infrared, and hyperspectral multimodal information, possessing comprehensive advantages of high precision, low redundancy, and strong robustness. It simultaneously acquires infrared and hyperspectral images, extracts road surface structure and spectral features, and stitches multiple frames together to enhance representation. Real-time ultrasonic detection of road surface structural differences prioritizes candidate regions with significant features, serving as trigger conditions for the image fusion process. An innovative "target detection first, then local fusion" strategy is adopted, performing hyperspectral registration and cross-modal fusion only on structurally abnormal regions to avoid redundancy interference across the entire image. The fusion results are structurally matched with standard targets in the database, combined with frame-to-frame verification and anomaly detection mechanisms to achieve high-precision and robust tunnel segment positioning. Based on the identified road segment number and current time, historical traffic status is queried, and combined with real-time vehicle speed calculated by a wheel speed encoder, the system determines whether congestion, high risk, or critical fluctuation states exist. Driving prompts, including visual identifiers and voice announcements, are generated based on the judgment results. Each prompt event records the road segment number, vehicle speed, prompt type, and user feedback for subsequent system optimization. The system also regularly resamples data and evaluates accuracy to ensure that the accuracy of road segment identification and traffic judgment remains stable at over 80%. If necessary, it can also resample and rebuild the database.

[0030] This disclosed method for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections is primarily applied to closed road environments with weak GPS or other positioning signals and limited lighting and environmental perception. It is particularly suitable for special road conditions where traditional navigation methods are ineffective, such as ultra-long tunnels and undersea tunnels. This method enables precise vehicle positioning within road sections, real-time traffic status alerts, and intelligent early warnings of abnormal situations. By using an onboard infrared thermal imager and hyperspectral camera to collect information from the road surface from a long distance, without the need for physical markers or sensors on the road surface, and without relying on road reconstruction or modification, it achieves efficient perception and positioning support for road section features, exhibiting good engineering applicability and deployment flexibility.

[0031] The vehicle positioning and traffic status indication method disclosed herein for ultra-long undersea tunnels utilizes the temperature distribution characteristics of infrared images and the material reflection characteristics of hyperspectral images to construct a multimodal fusion model. This model extracts weak but stable "invisible" physical differences (such as thermal inertia and spectral response) between road segments, effectively overcoming the difficulty of recognizing RGB images in low-light and highly similar road surface environments, and significantly improving the positioning robustness and accuracy in complex scenarios such as tunnels.

[0032] This disclosed method for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections proposes a "target-guided dynamic local fusion" mechanism, overcoming the limitations of traditional whole-image fusion methods in terms of computational efficiency and anti-interference capability. The system detects structurally anomalous target regions using infrared images and guides hyperspectral images to perform sub-pixel-level registration and cross-modal feature fusion within key regions, significantly improving the targeting and effectiveness of the fusion. Simultaneously, it introduces thermal-spectral joint confidence scoring and feature complementarity evaluation to ensure the fused region possesses actual discriminative power and avoids interference from invalid information. This method, combined with spatiotemporal consistency matching between consecutive frames and a multimodal interference removal mechanism, achieves more stable, accurate, and deployable road segment positioning capabilities in complex tunnel environments.

[0033] This disclosed method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections implements a precise synchronous sampling mechanism based on wheel speed encoders. It utilizes wheel speed sensors to achieve equidistant sampling of infrared and spectral images, rather than equidistant sampling, thus addressing the impact of vehicle speed variations on image sampling uniformity. The image spacing is calculated using a formula to ensure a 30% overlap rate. Simultaneously, encoder pulse control precisely triggers infrared and spectral image sampling, achieving precise hardware-level synchronization at the sampling level. This provides a high-precision data foundation for subsequent image stitching and comparison.

[0034] This disclosure presents a method for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections, proposing a dynamic database self-learning and update mechanism. The system does not rely on a static template library but instead uses a "candidate update pool" to automatically replace images in the database when their quality is determined to be higher than the templates. A fusion mechanism for image quality assessment and historical version management is designed, enabling the database to self-evolve and self-optimize. The dynamically iterable image template library allows the system to adapt to changes in road conditions, aging, and pollution.

[0035] The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections disclosed herein identifies and locates abnormal regions in the images through joint analysis of temperature features in infrared images and spectral features in hyperspectral images. Then, a mask removal and image restoration unit is used to cover up and repair the abnormal regions in subsequent image processing to improve the overall recognition robustness and system adaptability. Attached Figure Description

[0036] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0037] Figure 1 Flowchart for constructing the image-traffic state data index library according to embodiments of this disclosure;

[0038] Figure 2This is a flowchart illustrating the fusion of infrared and hyperspectral images according to an embodiment of the present disclosure;

[0039] Figure 3 This is a flowchart illustrating road segment positioning in an embodiment of this disclosure. Detailed Implementation

[0040] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0041] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0042] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] Example 1

[0044] One embodiment of this disclosure provides a method for vehicle positioning and traffic status alerts in an ultra-long undersea tunnel section, the steps of which include:

[0045] Step 1: Pre-build a map-traffic status data index library for the designated road segments;

[0046] Step 2: Acquire real-time road infrared and hyperspectral images, preprocess them, stitch the preprocessed infrared texture map and spectral feature image together, and use the feature point-based stitching RANSAC algorithm to register overlapping regions to obtain continuous infrared and continuous spectral images.

