Method for constructing a road facility database, and computer program recorded on record-medium for executing method therefor

Abstract

The present invention proposes a method for constructing a road facility database that generates three-dimensional point cloud data containing depth information based on images acquired while driving on a road, automatically identifies road facilities through this, and automatically constructs a database containing information about the identified road facilities. The method may include the steps of: a management server collecting a plurality of images continuously captured during the process of driving on a road from an information collection device; the management server extracting feature points within the collected plurality of images; the management server matching the extracted feature points of the collected plurality of images to acquire three-dimensional point cloud data containing depth information; the management server identifying an object corresponding to a road facility based on the acquired three-dimensional point cloud data; and the management server constructing a database for road facilities based on the identified object.

Description

Technology Field

[0001] The present invention relates to a technology for constructing a road facility database. More specifically, it relates to a method for constructing a road facility database that can generate three-dimensional point cloud data including depth information based on images continuously captured while driving on a road, and construct a database regarding road facilities through the generated three-dimensional point cloud data, and a computer program recorded on a recording medium to execute the same. Background Technology

[0002] Road facilities refer to various structures and equipment installed on or around roads to ensure road safety and efficiency. For example, road facilities can include various types of equipment such as bollards, guardrails, traffic signs, hydrants, and drains. These road facilities play a crucial role in securing traffic safety, facilitating the smooth movement of pedestrians and vehicles, and enabling rapid response in emergency situations.

[0003] In particular, road facilities can be damaged or their functionality deteriorates due to various factors such as long-term use, weather conditions, external impacts, and pollution. For example, if bollards tilt due to vehicle impacts or drains become clogged with fallen leaves or debris, serious problems may arise regarding traffic safety and the urban environment.

[0004] Accordingly, regularly inspecting and maintaining road facilities is emerging as an essential task for maintaining the safety and efficiency of road traffic. In particular, local governments and public institutions are currently facing the challenge of managing a vast number of facilities, and there is a strong demand for automated management technologies capable of performing this task efficiently.

[0005] Most existing road facility databases are constructed through manual field surveys. This method not only requires significant time and cost but also poses a major problem: there is a high risk that visual characteristic information, such as the location and shape of facilities, may be omitted or recorded inaccurately.

[0006] Accordingly, recently, technologies for building databases of facilities using lidar or expensive 3D measurement equipment have been proposed, but there have been limitations in building precise databases that reflect the visual characteristics of facilities because the equipment installation costs are excessively high and they cannot include color or texture information. Prior art literature

[0007] Korean Registered Patent Publication No. 10-0912463, 'Integrated Road Management System Based on Electronic Map and Method for Controlling the Same', (Registered Aug. 10, 2009) The problem to be solved

[0008] One objective of the present invention is to provide a method for constructing a road facility database that can generate three-dimensional point cloud data including depth information based on images acquired while driving on a road, automatically identify road facilities through this, and automatically construct a database including information about the identified road facilities, and a computer program recorded on a recording medium to execute the same.

[0009] Another objective of the present invention is to provide a method for constructing a road facility database that can manage facility information accurately and efficiently by constructing a 3D semantic map classified by road facility object for 3D point cloud data generated based on images, and a computer program recorded on a recording medium to execute the same.

[0010] The technical problems of the present invention are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0011] To achieve the technical objectives described above, the present invention proposes a method for constructing a road facility database that generates three-dimensional point cloud data containing depth information based on images acquired while driving on a road, automatically identifies road facilities through this, and automatically constructs a database containing information about the identified road facilities. The method may include the steps of: a management server collecting a plurality of images continuously captured during the process of driving on a road from an information collection device; the management server extracting feature points within the collected plurality of images; the management server matching the extracted feature points of the collected plurality of images to acquire three-dimensional point cloud data containing depth information; the management server identifying an object corresponding to a road facility based on the acquired three-dimensional point cloud data; and the management server constructing a database for road facilities based on the identified object.

[0012] The above-mentioned collecting step is characterized by collecting a plurality of images captured by driving a shutter in proportion to the number of rotations of the vehicle wheels during the driving process of the vehicle in which the information collecting device is installed.

[0013] The above-mentioned acquisition step is characterized by estimating depth information of each feature point through triangulation based on the correspondence relationship between the extracted feature points, and calculating the 3D coordinates of the 3D point cloud data through the estimated depth information.

[0014] The above identification step is characterized by applying a pre-trained object detection model to each of the plurality of images to specify the location of the object of interest as a bounding box, and extracting 3D point cloud data associated with feature points included in the specified bounding box to identify an object corresponding to a road facility.

[0015] The above identification step is characterized by clustering the acquired 3D point cloud data by class, calculating the clustering confidence of each point, excluding points where the calculated clustering confidence is less than a preset value, and classifying each point into buckets according to an image identifier and a bounding box identifier.

[0016] The above-described identifying step is characterized by calculating the centroid of the classified bucket, merging the buckets to assign a single label when the centroid distance between different buckets is less than a preset value and the image sequence interval is less than a preset value, and reconstructing the point cloud based on the label.

[0017] The above-described construction step is characterized by calculating a reference plane corresponding to a road surface for the acquired three-dimensional point cloud data, and correcting the three-dimensional point cloud data by repeatedly applying a rotation transformation to convert the calculated reference plane into a horizontal plane.

[0018] The above-described construction step is characterized by calculating transformation parameters by correlating the corrected camera pose and position information recorded during the shooting process in segment units, and aligning the 3D point cloud data obtained based on the calculated transformation parameters with a real-world coordinate system.

[0019] The above-described construction step is characterized by labeling each point of the above-described 3D point cloud data with object class information to construct a 3D semantic map classified by road facility object.

[0020] The above 3D semantic map is characterized by being stored in cluster units that combine a Universally Unique Identifier (UUID) for an object corresponding to the identified road facility, and metadata including type information, shape information, and location information for the object.

[0021] The above-described construction step is characterized by extracting an image patch for an object that appears commonly in at least two of the plurality of images, and extracting a visual feature (deep feature) from the extracted image patch that is invariant to changes in lighting, shooting angle, and partial occlusion.

[0022] The above-described construction step is characterized by converting the extracted features into embedding vectors based on a pre-trained embedding model, compressing the converted embedding vectors into a single representative vector representing the facility at a given time point through an aggregator, and storing the compressed single representative vector.

