Road disease detection method based on multi-temporal light inversion of low-altitude unmanned aerial vehicle and vehicle-mounted camera linkage

Abstract

The application discloses a kind of low-altitude unmanned aerial vehicle and vehicle camera linkage road disease detection method based on multi-temporal light shadow inversion.The method is through low-altitude unmanned aerial vehicle and vehicle camera cooperative collection, obtains the multi-temporal, multi-view image data of target section;Based on space-time synchronization and feature association, construct the spatial mapping of double-source data;Introduce light shadow inversion mechanism, use the geometric change of crack shadow under different solar elevation angles, and the real three-dimensional form and width information of crack are inverted, and the invisible implicit microcrack under normal illumination is identified;Finally, through fusion completion, form the complete road disease information set, realize high-precision disease identification and quantitative evaluation.The application takes "multi-temporal light shadow inversion + double-view linkage collection + feature association mapping" as core, breaks through the dependence of traditional visual detection on illumination condition, significantly improves microcrack detection capability and disease quantization precision.

Description

Technical Field

[0001] This invention belongs to the field of road inspection technology, specifically involving a road defect detection method based on multi-temporal light and shadow inversion and linkage between low-altitude UAVs and vehicle-mounted cameras. It is applicable to various road scenarios such as highways, urban roads, and rural roads. Through the deep integration of multi-temporal light and shadow inversion technology and dual-view collaborative acquisition, it achieves high-precision detection and three-dimensional quantitative assessment of hidden micro-cracks in roads, providing technical support for road traffic safety assurance and maintenance management. Background Technology

[0002] Road defect detection is a crucial aspect of road maintenance and management, and its effectiveness directly impacts the scientific rigor and timeliness of maintenance decisions. Traditional detection methods have significant shortcomings: manual detection is inefficient, highly subjective, and struggles to achieve both comprehensive coverage and detailed identification; while single-drone detection can acquire global road information and achieve full road segment coverage, it is limited by perspective and resolution, lacking detailed information on minor road surface defects, easily leading to missed detections; and while single-vehicle camera detection can capture fine details of local road surfaces, it is limited by driving trajectory, resulting in numerous spatial blind spots and failing to achieve complete detection of the entire road segment.

[0003] The core shortcomings of existing technologies are as follows: First, the information dimension is too narrow, and a single data source cannot simultaneously meet the dual requirements of full coverage and precise details, resulting in information gaps. Second, the data correlation is insufficient, and there is a lack of effective technical means to establish the correlation between data from different perspectives, making it impossible to achieve information complementarity. Third, the information integration is inefficient, and a scientific logic for fusion and supplementation has not been formed, making it difficult to fill the information limitations of each source of data, which leads to limitations in the completeness and accuracy of the detection results.

[0004] More importantly, existing visual inspection technologies generally suffer from a fundamental limitation—a passive dependence on lighting conditions. Tiny cracks, especially microcracks that are similar in color to the road surface, are easily overlooked under conventional single-image conditions; the true depth, opening width, and other three-dimensional morphological parameters of cracks cannot be obtained from a single image; and latent defects such as underground cavities and loose subgrade cannot be identified through surface visual inspection at all. Current technologies lack methods to actively utilize changes in lighting to enhance detection capabilities, resulting in a persistently high rate of missed detections for microcracks and a lack of quantitative basis for assessing the severity of defects. Summary of the Invention

[0005] In view of the shortcomings of the above-mentioned background technology, the present invention provides a method for detecting road defects by linking a low-altitude UAV and a vehicle-mounted camera based on multi-temporal light and shadow inversion.