[0047] Step 3: Based on the echo response mechanism, candidate regions with structural differences in the road surface structure are screened out. Based on the target detection algorithm of the structural saliency rule, the continuous infrared image of the candidate region is used to detect the target. By the thermal radiation continuity and local gradient change of the infrared image, the ROI region with road structure characteristics is extracted.

[0048] Step 4: Based on the ROI region and the spatial mapping relationship of the infrared-hyperspectral camera, registration is performed with the corresponding region of the hyperspectral image as the reference to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on the infrared-hyperspectral target region pairs to obtain local target feature vectors. The fusion value is evaluated using the cross-modal feature complementarity quantification index to obtain key fusion features.

[0049] Step 5: Perform target-level feature matching between the key fusion features and the pre-generated local target feature vectors in the data index, calculate the comprehensive matching score, and select the database road segment number with the highest score as the preliminary positioning result;

[0050] Step 6: Based on the road segment number corresponding to the location result and the current timestamp, obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time vehicle speed. Input the historical traffic status data of the current road segment and the real-time vehicle speed into the status judgment model, output the status judgment result, and provide a status prompt.

[0051] As one embodiment, this disclosure discloses a method for vehicle positioning and traffic status alerts in ultra-long undersea tunnel sections. Based on an echo response mechanism, candidate regions with structural differences in the road surface structure are selected. An ultrasonic detection mechanism utilizing structural reflection characteristics is used to initially determine whether structurally abnormal regions exist in the current road section, serving as a trigger condition for subsequent image fusion. Furthermore, a target detection algorithm based on structural saliency rules is employed to detect targets in the continuous infrared images of the candidate regions. By leveraging the thermal radiation continuity and local gradient changes in the infrared images, Regions of Interest (ROIs) with road structural characteristics are extracted. An innovative "target detection first, then local fusion" strategy is adopted, performing hyperspectral registration and cross-modal fusion only on structurally abnormal regions to avoid redundancy interference in the entire image. The fusion result is structurally matched with standard targets in the database and compared with a pre-constructed standard road section database to achieve high-precision current road section identification and positioning. Based on the positioning results, current vehicle speed, and historical traffic data, a comprehensive assessment of the current and forward road conditions is performed, and driver assistance feedback is provided through image prompts and voice broadcasts. The specific implementation process is as follows:

[0052] Step 1: Pre-build a pre-defined road segment image-traffic status data index library, including:

[0053] The designated road segment is divided into segments, and each segment unit is numbered sequentially. Road images, road surface spectral data, and historical traffic data are acquired. The historical traffic data includes vehicle speed and congestion probability. Each image sequence is bound to the road segment number, infrared image sequence, hyperspectral image sequence, acquisition time, corresponding road marker location signal, and historical traffic data to construct an image-traffic status data index library corresponding to the road segment number, image spectral data, and historical traffic data.

[0054] Specifically, step 1-1: Divide the tunnel into equal intervals according to a fixed length, with each 50-meter segment as a unit, and number them sequentially in the format: Tunnel-XXX, where XXX is an integer number (such as Tunnel-001, Tunnel-002, etc.). Each numbered segment is bound to its start and end mileage coordinates, which serve as the main index for the image and traffic information of that segment.

[0055] Steps 1-2: An infrared image acquisition unit and a spectral image acquisition unit are installed at the rear of the image acquisition vehicle. The thermal infrared imager captures the road surface temperature distribution and texture, while the hyperspectral camera captures the spectral reflectance characteristics of the road surface. Both devices are coaxially mounted, with their fields of view calibrated and overlapped to ensure a consistent sampling area.

[0056] Steps 1-3: The installation height of the image acquisition device is... The angle of inclination is Then the ground projection length (coverage length) D of a single frame image is:

[0057]

[0058] To ensure the accuracy of subsequent image stitching, a certain overlap rate between consecutive images is required. Assuming an image overlap rate of 30%, the continuous sampling interval is:

[0059]

[0060] That is, every time the vehicle moves forward X meters, it completes the acquisition of one infrared image and one hyperspectral image, ensuring continuous stitching without any breaks.

[0061] Steps 1-4: To achieve consistent image spacing and synchronized sampling, the device is equipped with a photoelectric wheel speed encoder, installed on the non-drive wheels, which outputs pulse signals. This encoder has the following characteristics: it outputs pulse signals, producing N pulses per revolution; the pulse frequency and cumulative pulse count can be read via a microcontroller; and when combined with the vehicle's tire radius R, it can accurately calculate the vehicle's travel distance. Every time the vehicle travels a distance L, the encoder will output... One pulse.

[0062] Based on the preset sampling interval X, the current total number of accumulated pulses inside the controller is ΣP. Whenever: This triggers a sampling event, after which the controller counter resets to zero, and the next sampling cycle begins.