[0023] The above-described construction step is characterized by configuring structured attributes including type, spatial information, shape information, and collection time for each road facility, unstructured attributes including image patches and person-in-chief notes, and the single representative vector into a graph-based semantic database framework.

[0024] The above computer program may be combined with a computing device comprising a transceiver, a memory, and a processor that processes instructions residing in the memory. Furthermore, the above computer program may be a computer program recorded on a recording medium to execute the steps of: the processor collecting a plurality of images continuously captured while driving on a road from an information collection device; the processor extracting feature points within the collected plurality of images; the processor matching the extracted feature points of the collected plurality of images to obtain three-dimensional point cloud data including depth information; the processor identifying an object corresponding to a road facility based on the obtained three-dimensional point cloud data; and the processor building a database for the road facility based on the identified object.

[0025] Specific details of other embodiments are included in the detailed description and drawings. Effects of the invention

[0026] According to various embodiments of the present invention, three-dimensional point cloud data is generated based on images acquired while driving on a road, and road facilities are automatically identified through this, thereby enabling the accurate and efficient construction of a database including the location, shape, and visual features of the facilities.

[0027] In addition, according to various embodiments of the present invention, by generating a 3D semantic map labeled for each road facility object and reflecting it in a database, the management and updating of facilities can be automated, and the reliability and usability of data can be significantly improved compared to existing manual-based surveys.

[0028] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art to which the present invention pertains from the description in the claims. Brief explanation of the drawing

[0029] FIG. 1 is a configuration diagram of a road facility management system according to one embodiment of the present invention. FIG. 2 is a logical configuration diagram of a management server according to one embodiment of the present invention. FIGS. 3 to 5 are illustrative diagrams for explaining a database construction unit according to an embodiment of the present invention. FIG. 6 is an illustrative diagram for explaining a change detection unit according to one embodiment of the present invention. FIG. 7 is a hardware configuration diagram of a management server according to one embodiment of the present invention. FIG. 8 is a flowchart illustrating a road facility management method according to an embodiment of the present invention. Specific details for implementing the invention

[0030] It should be noted that technical terms used in this specification are used merely to describe specific embodiments and are not intended to limit the invention. Furthermore, unless specifically defined otherwise in this specification, technical terms used in this specification should be interpreted in the sense generally understood by those skilled in the art to which the invention pertains, and should not be interpreted in an overly broad or overly narrow sense. Additionally, if a technical term used in this specification is an incorrect technical term that fails to accurately express the spirit of the invention, it should be understood as being replaced by a technical term that can be correctly understood by those skilled in the art. Moreover, general terms used in this invention should be interpreted according to their prior definitions or the context, and should not be interpreted in an overly narrow sense.

[0031] Additionally, singular expressions used in this specification include plural expressions unless the context clearly indicates otherwise. In this application, terms such as "composed of" or "have" should not be interpreted as necessarily including all of the various components or steps described in the specification, and should be interpreted as potentially including some of the components or steps, or including additional components or steps.

[0032] Additionally, terms including ordinal numbers, such as first, second, etc., used herein may be used to describe various components, but said components shall not be limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.

[0033] When it is stated that one component is "connected" or "connected" to another component, it may be directly connected or connected to that other component, or there may be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0034] Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. Identical or similar components regardless of drawing symbols are given the same reference number, and redundant descriptions thereof will be omitted. Furthermore, in describing the present invention, if it is determined that a detailed description of related known technology may obscure the essence of the present invention, such detailed description will be omitted. Additionally, it should be noted that the attached drawings are intended only to facilitate an easy understanding of the concept of the present invention and should not be interpreted as limiting the concept of the present invention. The concept of the present invention should be interpreted as extending to all modifications, equivalents, and substitutions other than those shown in the attached drawings.

[0036] FIG. 1 is a configuration diagram of a road facility management system according to one embodiment of the present invention.

[0037] Referring to FIG. 1, the information collection device (100) is mounted on a vehicle and can perform the function of continuously capturing images of the surrounding environment, including road facilities, while driving on the road.

[0038] Specifically, the information collection device (100) may be configured to include a camera, and images can be acquired at uniform distance units by driving the shutter in conjunction with the number of rotations of the wheel. Unlike a method of simply extracting images at time intervals, such a shooting method can secure image data at constant spatial intervals regardless of the vehicle's driving speed, and thus can contribute to improving the accuracy of subsequent 3D point cloud data reconstruction.

[0039] Additionally, the information collection device (100) can collect metadata such as location coordinates (GPS), the vehicle's driving speed, and the time of shooting along with the captured video, and the collected data can be transmitted to a management server (200) via a network. In some embodiments, an IMU (Inertial Measurement Unit) sensor may be installed to provide additional information regarding changes in the vehicle's attitude or movement trajectory.

[0040] The information collection device (100) is depicted as being installed on a vehicle, but is not limited thereto and may be installed on various means of transportation such as a drone, a train, a ship, or a bike. Such an information collection device (100) can provide basic data for detecting changes in road facilities through images acquired at each driving time. In addition, the information collection device (100) can provide source data for constructing three-dimensional point cloud data and a semantic database.

[0041] Next, the management server (200) may have the function of performing an analysis of road facilities based on a plurality of images received from the information collection device (100).

[0042] In one embodiment, a management server (200) can collect a plurality of images continuously captured while driving on a road from an information collection device (100). Subsequently, the management server (200) extracts feature points within the collected plurality of images, matches the extracted feature points of the collected plurality of images to obtain three-dimensional point cloud data including depth information, and can identify an object corresponding to a road facility based on the obtained three-dimensional point cloud data. Then, the management server (200) can build a database for the road facility based on the identified object.

[0043] In another embodiment, the management server (200) can collect a plurality of images continuously captured during the process of driving on the road from the information collection device (100). Subsequently, the management server (200) can extract feature points within the collected plurality of images, match the extracted feature points of the collected plurality of images to obtain 3D point cloud data including depth information, and identify an object corresponding to a road facility based on the obtained 3D point cloud data. Then, the management server (200) can detect changes in the road facility by comparing the 3D point cloud data corresponding to the road facility based on a pre-stored facility database.

[0044] Such a management server (200) can be implemented as a local server, a cloud server, an edge server, etc., and the embodiment of the present invention is not limited to a specific server type.

[0045] The information collection device (100) and management server (200) described above can transmit and receive data using a network that combines one or more of a secure line directly connecting the devices, a public wired communication network, or a mobile communication network.

