A High-Precision Map Incremental Update Method and Device Based on Road-Side Cameras

Abstract

The present invention discloses a method and device for incrementally updating high-precision maps based on roadside cameras. The method comprises: acquiring image data from multiple cameras, stitching the images, and cropping the stitched images to obtain a road image that retains only the road area; comparing road images from previous and subsequent time series, and initiating a map update if a newly added area exists; otherwise, terminating the process; after initiating the map update, extracting road traffic map elements from the road image at the current time series, and constructing map element identifiers based on road traffic map rules; and integrating the road traffic map elements and map element identifiers with the current latest map to complete the incremental update of the high-precision map. The present invention reduces the cost of updating high-precision maps and increases the frequency of map updates.

Description

Technical Field

[0001] The present invention belongs to the technical field of map incremental updating, and in particular relates to a high-precision map incremental updating method and device based on a roadside camera. Background Art

[0002] High-precision road maps, as the core technology for autonomous driving, have become a strategic advantage that global tech giants are vying for. The core goal of high-precision mapping technology is the production and updating of map data. Existing road traffic maps are primarily based on data from ground-based mapping vehicles, aerial photogrammetry, and satellite remote sensing. These systems struggle to automatically collect lane information, traffic signs, and verify the topological relationships and association rules of map elements. Mobile mapping technologies, whether based on mobile mapping vehicles, aerospace observation systems, or large-scale vehicles based on crowdsourcing, struggle to achieve real-time road traffic map updates.

[0003] With the development of intelligent, connected transportation systems in cities, roadside sensors will become an effective data source for HD map updates. For the static and quasi-static layers of HD maps, roadside sensors like vision and lidar can detect changes in road channelization, road markings, and other areas in real time, reporting this information and facilitating HD map updates. For the dynamic and quasi-dynamic layers of HD maps, roadside and vehicle-based sensing capabilities are closely integrated to increase information redundancy, enable cross-verification, and integrate them, and provide more accurate and rich dynamic information for HD maps.

[0004] Existing high-precision map update methods include the Chinese patent document with application number CN202110243239.1, which discloses a crowdsourced high-precision map update method, system, and medium for autonomous driving. The method mainly collects and obtains on-board sensor data from multiple vehicles, transmits the multi-source information to the blockchain processing module for block map updates; transmits the updated map module in the network back to the vehicle side for vehicle-side updates, replaces the unupdated block map data, and completes the data update of the entire link. Another example is the Chinese patent document with application number CN201710359395.8, which discloses a high-precision dynamic map data processing system and method based on roadside sensors. The system is divided into three main parts: a data acquisition system, a map generation system, and a high-precision map visual editing system to achieve map updates.

[0005] Although existing map update methods can achieve map updates, they have disadvantages:

[0006] 1. Raw data from multiple sources can easily lead to uneven data quality, and the diversity of data formats and specifications is obvious, which seriously affects the stability and universality of high-precision map data processing algorithms.

[0007] 2. Intelligent, connected, and autonomous vehicles require real-time, high-precision maps, but crowdsourcing-based map production cannot meet the real-time updates required for autonomous driving.

[0008] 3. The map production model based on roadside sensors involves multiple sensor types, complex calculations, and high hardware costs and computing resource consumption. Summary of the Invention

[0009] One of the objectives of the present invention is to provide a high-precision map incremental update method based on roadside cameras, which reduces the cost of high-precision map updates and increases the frequency of map updates.

[0010] To achieve the above object, the technical solution adopted by the present invention is:

[0011] A method for incrementally updating a high-precision map based on a roadside camera, comprising:

[0012] Acquire image data from multiple cameras for image stitching, and crop the stitched image to obtain a road image that only retains the road area;

[0013] Compare the road images before and after the time series. If there is a new area, start the map update; otherwise, end it.

[0014] After starting the map update, extract the road traffic map elements in the road image at the current time sequence, and construct the map element identification based on the road traffic map rules;

[0015] The road traffic map elements and map element identifiers are integrated with the latest map to complete the incremental update of the high-precision map.

[0016] Several optional methods are also provided below, but they are not intended to be additional limitations on the above-mentioned overall solution. They are merely further supplements or optimizations. Under the premise that there are no technical or logical contradictions, each optional method can be combined separately for the above-mentioned overall solution, or multiple optional methods can be combined.

[0017] Preferably, the arrangement rules of the cameras are as follows:

[0018] Observation stations are set up at intervals along the road. Two cameras facing opposite directions are set at each observation station at a preset height above the road. The main axes of the two cameras are consistent with the radial direction of the road, and the angle between them and the plane of the road is θ. Image data is provided by observation station when the map is updated.