[0063] Steps 1-5: The controller outputs a continuous 5ms high-level pulse through the GPIO interface, connecting to the Trigger IN of the infrared thermal imager and the Trigger IN of the hyperspectral camera. Both devices are set to "external trigger mode," receiving signals and sampling immediately to achieve millisecond-level synchronization, avoiding splicing misalignment or spatial mismatch issues caused by time deviations.

[0064] Steps 1-6: Bind each acquired image sequence to the following data items: road segment number Tunnel-XXX, infrared image sequence, hyperspectral image sequence, image acquisition time and corresponding road marker location information, and historical traffic data (average vehicle speed; congestion probability; frequency of historical accidents or abnormal events). The time label consists of [weekday + holiday / weekday + time period], with off-peak or low-peak periods divided into 15-minute intervals and peak periods divided into 5-minute intervals.

[0065] This completes the process of pre-building a pre-defined image-traffic status data index library for specific road segments.

[0066] Step 2: Acquire real-time road infrared and hyperspectral images, preprocess them, stitch the preprocessed infrared texture map and spectral feature image together, and use the feature point-based stitching RANSAC algorithm to register overlapping regions to obtain continuous infrared and continuous spectral images.

[0067] Specifically, considering the actual tunnel environment, the visual differences between different road sections are very limited—the road surface materials are uniform, and the lighting conditions are similar, making it almost impossible to distinguish between 50-meter sections using traditional RGB visual information. However, differences still exist at the fine-grained level in the road surface, such as: local structural texture variations (construction joints, repair layers); material aging, roughness, and pollutant residue; asphalt oxidation, dust deposition, and water stain distribution; and "invisible" differences in spectral reflectance or thermal inertia on the road surface. These differences exhibit certain regularities in infrared and hyperspectral images and possess discriminative power. Therefore, an infrared image acquisition unit and a hyperspectral image acquisition unit are installed behind the vehicle to acquire infrared and hyperspectral images. The method described in step 1 is used to ensure consistent image spacing and field of view, facilitating subsequent image stitching. A fusion algorithm is then used to deeply fuse the two types of image information, thereby completing the identification of road sections.

[0068] First, step 2-1: Preprocess the acquired infrared image. Use a high-pass filter (such as Laplacian) for edge enhancement to highlight texture seams, cracks, and other structures. Use a temperature gradient enhancement algorithm (differential temperature map) to calculate the temperature difference between each pixel and its neighboring pixels, generating a local thermal heterogeneous map. Perform histogram equalization on the local thermal heterogeneous map to unify brightness differences at different acquisition times and enhance contrast. The final output is a 1-channel normalized infrared texture map. .

[0069] Furthermore, the hyperspectral image is preprocessed, and principal component analysis (PCA) is used to reduce the image dimensionality: the original hyperspectral image is denoted as... , For wavelength; PCA is used to decompose the spatial covariance of all bands, and the first three principal components are extracted. Corresponding to the directions of maximum variation, these three principal components form a pseudo-RGB image, which serves as the final spectral feature image. .

[0070] Step 2-2: Image stitching is performed. Since a single image frame (covering an area of ​​only a few meters) is insufficient to represent the features of the entire segment, image stitching is used to enhance the recognition effect. To avoid excessive computational overhead due to stitching the entire 50m segment, the number of stitched frames is specified as N. Based on the single-frame image coverage length specified in Step 1 as D, the sampling interval is... Then the total coverage length of the stitched image is:

[0071]

[0072] That is, the stitched image is approximately the length of a single frame. times.

[0073] As one embodiment, the specific steps of image stitching include: recording the travel distance using a vehicle wheel speed encoder; acquiring the current frame and the previous N-1 frames according to the image sampling order; registering overlapping regions using a feature-point-based stitching RANSAC algorithm; performing vertical stitching according to the ground travel direction; and outputting a continuous infrared image. Continuous spectrum .

[0074] Steps 2-3: Based on the echo response mechanism, candidate regions with structural differences in the pavement structure are selected.

[0075] Specifically, a candidate region screening mechanism driven by sound waves is employed. A simple ultrasonic sensor installed on the underside of the vehicle emits short-pulse high-frequency sound wave signals (such as narrow-band pulses around 40kHz) in real time, perpendicularly illuminating the road surface along the current driving path. By measuring the time delay distribution, amplitude variation, and spectral characteristics of the echo signal, the system can identify subtle differences in the road material structure. For example, when the sound wave encounters areas with different density, material, or surface roughness (such as joints, repairs, minor cracks, or embedded objects), the echo signal will exhibit characteristics such as enhanced energy reflection, abnormal attenuation, or spectral distortion. The system uses a lightweight anomaly detection algorithm (such as statistical rate of change threshold judgment within a sliding window) to quickly analyze the sound wave echo signal. Once a difference in sound wave response is detected in the current area, it is determined that the location has structural heterogeneity.