[0046] For example, public wired communication networks may include Ethernet, Digital Subscriber Line (xDSL), Hybrid Fiber Coax (HFC), and Fiber To The Home (FTTH), but are not limited thereto. Additionally, mobile communication networks may include Code Division Multiple Access (CDMA), Wideband CDMA (WCDMA), High Speed ​​Packet Access (HSPA), Long Term Evolution (LTE), and 5th generation mobile telecommunication, but are not limited thereto.

[0048] FIG. 2 is a logical configuration diagram of a management server according to one embodiment of the present invention.

[0049] Referring to FIG. 2, a management server (200) according to one embodiment of the present invention may be configured to include a communication unit (205), an input / output unit (210), a database construction unit (215), a change detection unit (220), and a storage unit (225).

[0050] To explain each configuration, the communication unit (205) can perform the role of transmitting and receiving data between the information collection device (100) and the management server (200).

[0051] Specifically, the communication unit (205) receives metadata in real time, such as multiple images continuously captured from an information collection device (100) mounted on various mobile bodies such as vehicles, drones, and autonomous robots, location coordinates at each shooting time, shooting time, and vehicle speed, and transmits this to the database construction unit (215) and the change detection unit (220) so that subsequent processing can be performed.

[0052] In addition, the communication unit (205) can perform the function of reliably transmitting raw images and metadata so that the database construction unit (215) can perform point cloud generation, object identification, semantic map construction, etc.

[0053] Additionally, the communication unit (205) can perform the function of transmitting point clouds and embedding vectors collected at a specific point in time so that the change detection unit (220) can compare existing information stored in the database with new data. Additionally, the communication unit (205) can transmit analysis results, such as detected change results or maintenance priorities, to an external management system or user terminal so that they can be utilized for management decision-making of road facilities.

[0054] Furthermore, the communication unit (205) supports bidirectional communication between the management server (200) and the external management infrastructure, so that it receives data requests, update commands, maintenance instruction signals, etc. from the external system and transmits them to the internal module, thereby enabling the entire system to operate organically.

[0055] With the following configuration, the input / output unit (210) can receive control commands executed by an operator or manager of the management server (200) or perform the function of visually outputting processing results.

[0056] Specifically, the input / output unit (210) can be used to control the operation of the database construction unit (215) and the change detection unit (220) by receiving control commands, setting information, policy rules, etc., input by a system administrator or operator. For example, the input / output unit (210) can adjust the processing criteria for image and point cloud data, the threshold setting of the object detection model for facility identification, the adjustment of the reference value for similarity comparison when detecting changes, the database storage cycle, and management policies according to the administrator's input.

[0057] Additionally, the input / output unit (210) can visually output the results performed by the management server (200) so that an administrator can verify them. For example, the input / output unit (210) may output facility identification results for a specific road section, location and shape information of each facility, damage type and grade based on change detection results, maintenance priority, etc., through a map-based interface or dashboard screen.

[0058] Furthermore, the input / output unit (210) may be linked with an external organization or system to perform the function of converting and outputting processing results according to the requested data format. For example, the input / output unit (210) may convert a part of the semantic database into a standard GIS (Geographic Information System) format or automatically convert change detection logs into a report format and provide them to an external system.

[0059] The database construction unit (215) can perform the function of generating three-dimensional point cloud data based on a plurality of images received from the information collection device (100), identifying road facilities through this, and automatically constructing a database containing information about the identified road facilities.

[0060] Specifically, the database building unit (215) can collect multiple images continuously captured during the process of driving on the road from the information collection device. At this time, the shooting is not simply performed at regular time intervals, but can be controlled so that the shutter is driven in proportion to the number of rotations of the vehicle wheels. For example, if the shutter is set to be driven 4 times when the vehicle wheels rotate one full turn, one image can be captured every time the vehicle moves about 1 meter. This method allows images to be acquired at the same spatial interval regardless of the vehicle's driving speed, thereby ensuring the homogeneity of the image data processed by the database building unit (215). For example, guardrails installed at 10-meter intervals on the same road section will appear repeatedly in images captured at 1-meter intervals, and the database building unit (215) can secure multiple observations of the same object by utilizing these duplicated images. Through this, the database building unit (215) can increase the precision of 3D coordinates and strengthen the reliability of object identification during the subsequent point cloud data generation process.

[0061] In addition, the database building unit (215) can collect additional metadata such as GPS (Global Positioning System) coordinates, shooting time, and vehicle speed along with continuously captured images. By utilizing this, the database building unit (215) can streamline the subsequent process of generating point cloud data and the georeferencing process of projecting point cloud data onto actual spatial coordinates, and maintain a high level of consistency with reality and reliability.

[0062] In addition, the wheel rotation-based shooting method adopted by the database construction unit (215) can solve the problems of time-interval shooting. That is, in time-interval shooting, the same object may be excessively captured when the vehicle is congested, and gaps may occur where objects are not captured when driving at high speeds. On the other hand, since wheel rotation-based shooting can secure images at the same interval regardless of speed, a stable and uniform dataset can be constructed across the entire road section.

[0063] As a result, the database construction unit (215) can consistently record the location and shape of road facilities by collecting continuous images taken in proportion to the number of wheel rotations, increase the reliability of point cloud data reconstruction by utilizing duplicate shooting of the same object, and enable the construction of a database without omission of facilities even in a high-speed driving environment.

[0064] The database construction unit (215) can extract feature points from multiple images collected from an information collection device. A feature point is a point that represents the uniqueness of the location of a subject or space within an image, and serves as a consistent location reference point across multiple images.

[0065] For example, the database construction unit (215) can correct the image using a pinhole camera model when the image is taken with a fisheye lens or a wide-angle lens to remove image distortion and increase the accuracy of the subsequent feature point extraction process.

[0066] The database building unit (215) can apply various algorithms to extract feature points. For example, the database building unit (215) can define feature points by describing the degree and direction of brightness change around pixels as vectors using the SIFT (Scale-Invariant Feature Transform) technique. This process enables stable extraction of the same object even when transformations such as enlargement, reduction, and rotation of the image occur. In addition, the database building unit (215) learns autocorrelation characteristics that consider the spatial distribution of feature points within an image using a deep learning-based matching model such as "SuperGlue," and based on this, can secure high matching performance even in images with day-night changes or large differences in viewpoint.

[0067] The database building unit (215) can, for example, extract corners of the same building from multiple images taken during the day and night, or from different angles. The database building unit (215) finds pixel patterns corresponding to the corners of the building in each image and defines them as common feature points, and in a subsequent step, can use these feature points to recognize that they are the same object.