[0006] The road defect detection method based on multi-temporal light and shadow inversion using a low-altitude UAV and vehicle-mounted camera includes linked acquisition, image preprocessing, feature association mapping, information fusion and completion, and defect identification and result output. The linked acquisition involves using a low-altitude UAV and a vehicle-mounted camera to collect global and local detail information of the target road segment, forming complementary dual-source data. This linked acquisition includes multi-temporal and multi-angle planned acquisition of the same road segment to obtain image sequences under different lighting conditions. The image preprocessing optimizes the dual-source data to improve information recognition. The feature association mapping extracts core features from the dual-source data and establishes correlations, constructing a global and local information mapping channel and extracting shadow features under different lighting conditions. The information fusion and completion, based on the correlation mapping relationship, fills the information gaps of a single data source through information complementarity logic, forming a complete road information set. A light and shadow inversion mechanism is introduced to invert the three-dimensional morphology and width information of cracks using shadow changes in multi-temporal images, identifying hidden micro-cracks invisible under normal lighting. The defect identification and result output uses intelligent algorithms to identify defect information from the complete information set and generate a standardized detection report.

[0007] The coordinated acquisition also includes multi-temporal acquisition planning: based on the orientation and geographical location of the target road segment, the drone and vehicle-mounted equipment are planned to repeatedly acquire images of the same road segment within multiple time windows with different solar altitude angles, thereby obtaining image sequences containing differences in light and shadow.

[0008] In the multi-temporal acquisition plan, the difference in solar altitude angle between different time windows is not less than a preset threshold to ensure that the crack shadow has significant geometric differences.

[0009] The feature association mapping also includes light and shadow feature extraction: extracting feature parameters of the length, direction, and width of the crack shadow as a function of the illumination angle from a multi-temporal image sequence, and constructing a light and shadow feature vector.

[0010] The light and shadow inversion mechanism includes: based on the geometric changes of crack shadows under different illumination angles, using an optical geometric model to infer the crack depth, opening width and cross-sectional shape, and using the inversion results as a supplement to local detail information to fill the lack of crack detail information in conventional images.

[0011] In the information fusion and completion process, the hidden microcracks identified through the light and shadow inversion mechanism are verified by feature association with the local detail images captured by the vehicle camera, confirming the real existence of the microcracks and accurately calibrating their spatial location.

[0012] In the disease identification and result output, the output detection report includes the three-dimensional morphological parameters of the cracks (depth, opening width, cross-sectional shape) and the distribution map of hidden microcracks.

[0013] The method also has a data quality verification function, which marks the collected data that does not meet the quality requirements, the data that has not established an effective association mapping, or the data that still has information defects after fusion and completion, and triggers re-collection or secondary processing.

[0014] Both sources of data contain spatiotemporal location information at the time of collection, providing a spatial reference for feature association mapping, information fusion and completion, and accurate disease location.

[0015] The linked data collection can dynamically adjust the collection strategy according to the type and road condition characteristics of the target road segment, ensuring the effectiveness and sufficiency of the correlation between the two sources of data.

[0016] The information fusion and completion process follows the principle of "global-local mutual verification", ensuring the accuracy and consistency of the completed information through cross-verification of dual-source information.

[0017] The inspection report includes basic road section information, data collection and processing procedures, a complete information summary, statistical data on road defects, precise location information, and maintenance recommendations, and supports export in a standardized format.

[0018] This invention proposes a technical solution of "multi-temporal light and shadow inversion + dual-view linkage acquisition + feature association mapping + information fusion and completion". By acquiring multi-temporal images of the same road segment under different solar altitude angles, the three-dimensional morphology of cracks is inverted by utilizing the geometric changes of crack shadows, and hidden micro-cracks invisible under normal lighting are identified. At the same time, the advantages of UAV global coverage and vehicle-mounted local details are combined to achieve full coverage and accurate quantification of road defects. Attached Figure Description

[0019] Figure 1 This is a system architecture diagram. Detailed Implementation

[0020] The road defect detection method based on multi-temporal light and shadow inversion and linkage between low-altitude UAV and vehicle-mounted camera of the present invention includes the following five steps, the specific process of which is as follows: S1: Linked Data Acquisition

[0021] Equipment preparation: Configure low-altitude drones and vehicle-mounted cameras with high-definition imaging capabilities, complete equipment function calibration, and ensure the clarity and stability of image acquisition; equip the drones and vehicle-mounted equipment with high-precision spatiotemporal positioning modules (including GPS / BeiDou positioning modules and high-precision clock synchronization modules) to achieve spatiotemporal synchronization calibration between devices and ensure that dual-source data have a unified spatiotemporal reference.