[0076] Such structural heterogeneity is not equivalent to severe damage, but rather a normal variation in conventional pavement structures. However, its acoustic response characteristics, which are significantly different from those of adjacent areas, can serve as an "image processing trigger point." The system then initiates the fusion processing of infrared and hyperspectral images of that area, skipping redundant areas that lack variability, significantly reducing ineffective processing, and improving the efficiency of sensing resource scheduling and the response speed of the processing chain.

[0077] By using ultrasonic waves to detect differences in road surface structure in real time, candidate regions with significant features are selected first and used as the triggering condition for the image fusion process.

[0078] Steps 2-4: Target Region Extraction and Dynamic ROI Generation Based on Structural Saliency Rules. At locations initially identified by the ultrasonic module as having structural differences, the system further performs physical prior-driven target detection on the infrared images of the corresponding candidate regions. This disclosure proposes a "structural saliency-guided" target detection method that requires no depth model and is entirely based on the physical prior of infrared images, achieving low-cost and highly robust target extraction in complex tunnel scenarios. The core idea of ​​this method is to mine regions with road structural features (such as cracks, joints, oil stains, potholes, strong reflections, etc.) through the thermal radiation continuity and local gradient changes of infrared images, and construct a set of dynamic, high-confidence ROI regions for subsequent fusion processing. Specific steps are as follows:

[0079] (1) Preliminary salient region extraction: Input the continuous infrared image corresponding to the candidate region Calculate its temperature gradient diagram This is used to characterize the intensity of local thermal changes at individual pixels. A threshold is set. Extract the set of pixels whose temperature gradient is higher than a threshold. This serves as the initial mask for the candidate target region. The process requires no parameter learning and relies entirely on the saliency of the thermal structure.

[0080] (2) Clustering of candidate structural regions: Clustering of the set of pixels with significant temperature gradients obtained in step (1). Connectivity region extraction is performed using a connectivity analysis method based on four-neighbor or eight-neighbor rules to identify spatially interconnected hot salient regions, and each connected region is defined as a candidate target region. Then, the statistical properties of each region (such as area, mean temperature, standard deviation of temperature, aspect ratio of the boundary, principal orientation angle, etc.) are calculated, and edge operators (such as Laplacian or Canny) can be combined to enhance the continuity and stability of the boundary profile.

[0081] (3) Multi-indicator target significance scoring: for each candidate target region Define a multi-factor confidence scoring function that comprehensively considers the mean of its thermal gradient. Regional temperature distribution concentration ( Information such as boundary smoothness and spatial dimension ratio is included. The specific scoring function is as follows:

[0082]

[0083] in, Score the saliency of the nth target region; The mean temperature gradient of the target region; It is the ratio of the standard deviation to the mean of the regional temperature, used to characterize the concentration of heat distribution; This is the boundary smoothness factor, reflecting the continuity and regularity of the target contour; The aspect ratio or area percentage of a region is used to constrain its spatial geometry. As an experience-weighted factor, satisfying This scoring function requires no training; it determines the threshold and weighting factors through empirical rules and domain knowledge, and possesses good interpretability.

[0084] (4) Dynamic ROI generation: based on confidence score With set threshold Select a set of high-confidence target regions. Each region is generated based on structural saliency, possessing stable spatial positioning and physical meaning.

[0085] The method disclosed herein enables the further refinement of high-confidence target regions within the structural change regions initially screened by ultrasound using infrared structural saliency rules, effectively avoiding the invalid processing of redundant background regions. It also achieves step-by-step convergence from acoustic anomaly perception to infrared image structural analysis, constructing a low-cost, highly robust local perception chain, laying the foundation for subsequent cross-modal fusion and tunnel segment identification. Its output can serve as the sole input region for subsequent hyperspectral registration and cross-modal fusion, fundamentally improving fusion effectiveness and reducing unnecessary computational overhead.

[0086] Steps 2-5: Local Region Extraction and Fine Registration. Based on the target region set detected in Step 2-4. Further local region mapping and sub-pixel level registration are performed, the specific process is as follows:

[0087] (1) For each infrared target region Based on the spatial mapping relationship of the infrared-hyperspectral camera obtained from the previous calibration, the location of the target region in the hyperspectral image is determined.

[0088] (2) Using the corresponding region of the hyperspectral image as a reference, the mutual information maximization registration algorithm is executed at a small scale to accurately search for the local optimal displacement vector so that the target region of the infrared image is strictly aligned with the corresponding region of the hyperspectral image.

[0089] (3) After registration, a precisely aligned infrared-hyperspectral target region pair is formed:

[0090]

[0091] Steps 2-6: Target-guided dynamic local fusion mechanism. Cross-modal feature fusion is performed on the registered target regions from Steps 2-5, significantly improving fusion efficiency and region specificity. The specific steps are as follows:

[0092] (1) Within the target area of ​​the infrared image, feature information based on thermal radiation characteristics is extracted, specifically including: temperature distribution statistics (such as regional average temperature, temperature standard deviation, and range), thermal distribution concentration (such as entropy or energy statistics), and thermal edge structure features obtained through edge detection algorithms (such as Canny or Sobel) (such as total edge length, principal orientation angle, curvature distribution, and shape complexity). The above features are used to characterize the thermal and geometric properties of the thermal anomaly region;

[0093] (2) In the region corresponding to the infrared target region in the hyperspectral image, the statistical features (such as mean and standard deviation) of the first few principal component images (e.g., the first 3 to 5) are extracted using the principal component analysis (PCA) method. At the same time, representative key bands (such as typical material absorption peaks) are selected from the original spectral data to extract reflectance features. The spectral angle (SAM) between the original and standard material spectral library is calculated as a material identification feature to comprehensively describe the spectral information features in the region.