[0068] The database construction unit (215) can utilize the extracted feature points in a subsequent matching process to establish correspondence relationships between objects and extend this process to calculating three-dimensional coordinates based on these correspondence relationships. Through this feature point extraction process, the database construction unit (215) can increase the precision of the point cloud data and enable stable identification of road facility objects.

[0069] The database construction unit (215) can obtain three-dimensional point cloud data by matching the correspondence between feature points extracted from multiple collected images and estimating the depth information of each feature point through triangulation.

[0070] The database building unit (215) can detect feature points corresponding to the same object in two or more images taken at different points in time and calculate the 3D coordinates of each feature point using the projection geometry of the camera. For example, if the corner of the same traffic sign appears at different locations in two images, the database building unit (215) can calculate the disparity between the two images and calculate the actual 3D coordinates (x, y, z) of the corner of the traffic sign by applying the camera's intrinsic parameters (focal length, principal point coordinates, etc.) and extrinsic parameters (shooting position, shooting attitude).

[0071] The database construction unit (215) can calculate the geometric relationship between two images by utilizing an essential matrix or a fundamental matrix in the trigonometry-based depth estimation process, and can remove incorrectly matched corresponding points and increase the reliability of the match by applying the RANSAC (Random Sample Consensus) algorithm.

[0072] The database construction unit (215) can restore three-dimensional coordinates by matching feature points extracted from each image and integrating multiple parallax calculation results, for example, when the same guardrail appears in five consecutive images. By utilizing multiple observations in this way, errors caused by noise are reduced, and consequently, the precision of the point cloud data can be improved. In addition, the database construction unit (215) tracks the process of deriving each point so that suitable points that minimize errors are included in the point cloud data, so that the point cloud data obtained from multiple images (100 or more) taken along the road is not fragmented or distorted and the consistency of the shape is ensured.

[0073] The database construction unit (215) can construct point cloud data that reflects the location and shape of road facilities in a real spatial coordinate system through this process, and this can be used as basic data for object identification and database recording thereafter. Therefore, the database construction unit (215) can reconstruct road facilities in three dimensions through feature point matching and depth information estimation, and increase the accuracy and reliability of the database.

[0074] The database construction unit (215) can specify the location of an object of interest as a bounding box by applying a pre-trained object detection model to each of the multiple images. The database construction unit (215) can identify an object corresponding to a road facility by extracting 3D point cloud data associated with feature points included in the area specified by the bounding box. For example, when a traffic sign is detected, the database construction unit (215) can form a traffic sign object in the actual spatial coordinate system by extracting a point cloud connected to feature points existing in the area of ​​the sign.

[0075] The database building unit (215) can cluster the acquired three-dimensional point cloud data by class and calculate the clustering reliability of each point to exclude points whose reliability is less than a preset value. Additionally, the database building unit (215) can classify each point into a bucket according to an image identifier and a bounding box identifier. For example, if the same bollard is repeatedly detected in a series of consecutive images, the database building unit (215) can manage the point clouds extracted from each image by assigning them to the same bucket.

[0076] The database building unit (215) can calculate the centroid of the classified buckets, and if the centroid distance between different buckets is less than a preset value and the image sequence interval is also less than a preset value, the buckets can be merged to be confirmed as a single object. For example, even if the same guardrail post is detected at different locations in five images taken while driving a vehicle, the database building unit (215) can merge the point clouds extracted from each image to identify it as a single guardrail object.

[0077] The database construction unit (215) can integrate the same object into a single object even if the same object is recognized redundantly in multiple images through this object identification process, and can secure unique 3D point cloud data and labels for each object. Therefore, the database construction unit (215) can prevent redundant recognition of road facilities through the object identification step and ensure consistency and accuracy of facility information stored in the database.

[0078] The database construction unit (215) can first estimate the road surface for the acquired 3D point cloud data and perform a correction process to level the entire point cloud by defining the road surface as a reference plane. For example, a point cloud acquired on a sloping hill may not reflect the slope of the actual terrain or may be bent due to drift phenomena. At this time, the database construction unit (215) can estimate a horizontal plane calculated using GPS data or statistical techniques as the road surface, and apply a rotation transformation to level the point cloud based on the estimated road surface, thereby ensuring the consistency of the object coordinates recorded in the database.

[0079] The database construction unit (215) calculates conversion parameters by matching the corrected camera pose and the location information recorded during the shooting process in segments, and can use this to align point cloud data with a real-world coordinate system. For example, by utilizing metadata including GPS coordinates to convert point cloud coordinates into geographic coordinates based on latitude and longitude, facility information stored in the database can be made to match the actual map.

[0080] The database construction unit (215) can construct a 3D semantic map that is distinguished by road facility objects by labeling each point of the matched point cloud data with class information of the object. For example, points belonging to guardrails can be labeled with the same class value and clearly distinguished from other facilities.

[0081] The database construction unit (215) can assign a Universally Unique Identifier (UUID) to clusters of objects and store metadata such as type information, shape information (e.g., center point, main axis vector, height, radius), and location information of each object. For example, metadata such as UUID 0001, location coordinates (37.123, 127.456), height 1.0m, radius 0.2m, and class "Fire Hydrant" can be combined with an object named "Fire Hydrant A".

[0082] The database construction unit (215) can extract image patches for objects that appear commonly in multiple images and extract visual features that are invariant to lighting, shooting angle, and partial occlusion. In this process, the extracted features are converted into embedding vectors through a pre-trained embedding model and can be compressed into a single representative vector through an aggregator and stored in the database. For example, even if the same traffic sign is photographed during the day and at night, the vectors extracted from the two images can be integrated into a single representative vector and recorded.

[0083] The database construction unit (215) can generate structured attributes such as type, spatial information, shape information, and collection time, unstructured attributes such as image patches and person-in-chief notes, and deep feature-based vector data for each road facility and store them in the database (225).

[0084] The database construction unit (215) can configure a graph-based semantic database framework to effectively load various attribute information and support retrieval services.

[0085] Since road facilities have interrelationships and dependencies, the database needs to reflect these characteristics. For example, a single bollard may be part of the pedestrian safety structure of a specific road section, a fire hydrant may be located next to it, and it may have a history of frequent repairs.

[0086] The database construction unit (215) can search and analyze various attributes by associating them with each other through a semantic database framework. Accordingly, the database construction unit (215) stores road facility information precisely and consistently, and can utilize it for future change detection and maintenance management.