[0019] Preferably, before acquiring image data of multiple cameras, a joint calibration of the multiple cameras is also included. The joint calibration includes: calibrating the multiple cameras based on the latest map to obtain camera distortion parameters.

[0020] Preferably, the camera is a visible light high-resolution camera.

[0021] Preferably, the step of acquiring image data from multiple cameras and performing image stitching comprises:

[0022] Based on the camera distortion parameters obtained by camera calibration, the image data is rectified to restore the spatial observation value of the image data;

[0023] Remove moving target areas from image data through linear change detection;

[0024] Based on corner detection, feature corners are extracted from image data to establish the topological features of the image;

[0025] Extract lane lines, stop lines, and lane markings from the latest map, perform feature point generalization calculations, and construct a feature point matrix.

[0026] The constructed feature point matrix is used as the main observation value, and the topological features of the established image are sequentially fused at multiple scales to complete the image stitching.

[0027] Preferably, extracting road traffic map elements from the road image in the current time sequence includes:

[0028] Using image edge detection operators, linear features are extracted from road images. Linear features are visible linear targets on the road surface, including road edge lines, lane lines, zebra crossings, stop lines, lane markings, and parking area edge lines.

[0029] Based on the hidden Markov chain model, point features are extracted from linear features, and the observation space, state space, state transition equation and output probability matrix are constructed to estimate the maximum likelihood road edge line combination, thereby determining the overall road surface width and lane line combination;

[0030] Based on the overall road width and lane line combination, the lane line combination is topologically associated according to the length and distance attributes of the lane line, and continuous lane lines are output;

[0031] According to the linear elements in the extracted road image, lane lines and edge lines in the linear elements are removed, and block marks of the road surface are output. The lane lines, edge lines and block marks are used as extracted road traffic map elements.

[0032] Preferably, the method of extracting point elements from linear elements based on a hidden Markov chain model and constructing an observation space, a state space, a state transfer equation and an output probability matrix includes:

[0033] Observation space: The linear features in the road image are extracted as the observation space, and the data format is line segments;

[0034] State space: extract discrete point features in the road area as state space, and the data format is point;

[0035] The state transfer equation is constructed as follows:

[0036]

[0037] Where a ij From state q i To state q j The probability of transition, q i With q j Refers to different states in the observation space, which are expressed as linear elements, and d is the linear element q i With the linear feature q j distance;

[0038] The output probability matrix is constructed as follows:

[0039]

[0040] Where, P(R l |O t ) is in state R l Lower observation value O t The output probability, R l is the lth state, that is, the lth state value in the state space, expressed as a point element, O t is the observed value at time t, distance between R l to O t Indicates state R l With the observation value O t The distance between k is the kth state in the sum.

[0041] Preferably, the fusion of the road traffic map elements and map element identifiers with the latest map includes:

[0042] The extracted and constructed road traffic map elements and map element identifiers are matched with the latest map. The maximum likelihood map element mapping relationship is constructed for the elements that can be matched. The elements that cannot be matched are regarded as abnormal elements and their features are expressed.

[0043] In the local environment, the map elements within the current observation station are reconstructed based on the maximum likelihood mapping relationship of map elements and the characteristic expression of abnormal elements; in the global environment, the map elements in the global scope are updated based on the characteristic expression of abnormal elements.

[0044] Preferably, the high-precision map incremental update method based on roadside cameras also includes performing a logical consistency check after the map is fused, and the rules based on which the logical consistency check is based include the geographical attributes of map elements, digital road network topology associations and map database construction rules.

[0045] The proposed method for incrementally updating high-precision maps based on roadside cameras primarily implements a novel approach to producing high-precision road traffic maps. The data content adheres to the high-precision map data standards of mainstream autonomous driving manufacturers. This not only reduces the total cost of sensor hardware but also enriches the map semantics, meeting the application requirements of intelligent connected autonomous vehicles. Consistent data quality ensures the stability and universality of map updates, without relying on crowdsourcing, significantly increasing the frequency of map updates.

[0046] The second purpose of the present invention is to provide a high-precision map incremental update device based on roadside cameras to reduce the cost of high-precision map updates and increase the frequency of map updates.