[0094] (3) The feature information of the two types of data is concatenated to form a fused local target feature vector:

[0095]

[0096] in, It represents the temperature characteristics of the target area in the infrared image, including the average temperature of the area, the standard deviation of the temperature, the range, and the concentration of heat distribution. The edge structure features of the target region, such as edge length, principal orientation angle, average curvature and complexity, reflect its geometric shape; It represents the spectral characteristics of the corresponding region in the hyperspectral image, including the mean statistical value of the principal component image, the reflectance of a specific key band, and the spectral angle (SAM) between the spectrum and the standard material spectrum. The geometric structure information of the target includes the area, aspect ratio, and centroid coordinates.

[0097] Furthermore, the aforementioned set of local target feature vectors is denoted as:

[0098]

[0099] Furthermore, to further enhance the effectiveness of the fusion region, this disclosure designs a cross-modal feature complementarity quantification index after constructing local fusion features. This index is used to dynamically evaluate the fusion value of each target region, avoiding information redundancy or ineffective fusion. Specifically, a thermal-spectral coupling coefficient is introduced. As a complementarity evaluation function, it is defined as follows:

[0100]

[0101] in, This represents a temperature gradient map of the infrared target region. This represents the intensity gradient map of the principal components of a hyperspectral image. This is the Pearson correlation coefficient. A value approaching 0 indicates a large difference and strong complementarity between infrared and hyperspectral modes in this region; while if... If the threshold is too high, it indicates that the changes in the two modes are converging, limiting the fusion value. The system can set a complementarity threshold. ,when When this area is identified as a fusion redundancy area, it can be selectively downweighted or eliminated to ensure that subsequent matching and positioning processes rely only on key area features with cross-modal gain, thereby improving overall recognition stability and anti-interference capability.

[0102] Unlike traditional full-image fusion strategies, this disclosure proposes a "target-guided dynamic local fusion" mechanism. By detecting targets in preceding infrared images, a set of high-confidence Regions of Interest (ROIs) is generated in real time. Hyperspectral registration and fusion operations are performed only on these ROIs. A thermal-spectral joint confidence scoring mechanism is introduced to determine the fusion weight and confidence of each region, reducing redundant computation and improving the effectiveness and anti-interference capability of the fusion.

[0103] Step 2-7: Target matching and localization based on region fusion features. Based on the key region feature set of the current frame obtained in Step 2-6, match and identify the target region fusion features pre-generated for each road segment in the standard database:

[0104] (1) The standard database pre-constructs the fusion feature set of each road segment using the same process. ;

[0105] (2) Perform target-level feature matching to determine the feature distance (such as cosine distance or Euclidean distance) between each target in the current frame and the most similar target in the database.

[0106] (3) Construct the feature matching results into a target graph (with the target region as the node and the spatial position relationship between the nodes as the edge);

[0107] (4) Taking into account the feature similarity and spatial structure consistency between targets, calculate the comprehensive matching score:

[0108]

[0109] in, The feature similarity score represents the score for each fused feature vector in the current image frame. The inverse ratio to the distance to the most similar target in the database can be calculated using cosine similarity or Euclidean distance weighted average, reflecting the consistency of the fused features themselves; Spatial structure consistency scoring represents the degree of similarity in spatial arrangement (such as distance between nodes, relative angle, topology, etc.) between the target area in the current frame and the standard road segment. This can be achieved through a structural similarity measurement function between node graphs. The completeness score represents the degree of matching coverage between the target set and the standard template in the current frame, such as the ratio of the number of successfully matched targets to the total number of targets; α, β, and γ are the weight coefficients of the above three scoring components, which are used to balance the influence of feature similarity, structural consistency and matching coverage in the total score, satisfying α+β+γ=1.

[0110] Finally, the database road segment number with the highest score was selected as the preliminary location result.

[0111] Steps 2-8: To further improve the robustness and accuracy of positioning, if the preliminary positioning results have uncertainties or score thresholds, then perform candidate region expansion verification:

[0112] (1) Starting from the area where the initial target is located, expand the search to the neighboring area to find other structural feature targets;

[0113] (2) Perform the target detection → cross-modal fusion → feature matching process again on the extended region;

[0114] (3) Verify the spatial structure consistency between the extended target regions and confirm the stability and consistency of feature matching before and after;

[0115] (4) If the verification is successful, the preliminary matching result is confirmed; otherwise, the matching process is re-executed.