[0087] For a moment, with reference to FIGS. 3 to 5, a database construction unit (215) according to one embodiment of the present invention will be described in more detail.

[0088] FIGS. 3 to 5 are illustrative diagrams for explaining a database construction unit according to an embodiment of the present invention.

[0089] Referring to FIGS. 3 to 5, the database construction unit (215) can apply an algorithm to solve various problems that occur in the prior art during the process of constructing a database regarding road facilities based on identified objects.

[0090] The database construction unit (215) can solve the problem of incomplete labeling that occurs during the process of converting 2D object detection results into 3D point clouds. In the prior art, detection is performed even when bounding boxes overlap or facilities are partially obscured, so there was a problem of generating object labels that are different from the actual ones.

[0091] Accordingly, the database construction unit (215) can cluster the three-dimensional point cloud by class, remove points with low reliability, and then manage the buckets by separating them according to the image identifier (image_id) and the bounding box identifier (box_id). Subsequently, the database construction unit (215) can calculate the center point of each bucket and, if the distance between center points and image order conditions are satisfied, merge adjacent buckets to confirm the same object as a single label. As illustrated in FIG. 3, 'Drain' objects, 'Bollard' objects, 'Fence' objects, 'Manhole' objects, and 'Box' objects can be reliably labeled at different locations and recorded in the database. Through this, the database construction unit (215) can reliably identify objects even in urban environments with significant overlap or obscuration. For example, even when a guardrail appears overlapping in multiple continuously captured images, the database construction unit (215) can reliably identify it as a single object through the bucket merging process.

[0092] Additionally, the database construction unit (215) can correct drift phenomena occurring in a 3D point cloud generated based on a long-duration sequence. Conventional SfM (Structure from motion) or SLAM (Simultaneous Localization and Mapping) based techniques had a problem where the point cloud became bent as the sequence lengthened, and bundle adjustment or closure techniques to correct this were difficult to apply because they required a large amount of computation or constraints.

[0093] Accordingly, the database construction unit (215) can repeatedly apply a rotation vector to extract a road surface from a point cloud and make it into a horizontal plane. In each iteration, the upper and lower point clouds are cut to minimize residual error, and finally, a stabilized horizontal point cloud can be generated within a few seconds.

[0094] Figure 4 (a) shows point cloud data before correction, where the road appears curved and the object appears tilted, and (b) shows point cloud data after correction, where the road surface is leveled and the object is corrected into an upright form. The database construction unit (215) can ensure the consistency of the facility coordinates recorded in the database through such correction.

[0095] The database construction unit (215) can solve the distortion problem that occurs during the geographic coordinate system alignment process using GPS signals. Conventional GPS has limitations in that it is difficult to use directly in urban environments due to frequent signal jumps or drifts, and it is difficult to secure multiple reference points for correction through GCPs (Ground Control Points).

[0096] Accordingly, the database construction unit (215) can correspond the camera pose sequence and the GPS sequence in units of a certain interval, and estimate rotation, translation, and scale parameters for each interval and apply them to the 3D point cloud data. In addition, the database construction unit (215) can supplement missing GPS information by interpolating the preceding and succeeding data. Through this, the database construction unit (215) can stably align the location of an object with the real map coordinate system and ensure that the facilities stored in the database correspond to the actual road network.

[0097] In this way, the database construction unit (215) can effectively solve problems such as incomplete labeling, distortion due to long-term sequences, and instability of GPS-based alignment that occurred in existing technology. Therefore, the database construction unit (215) can accurately record location, shape, and relationship information of road facilities and build a reliable semantic database.

[0098] Additionally, the database construction unit (215) can configure the spatial relationships between identified road facility objects into a graph-based semantic database framework.

[0099] Specifically, the database construction unit (215) can define the center point of each road facility object as a node. For example, facilities such as guardrails, bollards, trees, traffic signs, and drains can have their center points calculated on a per-object basis and be represented as nodes.

[0100] The database construction unit (215) can define the relative positional relationship between two nodes as an edge and record it in the form of a graph. The edge can be generated according to the distance, direction, arrangement pattern, etc. between the nodes, for example, bollards arranged in a line at regular intervals can be connected by a straight edge, and trees and guardrails can have an edge defined by a relationship of "proximity."

[0101] As illustrated in FIG. 5, the database construction unit (215) can form a network-shaped graph by representing the center points of road facility objects as nodes based on images and point cloud data captured in an actual road environment and connecting the relationships between objects with edges. This graph can be used to reflect not only simple object location information but also the arrangement patterns and interrelationships between facilities in the database.

[0102] The database construction unit (215) can provide a basis for road facility managers to consider not only specific objects but also relationships with surrounding facilities by constructing a graph-based semantic database framework. For example, the graph can be usefully utilized to establish a management strategy to review the condition of other guardrails in the same line when a guardrail in a specific section is damaged.

[0103] In the following configuration, the change detection unit (220) can generate three-dimensional point cloud data based on multiple images collected while driving on the road, identify road facility objects through this, and then detect changes in facilities by comparing the identified objects with existing objects recorded in the database.

[0104] Meanwhile, the object identification process of the change detection unit (220) has substantially the same configuration as the object identification function of the database construction unit (215) described earlier. Therefore, redundant descriptions will be omitted.

[0105] The change detection unit (220) can look up past information of the same facility by matching the identified road facility object with an object recorded in the database based on a Universally Unique Identifier (UUID). For example, if a specific guardrail is identified as "UUID 0001" at the time of driving "T", the change detection unit (220) can directly compare the state of the two times by calling up the object information of the past time "T-1" which has the same UUID.

[0106] The change detection unit (220) receives image patches extracted from multiple images of the same object as input and can extract visual features (deep features) that are invariant even when the shooting environment changes. To this end, the change detection unit (220) can apply a convolutional neural network (CNN)-based embedding model, and, for example, can use a general-purpose neural network such as "ResNet" to convert texture, edge, and contour information of the object surface into high-dimensional vectors. This process enables the essential pattern to be represented by the same feature vector, even if the same road sign appears bright due to sunlight reflection during the day and its hue changes due to streetlights at night.

[0107] When multiple visual feature vectors are generated for the same object, the change detection unit (220) can apply an aggregator to integrate them into a single representative vector. At this time, a simple average may be applied, and an attention mechanism may be used to give greater weight to image patches with higher confidence.