[0022] Data Acquisition Planning: Based on the length, width, road conditions, and geographical orientation of the target road segment, plan the flight path of the UAV and the acquisition route of the vehicle-mounted camera. Specifically, for road segments or key monitoring areas with high crack detection accuracy requirements, develop a multi-temporal acquisition scheme: Based on the latitude and longitude of the target road segment and its direction, combined with solar calendar data, select two or more time windows with different solar altitude angles (e.g., selecting three time points with solar altitude angles of approximately 30°, 50°, and 70°, corresponding to morning, noon, and afternoon). Plan for the UAV and vehicle-mounted equipment to repeatedly acquire data on the same road segment within the corresponding time windows. The difference in solar altitude angle between different time windows should not be less than a preset threshold (e.g., 15°) to ensure that the crack shadows have significant geometric differences, providing a sufficient data foundation for subsequent light and shadow inversion. The acquisition plan also needs to specify the UAV's flight altitude, flight speed, gimbal angle, and the vehicle-mounted camera's acquisition frame rate and resolution, ensuring the spatiotemporal synchronization of the dual-source data and that the image quality meets the requirements for subsequent processing.

[0023] Dual-view acquisition: The drone flies along a planned path, continuously acquiring global images of the target road segment, achieving complete coverage of the road surface and the edges of both sides of the road; the vehicle travels at a constant speed along the target road segment, and the onboard camera acquires local detailed images of the road surface, focusing on the road surface area. For multi-temporal acquisition tasks, the drone and the onboard equipment repeatedly perform acquisition tasks according to predetermined time windows. Within each time window, a complete full-coverage acquisition of the road segment is completed, recording the precise timestamp, solar altitude angle, solar azimuth angle, equipment attitude parameters, and other information for each acquisition. This information is then stored as metadata in association with the image data.

[0024] Anomaly Handling: Monitor the equipment status and image quality in real time during the acquisition process. If equipment failure, image occlusion (such as being blocked by other vehicles), poor environment causing blurred images, or overexposure or underexposure of light causing substandard image quality occur, mark the corresponding data as invalid, record the cause of the anomaly and its location information, and re-acquire images of the area after the fault is resolved or the weather conditions improve, to ensure the integrity and availability of the acquired data. S2: Image Preprocessing

[0025] Noise Removal: The global image of the UAV and the local image of the vehicle are processed separately. Based on the imaging characteristics of the equipment and the characteristics of the acquisition environment, an adaptive filtering algorithm is used to eliminate environmental interference, equipment imaging noise, sensor noise, etc., while retaining the core feature information of the image and improving the signal-to-noise ratio.

[0026] Distortion correction: Based on the imaging characteristics of the two types of devices—drone cameras typically suffer from wide-angle distortion, while vehicle-mounted cameras suffer from lens distortion and perspective distortion caused by vehicle motion—geometric distortion correction is performed on the dual-source images to correct the effects of viewing angle deviation and lens distortion. At the same time, motion blur correction is performed on the vehicle-mounted images in conjunction with vehicle motion parameters to ensure the geometric accuracy and clarity of the images.

[0027] Feature enhancement: Image optimization processes such as contrast stretching, histogram equalization, and edge enhancement are used to improve the feature differences between the road surface and the defects, and enhance the recognition of key features such as crack edges, texture details, and defect outlines. In particular, for multi-temporal image sequences, it is necessary to maintain the consistency of enhancement parameters for images in different time windows to avoid distortion of light and shadow features caused by differences in processing parameters, thus laying the foundation for subsequent feature extraction and matching.