[0116] Steps 2-9: Joint Comparison Mechanism Between Previous and Previous Frames. After locating the current frame image, the location result of the previous frame (the previous sampling point) is simultaneously acquired. If the current segment and the previous segment simultaneously meet the following conditions: the similarity with the corresponding numbered segment image in the database is higher than a set threshold (e.g., 85%); the numbers of the previous segment and the current segment are consecutive in the numbering order; the timestamp and the driving distance are consistent (determined by the wheel speed encoder); and the comparison result of a continuously stitched image is higher than a unified threshold, then the location result is considered reliable and confirmed as the tunnel segment number where the current vehicle is located. If only the current segment or the previous segment matches, an additional frame of data needs to be added for judgment (sliding window mechanism).

[0117] Steps 2-10: Interference Cancellation and Abnormal Image Processing Mechanism. Actual tunnel pavement images often contain interference factors such as debris, construction equipment, and vehicle projections. These foreign objects may damage local structures or spectral characteristics, affecting positioning accuracy. To enhance the system's anti-interference capability, this disclosure designs the following multi-stage collaborative processing flow:

[0118] Interference detection methods: In infrared images, the temperature difference threshold method is used to detect local abnormal heat sources (such as patches with a temperature difference exceeding 3°C). Morphological processing (opening and closing operations) is used to remove small noise points, retaining only suspicious foreign object areas. In hyperspectral images, based on spectral angle mapping (SAM) or the anomaly detection algorithm RXDetector, abnormal areas of reflectance spectrum are identified and compared with the normal road surface spectrum library using cosine similarity. Areas below the threshold (such as 0.85) are considered abnormal. If the abnormal areas detected in the infrared and spectral images overlap in space, they are marked as "valid foreign object areas".

[0119] Mask Removal and Image Restoration: A mask is generated for the detected foreign object region; during the image comparison stage, the mask region is ignored, and image matching is performed only on the unmasked part; if the area of ​​the abnormal region exceeds the threshold (e.g., 15%) of the image area, the current frame image is invalidated and recognition is skipped; if the area is small, the system can use edge interpolation based on image context (e.g., Navier-Stokes restoration) or texture filling algorithm (e.g., PatchMatch) to complete the occluded area and improve image integrity.

[0120] S2-11: Dynamic database update mechanism. After successful localization (meeting the requirements of continuous comparison of two segments and no occlusion), the current image pair enters the "candidate update pool"; if the clarity and texture intensity are higher than the original template, it is automatically overwritten and updated; the new image is marked with a timestamp to form a traceable evolution history; a four-dimensional data structure of "time-road segment-image-traffic" is constructed for subsequent trend analysis or learning.

[0121] Step 3: Based on the road segment number corresponding to the location result and the current timestamp, obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time vehicle speed. Input the historical traffic status data of the current road segment and the real-time vehicle speed into the status judgment model, output the status judgment result, and provide a status prompt.

[0122] Specifically, step 3-1: Obtain the current location result as the road segment number Tunnel-XXX, and at the same time obtain the current timestamp T_now, and parse out the time tag. The time tag includes the week tag, such as Monday to Sunday; or the holiday tag: such as weekday / statutory holiday; and the time period tag: extract the current segment (such as 08:30~08:45) according to the set granularity (such as 15-minute segment).

[0123] In the traffic database, a composite index of {Tunnel-XXX + date tag + time period} is used to extract historical traffic status information for the current segment within that time period, including but not limited to: average vehicle speed, congestion probability, average transit time, frequency of abnormal events, and estimated effective lane capacity. Simultaneously, data from 2-3 consecutive road segments ahead are queried to predict traffic flow trends.

[0124] Step 3-2: Based on the pulse signal output by the photoelectric wheel speed encoder installed on the bottom of the vehicle for each wheel rotation, combined with the tire radius R and the number of pulses per revolution N, the distance the vehicle travels per unit time can be calculated, thereby estimating the vehicle speed in real time.

[0125]

[0126] in, for The number of pulses within the time window; The sampling period (e.g., 1 second); This represents the instantaneous speed of the vehicle.

[0127] Step 3-3: Calculate the average speed of historical traffic indicators for the current road segment. Congestion probability Real-time speed of the vehicle The joint input state judgment model sets up judgment logic to determine whether there is an abnormal state or an event that needs to be prompted, specifically including:

[0128] (1) Determining congestion or high-risk conditions:

[0129] If any of the following conditions are met, the current situation is considered to be congested or high-risk:

[0130] 1) And the duration is ≥10s;

[0131] 2) Current segment congestion probability ;

[0132] 3) The frequency of accidents in the current section or downstream section is ≥3 times / week;

[0133] 4) The actual test showed that the time taken to pass was more than 30% higher than the historical average.

[0134] (2) Smooth passage determination:

[0135] 1) Furthermore, the historical speed of vehicles on the two consecutive sections of road behind the intersection was ≥30km / h;

[0136] 2) Congestion probability ≤ 20%, accident frequency < 1 / week;

[0137] (3) Critical fluctuation state: The current segment is in normal condition, but the probability of congestion in the downstream segment increases.

[0138] Steps 3-4: Based on the comprehensive judgment results, generate multimodal prompts:

[0139] (1) Visual cues (central control display screen):

[0140] The current status is displayed by road segment number and image recognition. Red represents congestion or high-risk status, yellow represents critical status, and green represents smooth traffic status.