[0108] The change detection unit (220) can calculate the Euclidean distance to compare representative vectors for the same object at two points in time. However, it is not limited to this, and the change detection unit (220) may apply various distance calculation methods such as cosine similarity and Mahalanobis distance, and furthermore, the distance calculation method may be included in the model's learning process to produce an optimized indicator through an aggregator.

[0109] For example, if the change detection unit (220) determines that there is no change because the distance between the two vectors is calculated to be very small, such as [0.11, 0.32, 0.95] at time point "T" and [0.12, 0.35, 0.97] at time point "T+1", the change detection unit (220) can determine that there is no change if a vector with a large difference, such as [0.52, 0.80, 0.10], is generated at time point "T+1".

[0110] The change detection unit (220) can calculate the amount of change based on the calculated Euclidean distance and estimate the type of change by referring to a pre-stored threshold table for each type of change. For example, the change detection unit (220) can determine that if the distance is 0.1 or less, it is simple contamination; if it is 0.3 or more, it is a crack; and if it is 0.5 or more, it is damage. These threshold values ​​can be derived through past learning data and statistical analysis.

[0111] The change detection unit (220) can classify the importance of maintenance based on the estimated type of change and determine the priority of repairs according to the grade. For example, the change detection unit (220) may classify discoloration on the surface of a sign as low importance and repair it during a regular inspection, and classify structural damage to a guardrail as high importance and determine that urgent repairs are required. Through this, the change detection unit (220) can simultaneously ensure efficiency and safety in managing road facilities.

[0112] Additionally, the change detection unit (220) can identify multiple facilities of the same type that exist within a certain distance among the road facilities recorded in the database. The change detection unit (220) can compare the similarity between single representative vectors for these adjacent facilities and determine that facilities with a similarity lower than a reference value are abnormal.

[0113] For example, the change detection unit (220) may define 10 bollards installed at regular intervals along a road section as a set of adjacent objects. In this case, if the representative vectors of 9 bollards maintain a high degree of similarity to each other, while the representative vector of a specific bollard is significantly different from the others, the change detection unit (220) may estimate that deformation or damage has occurred in the bollard.

[0114] The change detection unit (220) may apply various metrics, such as cosine similarity and Euclidean distance, to calculate the similarity of representative vectors. For example, if the change detection unit (220) is a normal pattern for guardrail objects of the same type to show a cosine similarity of 0.95 or higher, it may classify a specific object as an outlier if its similarity is calculated to be 0.6 or lower.

[0115] The change detection unit (220) can evaluate relative differences within a group of facilities of the same type, rather than judging individual objects based solely on absolute standards through a comparison process between adjacent facilities. This allows for the exclusion of the influence of lighting changes or seasonal color changes that occur overall due to environmental factors, and enables the effective detection of abnormalities occurring only in specific objects.

[0116] Additionally, the change detection unit (220) can extract the geometric attributes of a road facility object identified without obscuration at close range and calculate the presence and degree of difference by comparing it with the past attributes of the object recorded in the database. The geometric attributes may include the object's centroid, principal axis vector, height, radius, etc. At this time, since the accuracy of the attribute estimation is proportional to the number of point clouds representing the object, a limit error is calculated together.

[0117] Additionally, the change detection unit (220) can calculate the difference in geometric attributes by comparing the identified road facility object with the three-dimensional point cloud data of the object recorded in the database. Geometric attributes may include the object's centroid, principal axis vector, height, radius, etc., and the change detection unit (220) can detect changes in shape by comparing these attribute values ​​at different points in time.

[0118] For example, if the change detection unit (220) determines that the amount of movement of the center point has exceeded a specific threshold value, if the center point coordinates of the guardrail post were recorded as (x=12.35, y=45.22, z=0.90) at a past time point but are measured as (x=12.50, y=45.30, z=0.80) at the current time point. In this case, the change detection unit (220) can estimate that the post has tilted or been partially damaged due to an impact.

[0119] The change detection unit (220) can detect the tilt of an object by comparing the main axis vector. For example, the change detection unit (220) can determine that the sign is tilted due to wind or impact if the main axis vector of the traffic sign was previously almost parallel (0° to 2°) to the z-axis direction but is now calculated to be tilted by 15° or more.

[0120] The change detection unit (220) can detect a change in size by comparing the height and radius of an object. For example, if the change detection unit (220) calculates that the radius of the drain was previously recorded as 0.25m but is now 0.18m, it can estimate that it is partially blocked by fallen leaves or debris. Conversely, if the height of the tree plate has increased compared to the past, it can suspect the possibility of subsidence of the surrounding ground or replacement of the structure.

[0121] In this way, the change detection unit (220) calculates a difference in geometric attributes, and if the difference is observed to be greater than a threshold value, it can determine that a change in the shape of the object has occurred. This comparison of geometric attributes may be useful for detecting structural deformation that cannot be explained by simple color changes or contamination.

[0122] As a result, the change detection unit (220) can precisely detect shape changes such as tilting, damage, and blockage of road facilities based on differences in geometric attributes such as center point, main axis vector, height, and radius.

[0123] Additionally, the change detection unit (220) can accumulate and record the amount of shape change of the identified road facility object at past points in time. The amount of shape change can be calculated as differences in geometric attributes such as movement of the object's center point, change in the slope of the main axis vector, and increase or decrease in height and radius, and the change detection unit (220) can store this as time-series data.

[0124] For example, if the center point coordinates of a specific guardrail post were (x=10.0, y=20.0, z=0.9) at time point "T0", but changed to (x=10.1, y=20.1, z=0.85) at "T1" and (x=10.3, y=20.3, z=0.80) at "T2", the change detection unit (220) can record the amount of movement at each time point and reflect in the database that the post is gradually tilting.

[0125] The change detection unit (220) can analyze patterns from accumulated change records to predict the possibility of damage to the facility at a future point in time. For example, if the change in the tilt of the sign shows a pattern of increasing by a certain angle every month, the change detection unit (220) can estimate the possibility that the sign will exceed a critical angle within the next 6 months. In addition, if the radius of the drain shows a pattern of decreasing over time, it can predict the time when it will be completely blocked by fallen leaves and sediment.

[0126] In this way, the change detection unit (220) does not merely record changes at past points in time individually, but can estimate the possibility of future damage in advance by utilizing the patterns of accumulated data. This can be used as important supporting data for road managers to establish a prior maintenance plan.

[0127] As a result, the change detection unit (220) can accumulate and record the amount of shape change at each past point in time and estimate the possibility of future damage to the facility based on the accumulated change pattern, thereby enabling predictive maintenance of road facility management.