[0028] Image registration preprocessing: Perform preliminary geometric alignment on multi-temporal image sequences to eliminate image position shifts caused by UAV flight trajectory deviations or vehicle driving path differences, ensuring that different temporal images of the same spatial location can correspond accurately. S3: Feature Association Mapping

[0029] Feature Extraction: Core features with stability and uniqueness are extracted from the preprocessed dual-source data. The UAV global data focuses on extracting global features such as road segment outlines, landmark features (e.g., road markings, manhole covers, curbs), and macroscopic damage outlines; the vehicle-mounted local data focuses on extracting microscopic features such as damage edges, road surface texture details, local feature markers, and aggregate distribution, constructing a dual-source feature set. For multi-temporal image sequences, additional light and shadow features are extracted, including: the length, width, and direction of crack shadows, the relative position of the shadow and the crack itself, the gradient parameters of the shadow as the illumination angle changes, and the contrast variation characteristics of cracks under different illumination conditions, forming a light and shadow feature vector.

[0030] Feature association: Based on the spatiotemporal location information and common feature regions of dual-source data, global and local features are compared and matched. Methods such as feature descriptor matching and mutual information maximization are used to filter out common feature points across data sources, establishing spatial correspondences between global and local features, forming a "global-local" feature association network. For multi-temporal data, it is also necessary to establish feature correspondences for the same spatial location in different time windows.

[0031] Mapping Construction: Based on the feature association network, a logical mapping channel between global and local information is constructed to clarify the information correspondence of the same target from different perspectives, achieving accurate association between global and local data. During the mapping construction process, spatial transformation models (such as perspective transformation and thin-plate spline transformation) are used to describe the coordinate mapping relationship between the global and local images, providing accurate spatial transformation parameters for subsequent information fusion and completion.

[0032] Light and shadow feature association: The light and shadow feature vectors in multi-temporal image sequences are associated with the crack features at corresponding spatial locations to establish a mapping database of "illumination conditions (solar altitude angle, solar azimuth angle) - shadow geometric parameters (length, width, direction) - crack morphology," providing input for light and shadow inversion. By analyzing the shadow variation patterns of the same crack under different illumination conditions, a functional relationship model between the three-dimensional morphology of the crack and the shadow geometry is constructed.

[0033] Association Validation: The validity of the constructed association mapping relationship is verified to confirm the accuracy and sufficiency of feature association. Cross-validation is used, selecting a subset of feature points as a validation set to evaluate the mapping error. If the association success rate does not reach the preset standard (e.g., 90%), or the mapping error exceeds the threshold, feature extraction is repeated, or supplementary data on common feature regions are collected until the association quality meets the requirements. S4: Information Fusion and Completion

[0034] Coverage Gap Filling: Based on feature association mapping, the full-segment coverage advantage of UAV global data is used to fill the spatial coverage blind spots caused by the limitations of vehicle-mounted local data due to driving trajectory (such as road edges, shoulders, and both sides of the central median, which are difficult for vehicle-mounted cameras to cover). The basic road surface conditions in the blind spot areas are clarified to ensure the integrity of the entire road information and to avoid missing easily overlooked defects in edge areas, both sides of the road, and other locations.

[0035] Detailed information completion: By leveraging the microscopic detail advantage of vehicle-mounted local data, the missing details in the global data of UAVs caused by resolution limitations are supplemented, providing microscopic feature support for macroscopic defects in the global data, clarifying the specific shape, texture, boundary and other detailed information of defects, and avoiding misjudgment of defects or misclassification of defects due to insufficient details.

[0036] Light and shadow inversion and completion: For multi-temporal image sequences, a light and shadow inversion mechanism is used to infer the three-dimensional morphological parameters of cracks based on the geometric changes of crack shadows under different illumination angles through an optical geometric model. Specifically, a geometric optical model is established with solar altitude angle, solar azimuth angle, crack direction, crack depth, and crack opening width as variables. The shadow length and shadow width under different illumination conditions are used as observed values ​​to solve for the optimal estimates of crack depth and opening width. During the inversion process, crack texture information from vehicle-mounted local detail images is incorporated as a constraint to improve inversion accuracy. The inverted three-dimensional parameters such as crack depth, opening width, and cross-sectional morphology are used as supplementary local detail information. In particular, for latent microcracks with extremely low contrast and invisible to the naked eye in conventional images, reliable identification and quantitative assessment are achieved through light and shadow differences, incorporating latent microcracks into the disease information set.