[0141] (2) Voice prompts: For example, “You are currently on Tunnel-026. The historical congestion probability for this period is 68%. The current vehicle speed is too low. It is recommended to slow down and observe the situation ahead.” or “The two sections ahead are expected to be clear. The vehicle speed is expected to be above 40km / h.”

[0142] As one implementation, each notification event is cached locally, recording the following information: road segment number, notification category (congestion / abnormal / suggestion), timestamp, current speed, and user feedback (e.g., "ignore notification," "discussion is accurate"). This feedback information can be used for later model optimization and backend data training. Offline caching on the vehicle side combined with periodic data uploads is supported.

[0143] The detection vehicle (data collection vehicle) periodically (e.g., monthly) completes a full run within the target tunnel, collecting a complete set of images along with actual location and traffic assessment data, which serves as the system's self-testing sample. The system compares the detection vehicle's current road segment identification results with the actual known segments; simultaneously, it compares the traffic state prediction with the actual vehicle speed / travel time, calculating: road segment positioning accuracy; and traffic state assessment accuracy. If either indicator is <80%, it is marked as "system degradation," and the process proceeds to step 1, resampling and updating the database.

[0144] Example 2

[0145] One embodiment of this disclosure provides a vehicle positioning and traffic status alert system for an ultra-long undersea tunnel section, including:

[0146] The vehicle positioning and traffic status alert system for ultra-long undersea tunnel sections includes:

[0147] The initialization module is used to pre-build a pre-defined road segment image-traffic status data index library;

[0148] The image acquisition and preprocessing module is used to acquire real-time road infrared images and hyperspectral images, preprocess them, stitch the preprocessed infrared texture map and spectral feature image together, and use the feature point-based stitching RANSAC algorithm to register overlapping regions to obtain continuous infrared images and continuous spectral images.

[0149] The localization module is used to filter candidate regions with structural differences in the road surface structure based on the echo response mechanism. A target detection algorithm based on structural saliency rules performs target detection on the continuous infrared images of the candidate regions. By leveraging the thermal radiation continuity and local gradient changes of the infrared images, Regions of Interest (ROIs) with road structural characteristics are extracted. Based on the ROI regions and the spatial mapping relationship between the infrared and hyperspectral cameras, registration is performed using the corresponding regions in the hyperspectral images as a benchmark to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on these pairs to obtain local target feature vectors. The fusion value is evaluated using a cross-modal feature complementarity quantification index to obtain key fusion features. The key fusion features are then matched with pre-generated local target feature vectors in the data index at the target level, and a comprehensive matching score is calculated. The database road segment number with the highest score is selected as the preliminary localization result.

[0150] The status judgment module is used to obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time speed of the vehicle, based on the road segment number corresponding to the positioning result and the current timestamp. The historical traffic status data of the current road segment and the real-time speed of the vehicle are jointly input into the status judgment model, output the status judgment result, and provide status prompts.

[0151] Example 3

[0152] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the vehicle positioning and traffic status prompting method for ultra-long undersea tunnel sections.

[0153] Example 4

[0154] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the vehicle positioning and traffic status prompting method for ultra-long undersea tunnel sections.

[0155] Example 5

[0156] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the vehicle positioning and traffic status prompting method for the ultra-long undersea tunnel section.

[0157] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0158] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0159] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections, characterized in that, include: Pre-build a pre-defined road segment image-traffic status data index library; Real-time infrared and hyperspectral images of the road are acquired and preprocessed to obtain continuous infrared and continuous spectral images. Candidate regions with structural differences in the road surface structure are screened based on the echo response mechanism. Target detection is performed on the continuous infrared images of the candidate regions based on the structural saliency rule. ROI regions with road structural features are extracted by utilizing the thermal radiation continuity and local gradient changes of the infrared images. Based on the ROI region and the spatial mapping relationship of the infrared-hyperspectral camera, registration is performed with the corresponding region of the hyperspectral image as the benchmark to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on the infrared-hyperspectral target region pairs to obtain local target feature vectors. The fusion value is evaluated using the cross-modal feature complementarity quantification index to obtain key fusion features. The key fusion features are then matched with pre-generated local target feature vectors in the data index library at the target level, and a comprehensive matching score is calculated. The database road segment number with the highest score is selected as the preliminary positioning result. Based on the road segment number corresponding to the location result and the current timestamp, obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time vehicle speed. Input the historical traffic status data of the current road segment and the real-time vehicle speed into the status judgment model, output the status judgment result, and provide status prompts.

2. The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections as described in claim 1, characterized in that, The designated road segment is divided into segments, and each segment unit is numbered sequentially. Road images, road surface spectral data, and historical traffic data are acquired. The historical traffic data includes vehicle speed and congestion probability. Each image sequence is bound to the road segment number, infrared image sequence, hyperspectral image sequence, acquisition time, corresponding road marker location signal, and historical traffic data to construct an image-traffic status data index library corresponding to the road segment number, image spectral data, and historical traffic data.