[0128] Additionally, the change detection unit (220) can identify a moving object whose position changes over time within a plurality of images. Here, the moving object may include a moving object that temporarily occupies the road, such as a vehicle, pedestrian, bicycle, or motorcycle. The change detection unit (220) can separate such moving objects through an object detection model and time series analysis.

[0129] The change detection unit (220) can mask the three-dimensional point cloud data corresponding to the flowing object so that it is not considered during the database comparison process. For example, if a vehicle is parked at a specific point in time and there is no vehicle at another point in time on the same road section, the change detection unit (220) can remove or ignore the point cloud data so that the vehicle is not mistakenly recognized as a road facility.

[0130] The change detection unit (220) can analyze positional changes between time series frames to effectively identify moving objects. For example, a set of point clouds moving at a constant speed in multiple consecutive images can be determined as a vehicle and can be distinguished from stationary facilities such as streetlights or guardrails. Additionally, the change detection unit (220) can apply an object detection neural network such as "YOLO" or "Faster R-CNN" to classify vehicles and pedestrians into pre-trained classes and process them separately from road facilities.

[0131] The change detection unit (220) can prevent unnecessary change detection when compared with fixed facilities recorded in the database through a fluid object masking process. For example, even if a temporarily parked vehicle appears in the data captured daily in a specific road section, the change detection unit (220) can accurately detect changes in actual facilities by ignoring it without recording it as a change.

[0132] As a result, the change detection unit (220) can prevent false detections caused by temporary objects such as vehicles or pedestrians by identifying the flowing object in advance and masking the point cloud data, and can increase the accuracy of detecting changes in road facilities.

[0133] Additionally, the change detection unit (220) can detect a change by extracting the visual features of the identified road facility object and comparing them with the visual features of the same object at the same time recorded in the database.

[0135] Referring briefly to FIG. 6, a change detection unit (220) according to one embodiment of the present invention will be described in more detail.

[0136] FIG. 6 is an illustrative diagram for explaining a change detection unit according to one embodiment of the present invention.

[0137] Referring to FIG. 6, the change detection unit (220) can receive multiple image patches as input for the same object and apply an embedding model. The embedding model can produce a high-dimensional embedding vector from each image patch, and the produced vector can numerically represent the visual features of the object. For example, a feature matrix A can be extracted from the first image patch, a feature matrix B from the second image patch, and a feature matrix C from the third image patch.

[0138] The change detection unit (220) can input the calculated multiple embedding vectors into an aggregator to compress them into a single representative vector. Here, the aggregator can generate a representative vector by applying a simple average, a weighted average, or an attention-based integration method to give greater weight to features with higher reliability among the multiple vectors.

[0139] For example, if there are three images of the same traffic sign taken under different lighting conditions, the change detection unit (220) can extract visual features from each image to form an embedding vector, and then compress it into a single representative vector through an aggregator. This single representative vector can be used for database comparison as a unique numerical value representing the appearance of the sign.

[0140] Accordingly, the change detection unit (220) can mitigate the effects of lighting, shooting angle, and partial occlusion that may occur in multiple image patches, and can consistently manage the intrinsic visual characteristics of an object as a single representative vector. This enables time-series comparison of the same object and improves the accuracy of facility change detection.

[0142] Below, hardware for implementing the logical components of the management server (200) as described above will be explained in more detail.

[0143] FIG. 7 is a hardware configuration diagram of an analysis server according to one embodiment of the present invention.

[0144] As illustrated in FIG. 7, the management server (200) may be configured to include a processor (250), memory (255), transceiver (260), input / output device (265), data bus (270) and storage (275).

[0145] Specifically, the processor (250) can implement the operation and function of the management server (200) based on instructions according to software (280a) which implements a road facility management method and a road facility database construction method residing in memory (255).

[0146] Software (280a) in which a road facility management method and a road facility database construction method according to embodiments of the present invention are implemented can be loaded in the memory (255).

[0147] The transceiver (260) can transmit and receive data with the information collection device (100).

[0148] The input / output device (265) can receive signals necessary for the operation of the management server (200) or output calculation results to the outside according to the command of the processor (250).

[0149] The data bus (270) is connected to the processor (250), memory (255), transceiver (260), input / output device (265), and storage (275), and can serve as a passage for transferring data between each component.

[0150] Storage (275) can store an Application Programming Interface (API), library files, resource files, etc., necessary for the execution of software (280a) in which a road facility management method and a road facility database construction method according to embodiments of the present invention are implemented. Additionally, storage (275) can store software (280b) and a database (285) in which a method according to embodiments of the present invention is implemented.

[0151] According to one embodiment of the present invention, software (280a, 280b) for implementing a road facility management method that resides in memory (255) or is stored in storage (275) may be a computer program recorded on a recording medium for the purpose of executing the steps of: the processor (250) collecting a plurality of images that are continuously captured while driving on a road from an information collection device; the processor (250) extracting feature points within the plurality of collected images; the processor (250) matching the extracted feature points of the plurality of collected images to obtain three-dimensional point cloud data including depth information; the processor (250) identifying an object corresponding to a road facility based on the obtained three-dimensional point cloud data; and the processor (250) detecting a change in the road facility by comparing the three-dimensional point cloud data corresponding to the road facility based on a pre-stored facility database.

[0152] According to another embodiment of the present invention, software (280a, 280b) for implementing a method for constructing a road facility database that resides in memory (255) or is stored in storage (275) may be a computer program recorded on a recording medium to execute the steps of: the processor (250) collecting a plurality of images that are continuously captured while driving on a road from an information collection device; the processor (250) extracting feature points within the plurality of collected images; the processor (250) matching the extracted feature points of the plurality of collected images to obtain three-dimensional point cloud data including depth information; the processor (250) identifying an object corresponding to a road facility based on the obtained three-dimensional point cloud data; and the processor (250) constructing a database for a road facility based on the identified object.

[0153] More specifically, the processor (250) may include an Application-Specific Integrated Circuit (ASIC), other chipsets, logic circuits, and / or data processing devices. The memory (255) may include Read-Only Memory (ROM), Random Access Memory (RAM), flash memory, memory cards, storage media, and / or other storage devices. The transceiver (260) may include a baseband circuit for processing wired and wireless signals. The input / output device (265) may include input devices such as a keyboard, mouse, and / or joystick, image output devices such as a Liquid Crystal Display (LCD), Organic Light Emitting Diode (OLED), and / or Active Matrix OLED (AMOLED), and printing devices such as a printer and plotter.