[0037] Information cross-verification optimization: Following the principle of "global-local cross-verification," the bias of a single data source is corrected through cross-verification of dual-source information—global data is used to verify the spatial location accuracy of local data, and local data is used to verify the authenticity of the disease characteristics of global data. In particular, the latent microcracks identified by light and shadow inversion are verified by feature association with the local detail images on the vehicle to confirm the real existence of the microcracks and accurately pinpoint their spatial location, eliminating false detections caused by light and shadow interference or image noise; at the same time, the crack morphology identified in the local images on the vehicle is compared with the light and shadow inversion results to verify the rationality of the inversion parameters. If a deviation is found, the inversion model parameters are adjusted accordingly.

[0038] Completeness Verification: The quality of the fused and completed information set is checked to confirm that the spatial coverage is complete, the detailed features are clear, the information logic is consistent, and the lighting and shadow inversion parameters are reasonable. If there are information gaps, logical contradictions, or abnormal inversion results, the fusion and completion strategy is adjusted, or feature association mapping and lighting and shadow inversion are re-performed until the information quality meets the requirements. S5: Disease Identification and Result Output

[0039] Intelligent Model Loading: Loads a trained and optimized intelligent recognition model. This model is capable of recognizing various road defects (cracks, potholes, ruts, loose surfaces, etc.) and can effectively distinguish defects from non-defect interference targets such as road stains, manhole covers, road markings, and shadows. The model has been trained on a large number of labeled samples and has good generalization ability and robustness. The model has been specifically trained for the 3D parameters of cracks output by light and shadow inversion, and can integrate multi-dimensional features for comprehensive judgment.

[0040] Intelligent Disease Identification: The fused complete information set and the results of light and shadow inversion are input into the intelligent identification model. The model automatically scans the image area, identifies the location, boundaries, and type of the disease, and extracts the core information of the disease. For crack-type diseases, the model not only outputs the two-dimensional morphology of the crack (length, direction, distribution) but also integrates the three-dimensional parameters (depth, opening width, cross-sectional shape) output by light and shadow inversion to comprehensively assess the severity of the crack. For latent microcracks, the model marks them based on the light and shadow inversion results and includes them in the disease statistics. During the identification process, the model comprehensively utilizes global and local information, and improves the identification accuracy through multi-scale feature fusion.

[0041] Results integration and statistics: The identified disease information was classified and statistically analyzed, and the quantity, area, and severity distribution were counted according to disease type (cracks, potholes, ruts, loosening, etc.). Combined with the spatiotemporal location information at the time of collection, a disease distribution map was generated, and the location, type, and level of the disease were marked on the geographic information system base map. The severity level of the disease (mild, moderate, and severe) was divided according to preset standards (such as crack width and depth thresholds) to provide a basis for prioritizing maintenance.

[0042] Report Generation and Output: The system automatically generates standardized inspection reports, including: basic road segment information (segment name, length, width, pavement type, data collection time, etc.); data collection and processing procedures (data collection equipment parameters, multi-temporal data collection time windows, solar altitude angle information, preprocessing methods, feature association parameters, and light and shadow inversion parameters, etc.); fused images (including global and local detail images with disease annotations); a summary table of crack 3D morphology parameters (length, depth, opening width, cross-sectional shape, orientation, and location coordinates of each crack); a distribution map of latent microcracks (marking the location and morphology of microcracks not visible in regular images but identified through light and shadow inversion); disease statistics (quantity, area, and severity distribution of various diseases); disease distribution maps; and targeted maintenance recommendations (providing maintenance timing and methods based on different disease types and severity). The report supports exporting to standardized formats (such as PDF, Word, Excel, GIS layers, etc.) for use by maintenance management departments. Example 1