3. The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections as described in claim 1, characterized in that, Real-time infrared and hyperspectral images of the road are acquired. The infrared images are preprocessed, and a high-pass filter is used for edge enhancement to highlight texture seams and crack structures. A temperature gradient enhancement algorithm is used, which calculates the temperature difference between each pixel and its neighboring pixels to generate a local thermal heterogeneous map. Histogram equalization is used to unify the brightness differences at different acquisition times and enhance contrast. The final output is a single-channel normalized infrared image. The hyperspectral images are preprocessed, and principal component analysis is used to decompose the spatial covariance of all bands. The first three principal components are extracted and combined into a pseudo-RGB image as the final hyperspectral image.

4. The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections as described in claim 1, characterized in that, An ultrasonic sensor installed under the vehicle detects the reflection characteristics of the road surface structure and filters candidate areas with structural differences in real time. A target detection algorithm based on structural saliency rules performs target detection on the continuous infrared image. The continuous infrared image of the candidate area is input, its temperature gradient map is calculated, a threshold is set, and the set of pixels with a temperature gradient higher than the threshold is extracted as the initial mask of the candidate target area. Connected region clustering is performed on the pixel set to identify a series of spatially continuous target regions and calculate their statistical attributes. For each target region, a multi-factor confidence score function is defined, which comprehensively considers its mean thermal gradient, regional temperature distribution concentration, boundary smoothness, and spatial size ratio. Based on the confidence score and a set threshold, a set of high-confidence target regions is selected.

5. The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections as described in claim 1, characterized in that, The specific process of obtaining the local target feature vector is as follows: (1) For each infrared target region, based on the spatial mapping relationship of the infrared-hyperspectral camera obtained in the previous calibration, determine the position of the target region corresponding to it in the hyperspectral image; (2) Using the corresponding region of the hyperspectral image as a reference, execute the mutual information maximization registration algorithm at a small scale to accurately search for the local optimal displacement vector so that the target region of the infrared image is strictly aligned with the corresponding region of the hyperspectral image; (3) After registration, a pair of precisely aligned infrared-hyperspectral target regions is formed; (4) Cross-modal feature fusion is performed on the registered target region pair to extract the feature information in the infrared target region, including temperature distribution statistics and thermal edge structure features; extract the feature information in the corresponding hyperspectral region, and splice the feature information of the two data to form the fused local target feature vector.

6. The method for vehicle positioning and traffic status indication in ultra-long undersea tunnel sections as described in claim 1, characterized in that, The historical traffic status data of the current road segment is combined with the real-time vehicle speed and input into the status judgment model. The specific execution logic of the status judgment model includes: If any of the following conditions are met, the current situation is considered to be congested or high-risk: 1) And the duration is ≥10s; 2) Current segment congestion probability ; 3) The frequency of accidents in the current section or downstream section is ≥3 times / week; 4) The actual test showed that the time taken to pass was more than 30% higher than the historical average; Smooth passage determination: 1) Furthermore, the historical speed of vehicles on the two consecutive sections of road behind the intersection was ≥30km / h; 2) Congestion probability ≤ 20%, accident frequency < 1 / week; Critical fluctuation state: The current segment is in normal condition, but the probability of congestion in the downstream segment increases.

7. A vehicle positioning and traffic status alert system for ultra-long undersea tunnel sections, characterized in that, include: The initialization module is used to pre-build a pre-defined road segment image-traffic status data index library; The image acquisition and preprocessing module is used to acquire real-time road infrared and hyperspectral images, and preprocess them to obtain continuous infrared and continuous spectral images. The localization module is used to filter out candidate regions with structural differences in the road surface structure based on the echo response mechanism. The target detection algorithm based on the structural saliency rule performs target detection on the continuous infrared image of the candidate region. By using the thermal radiation continuity and local gradient changes of the infrared image, the ROI region with road structure characteristics is extracted. Based on the ROI region and the spatial mapping relationship of the infrared-hyperspectral camera, registration is performed with the corresponding region of the hyperspectral image as the benchmark to obtain infrared-hyperspectral target region pairs. Cross-modal feature fusion is performed on the infrared-hyperspectral target region pairs to obtain local target feature vectors. The fusion value is evaluated using the cross-modal feature complementarity quantification index to obtain key fusion features. The key fusion features are then matched with pre-generated local target feature vectors in the data index library at the target level, and a comprehensive matching score is calculated. The database road segment number with the highest score is selected as the preliminary positioning result. The status judgment module is used to obtain the historical traffic status data of the current road segment and the previous few set road segments, as well as the real-time speed of the vehicle, based on the road segment number corresponding to the positioning result and the current timestamp. The historical traffic status data of the current road segment and the real-time speed of the vehicle are jointly input into the status judgment model, output the status judgment result, and provide status prompts.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle positioning and traffic status prompting method for ultra-long undersea tunnel sections as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the vehicle positioning and traffic status prompting method for ultra-long undersea tunnel sections as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the vehicle positioning and traffic status prompting method for ultra-long undersea tunnel sections as described in any one of claims 1-6.

Images