[0154] When the embodiments included in this specification are implemented in software, the above-described method may be implemented as a module (process, function, etc.) that performs the above-described function. The module may reside in memory (255) and be executed by a processor (250). Memory (255) may be located inside or outside the processor (250) and may be connected to the processor (250) by various widely known means.

[0155] Each component illustrated in FIG. 7 may be implemented by various means, for example, hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, one embodiment of the present invention may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers, microprocessors, etc.

[0156] In addition, in the case of implementation by firmware or software, an embodiment of the present invention may be implemented in the form of a module, procedure, function, etc., that performs the functions or operations described above, and may be recorded on a recording medium readable through various computer means. Here, the recording medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the recording medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs (Compact Disk Read Only Memory) and DVDs (Digital Video Disks); magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions may include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. Such hardware devices may be configured to operate as one or more software to perform the operation of the present invention, and vice versa.

[0158] FIG. 8 is a flowchart illustrating a road facility management method according to an embodiment of the present invention.

[0159] Referring to FIG. 8, first, in step S110, the management server can collect multiple images continuously captured from an information collection device driving on the road. At this time, the information collection device can acquire images at regular spatial intervals by driving a shutter in proportion to the number of rotations of the vehicle wheels, and the management server can receive metadata such as GPS coordinates, shooting time, and vehicle speed along with the images.

[0160] In step S120, the management server can extract feature points from multiple collected images. After correcting image distortion, the management server can secure stable feature points unaffected by lighting, shooting angle, or partial occlusion by applying SIFT (Scale-Invariant Feature Transform) or SuperGlue-based algorithms.

[0161] In step S130, the management server can obtain 3D point cloud data containing depth information by matching the correspondence relationships between the extracted feature points. The management server can calculate the spatial coordinates of each feature point using triangulation and improve the reliability of the point cloud by applying a Random Sample Consensus (RANSAC)-based error removal process.

[0162] In step S140, the management server can identify road facility objects based on the acquired 3D point cloud data. The management server can apply an object detection model to identify objects of interest within the image using bounding boxes, and extract point cloud data included in the area to classify them into traffic signs, guardrails, bollards, drains, etc.

[0163] Subsequently, in step S150, the management server can construct a database based on the identified objects. The management server records UUIDs, types, locations, and visual features, along with geometric attributes such as the object's center point, principal axis vector, height, and radius, in the database, enabling management on a facility-by-facility basis. Additionally, it can generate a relationship graph based on the relative distance between objects and store it in the database in the form of a semantic network.

[0164] Meanwhile, in step S160, the management server can detect changes by performing a comparison with the database based on the identified objects. The management server can detect changes by mapping identical objects based on UUIDs and comprehensively analyzing similarity between representative vectors, differences in geometric attributes, and changes in visual features. Furthermore, it can determine maintenance priorities by classifying the change detection results by type and assigning importance grades.

[0166] As described above, preferred embodiments of the present invention have been disclosed in this specification and drawings; however, it is obvious to those skilled in the art that other variations based on the technical spirit of the present invention are possible in addition to the embodiments disclosed herein. Furthermore, although specific terms have been used in this specification and drawings, they are used merely in a general sense to facilitate the explanation of the technical content of the present invention and to aid in understanding the invention, and are not intended to limit the scope of the present invention. Accordingly, the detailed description above should not be interpreted restrictively in any respect and should be considered illustrative. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the present invention are included within the scope of the present invention. Explanation of the symbols

[0167] 100: Information gathering device 200: Management server 205 : Communications Unit 210 : Input / Output Unit 215: Database Construction Department 220: Change Detection Department 225 : Storage

Claims

Claim 1 A management server collects a plurality of images continuously captured from an information collection device while driving on a road; the management server extracts feature points within the collected plurality of images; the management server obtains three-dimensional point cloud data including depth information by matching the extracted feature points of the collected plurality of images; the management server identifies an object corresponding to a road facility based on the obtained three-dimensional point cloud data; and the management server constructs a database for the road facility based on the identified object. A method for constructing a road facility database, comprising: a method for obtaining a method wherein the obtaining step is characterized by estimating depth information of each feature point through triangulation based on the correspondence relationship between the extracted feature points, and calculating the 3D coordinates of the 3D point cloud data through the estimated depth information; the identifying step is characterized by applying a pre-trained object detection model to each of the plurality of images to specify the location of the object of interest as a bounding box, and extracting 3D point cloud data associated with the feature points included in the specified bounding box to identify an object corresponding to a road facility; and the identifying step is characterized by clustering the obtained 3D point cloud data by class, calculating the clustering confidence of each point, excluding points where the calculated clustering confidence is less than a preset value, and classifying each point into buckets according to an image identifier and a bounding box identifier. Claim 2 A method for constructing a road facility database according to claim 1, wherein the collecting step is characterized by collecting a plurality of images captured by driving a shutter in proportion to the number of rotations of the vehicle wheels during the driving process of a vehicle equipped with the information collecting device. Claim 3 A method for constructing a road facility database according to claim 1, wherein the identifying step calculates the centroid of the classified bucket, and if the centroid distance between different buckets is less than a preset value and the image sequence interval is less than a preset value, the buckets are merged to assign a single label, and the point cloud is reconstructed based on the label. Claim 4 A method for constructing a road facility database according to claim 3, wherein the constructing step comprises calculating a reference plane corresponding to a road surface for the acquired 3D point cloud data, and correcting the 3D point cloud data by repeatedly applying a rotation transformation to convert the calculated reference plane into a horizontal plane. Claim 5 A method for constructing a road facility database according to claim 4, wherein the constructing step involves calculating transformation parameters by correlating the corrected camera pose and position information recorded during the shooting process in segment units, and aligning the 3D point cloud data obtained based on the calculated transformation parameters with a real-world coordinate system. Claim 6 A method for constructing a road facility database according to claim 1, wherein the constructing step involves labeling each point of the 3D point cloud data with object class information to construct a 3D semantic map classified by road facility object. Claim 7 A method for constructing a road facility database according to claim 6, wherein the three-dimensional semantic map is stored in cluster units that combine a Universally Unique Identifier (UUID) for an object corresponding to the identified road facility and metadata including type information, shape information, and location information for the object. Claim 8 delete Claim 9 delete Claim 10 delete

Images