[0043] A road defect detection method based on multi-temporal light and shadow inversion and linkage between low-altitude UAVs and vehicle-mounted cameras was applied and verified in a major urban road. The execution process is as follows: Linked Data Acquisition: A 3.2-kilometer-long, four-lane, asphalt concrete road was selected as a main urban road. A DJI Matrice M300 RTK drone (equipped with a Zenmuse P1 full-frame camera) and a vehicle-mounted high-definition camera (20MP, 60fps) were configured, and equipment calibration and spatiotemporal synchronization were completed. Based on the road's latitude and longitude (39.9°N) and direction (nearly north-south), combined with solar calendar data, three time windows were selected for multi-phase data acquisition: 9:30 AM (solar altitude approximately 35°), 12:30 PM (solar altitude approximately 65°), and 3:30 PM (solar altitude approximately 40°). The drone flew along a pre-set path along one side of the road's centerline at a height of 50 meters, with the gimbal pointing vertically downwards to achieve global coverage of the entire road section; the vehicle traveled at a constant speed (30 km / h) along the road, with the vehicle-mounted camera focusing on capturing road surface details. A total of 2,400 sets of global data (each set containing images of the entire road segment within that time window) and 3,600 sets of local data were collected across three time windows. All data included precise timestamps, GPS coordinates, solar altitude angles, and device attitude parameters.

[0044] Image preprocessing: Adaptive median filtering is applied to the dual-source data to reduce noise and eliminate environmental and equipment interference; geometric distortion correction is performed based on the calibration parameters of the UAV and vehicle-mounted cameras to ensure data geometric accuracy; adaptive histogram equalization is used for feature enhancement to improve the identification of disease features, and the processing parameters of the three time windows are kept consistent to avoid introducing human differences.

[0045] Feature association mapping: Global features (road segment outline, markings, manhole cover location) and local features (road surface texture, crack edges, aggregate distribution) are extracted from the preprocessed data. Based on GPS coordinates and feature descriptor matching, a "global-local" feature association network is established, achieving an association success rate of 94.5%. Light and shadow features are extracted from multi-temporal image sequences: For each crack, the shadow length, shadow width, and shadow direction are measured under three time windows to construct a light and shadow feature vector. A mapping relationship database of "lighting conditions—shadow geometry—crack morphology" is established.

[0046] Information fusion and completion: Using global data from UAVs to fill in five edge area blind spots (roadside shoulder areas on both sides of the road) in the local vehicle data, clarifying the road surface conditions in the blind spots. Using local vehicle data to complete the detailed features of 18 macroscopic defects in the global UAV data.

[0047] Light and shadow inversion and completion: A microcrack with a width of approximately 0.3 mm (invisible to the naked eye in conventional images) was selected. This crack exhibited different shadow characteristics in three time windows: a shadow length of 8.2 mm in the morning, 1.5 mm at noon, and 7.8 mm in the afternoon. Based on the optical geometric model inversion, the crack depth was calculated to be approximately 4.2 mm, the opening width approximately 0.32 mm, and the cross-sectional shape V-shaped. The inversion results were in high agreement with the measured values ​​(depth 4.5 mm, width 0.30 mm) obtained from subsequent verification core sampling. Through light and shadow inversion, a total of 23 hidden microcracks invisible in conventional images were identified, and the three-dimensional morphological parameters of each crack were obtained. The light and shadow inversion results were verified by feature correlation with the local detail images on the vehicle, confirming that all 23 microcracks truly existed and ruling out false detections.

[0048] Disease Identification and Output: A trained YOLOv8+Transformer fusion identification model was loaded to identify various diseases from the complete information set: 31 visible cracks (including 18 previously identified macroscopic diseases), 23 latent microcracks (identified through light and shadow inversion), and 5 potholes. The model comprehensively assessed the severity of cracks based on crack width and depth information. Cracks with a width ≥2mm and a depth ≥5mm were classified as moderate cracks (7 locations), and the rest as minor cracks. A disease distribution map was generated by combining spatiotemporal information, and the location, type, severity level, and 3D parameters of the cracks were accurately marked on the GIS map. A standardized inspection report was generated, specifying that moderate cracks should be treated with grouting within one month, latent microcracks should be included in the key monitoring list, and potholes should be repaired within two weeks. The report was exported as a PDF and a GIS layer for submission to the maintenance management department for decision-making.

[0049] Completeness verification: The quality of the fused and completed information set is checked to confirm that the spatial coverage is complete without blind spots, the detailed features are clear, the light and shadow inversion results are consistent with the texture features of the vehicle image, and the information quality meets the detection requirements.

Claims

1. A method for detecting road defects using a low-altitude UAV and vehicle-mounted camera in conjunction with multi-temporal light and shadow inversion, characterized in that, The system includes coordinated data acquisition, image preprocessing, feature association mapping, information fusion and completion, and disease identification and result output. The coordinated data acquisition uses low-altitude UAVs and vehicle-mounted cameras to collect global and local detail information of the target road segment, forming dual-source complementary data. The coordinated data acquisition also includes multi-temporal and multi-angle planned acquisition of the same road segment to obtain image sequences under different lighting conditions. The image preprocessing optimizes the dual-source data to improve information recognition. The feature association mapping extracts the core features of the dual-source data and establishes association relationships, constructs a global and local information mapping channel, and extracts shadow features under different lighting conditions; the information fusion and completion is based on the association mapping relationship, fills the information gaps of a single data source through information complementarity logic, forms a complete road information set, and introduces a light and shadow inversion mechanism to invert the three-dimensional morphology and width information of cracks using shadow changes in multi-temporal images, and identifies hidden micro-cracks that are not visible under normal lighting; the disease identification and result output uses intelligent algorithms to identify disease information from the complete information set and generate a standardized inspection report.

2. The method according to claim 1, characterized in that, The coordinated acquisition also includes multi-temporal acquisition planning: based on the orientation and geographical location of the target road segment, the drone and vehicle-mounted equipment are planned to repeatedly acquire images of the same road segment within multiple time windows with different solar altitude angles, thereby obtaining image sequences containing differences in light and shadow.

3. The method according to claim 2, characterized in that, In the multi-temporal acquisition plan, the difference in solar altitude angle between different time windows is not less than a preset threshold to ensure that the crack shadow has significant geometric differences.

4. The method according to claim 1, characterized in that, The feature association mapping also includes light and shadow feature extraction: extracting feature parameters of the length, direction, and width of the crack shadow as a function of the illumination angle from a multi-temporal image sequence, and constructing a light and shadow feature vector.

5. The method according to claim 1, characterized in that, The light and shadow inversion mechanism includes: based on the geometric changes of crack shadows under different illumination angles, using an optical geometric model to infer the crack depth, opening width and cross-sectional shape, and using the inversion results as a supplement to local detail information to fill the lack of crack detail information in conventional images.

6. The method according to claim 1, characterized in that, In the information fusion and completion process, the hidden microcracks identified through the light and shadow inversion mechanism are verified by feature association with the local detail images captured by the vehicle camera, confirming the real existence of the microcracks and accurately calibrating their spatial location.

7. The method according to claim 1, characterized in that, In the disease identification and result output, the output detection report includes the three-dimensional morphological parameters of the cracks, such as depth, opening width, cross-sectional shape, and distribution map of hidden microcracks.

8. The method according to claim 1, characterized in that, It also has a data quality verification function, which marks the collected data that does not meet the quality requirements, the data that has not established an effective association mapping, or the data that still has information defects after fusion and completion, and triggers re-collection or secondary processing.

9. The method according to claim 1, characterized in that, Both sources of data contain spatiotemporal location information at the time of collection, providing a spatial reference for feature association mapping, information fusion and completion, and accurate disease location.

10. The method according to claim 1, characterized in that, The linked data collection can dynamically adjust the collection strategy according to the type and road condition characteristics of the target road segment to ensure the effectiveness and sufficiency of the association between the two sources of data; the information fusion and completion process follows the principle of "global-local mutual verification" and ensures the accuracy and consistency of the completed information through cross-verification of the two sources of information.

Images