[0047] To achieve the above object, the technical solution adopted by the present invention is:

[0048] A high-precision map incremental update device based on a roadside camera includes a processor and a memory storing a plurality of computer instructions. When the computer instructions are executed by the processor, the steps of the high-precision map incremental update method based on a roadside camera are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of the method for incremental updating of high-precision maps based on roadside cameras of the present invention;

[0050] Figure 2 This is a schematic diagram of the arrangement of multiple cameras of the present invention;

[0051] Figure 3 This is a flow chart of image stitching of the present invention;

[0052] Figure 4 This is a flow chart of extracting road traffic map elements according to the present invention;

[0053] Figure 5 Schematic diagram of the structure of the hidden Markov chain model of the present invention;

[0054] Figure 6 This is a flowchart of the map fusion of the present invention;

[0055] Figure 7 FIG. 1 is a structural diagram of a hardware platform in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0058] To overcome the shortcomings of map updates in the existing technology, this embodiment provides a high-precision map incremental update method based on roadside cameras, realizing a new method for producing high-precision road traffic maps. The data content complies with the high-precision map data standards of mainstream autonomous driving manufacturers. Not only is the total cost of sensor hardware lower, but the map semantics are also rich, meeting the application requirements of intelligent connected autonomous driving vehicles. Logical verification is performed based on factors such as spatial location constraints, traffic network topology rules, and traffic sign consistency understanding. The timeliness of map data reaches quasi-real time, which is the basic information service supporting fully autonomous driving.

[0059] Specifically, such as Figure 1 As shown, the high-precision map incremental update method based on roadside cameras of this embodiment includes the following steps:

[0060] Step S1: Acquire image data from multiple cameras, perform image stitching, and crop the stitched image to obtain a road image that only retains the road area.

[0061] This embodiment performs image processing on the image data collected by multiple cameras, including image correction, image feature point extraction, and multi-scale feature fusion. The image data of multiple cameras are stitched into a single image, and then the image is cropped based on the geographical scope of the traffic road, retaining road areas such as motor vehicle lanes, non-motor vehicle lanes, pedestrian areas, and the green belt area in the center of the road.

[0062] To improve the accuracy of image data, multiple cameras need to be calibrated before acquiring image data from them. This embodiment uses the most recent map since the last update as the reference map. Based on the reference map, multiple cameras are calibrated to obtain camera distortion parameters.

[0063] Considering that the amount of data collected by the camera will directly affect the map update speed, and the larger the range, the higher the map update frequency may be, this embodiment proposes a suitable camera arrangement. Figure 2 As shown, the camera layout rules are as follows:

[0064] Observation stations are set up at intervals along the road. At each observation station, two cameras facing opposite directions are set at a preset height above the road. The main axes of the two cameras are aligned with the radial direction of the road and the angle θ with the plane of the road is consistent. When the map is updated, image data is provided by observation station. In this case, the map update only needs to be processed based on the image data of two cameras, which is faster and has a smaller update range.

[0065] In the camera arrangement, for example, cameras N_L and N_R are at observation station N, and (N+1)_L and (N+1)_R are at observation station N+1 (see Figure b in the diagram). The cameras are all positioned 6 meters above the road surface (the height is adjustable), and the angle θ between the camera's main axis and the road surface plane is 28 degrees (adjustable), as shown in Figure a), aligning with the radial direction of the road. The area between adjacent observation stations also represents the overlapping field of view of the corresponding cameras at those stations, as shown in Figure c in the diagram. The black area represents the road surface, the diagonal lines (low on the left and high on the right) represent the field of view of camera (N+1)_L, and the diagonal lines (high on the left and low on the right) represent the field of view of camera N_R. The area between the two intersecting diagonal lines represents the overlapping field of view of the two cameras.

[0066] The camera in this embodiment is preferably a high-resolution visible light camera. In other embodiments, other types of cameras may be replaced or added, such as a night vision infrared camera. Since the update is based on the image data of multiple cameras (in this embodiment, two cameras located at the same observation station are used as a unit), image stitching is a key part of map increment processing, such as Figure 3 As shown, this embodiment provides an image stitching as follows:

[0067] Step 1: Based on the camera distortion parameters obtained from camera calibration, the image data is rectified to restore the spatial observation value of the image data. By grayscale stretching, the image processing accuracy and algorithm sensitivity are improved.

[0068] Step 2: Remove moving target areas in the image data through linear change detection, such as the areas where road traffic participants are located in the image.

[0069] Step 3: Based on corner detection, extract the characteristic corner points in the image data and establish the topological features of the image.

[0070] Step 4: Extract lane lines, stop lines, and lane markings from the latest map, perform feature point generalization calculations, and construct a feature point matrix, including the endpoints of linear features and the feature points of lane markings.

[0071] Step 5: Using the constructed feature point matrix as the main observation value, perform multi-scale feature fusion on the topological features of the established image in turn to achieve the fusion of the new image and the main observation value, and complete the image stitching.

[0072] The multi-scale feature fusion can be implemented based on existing methods, such as concat fusion, add fusion, or DCA feature fusion. Furthermore, the image cropping in this embodiment can be performed using existing image cropping algorithms (such as the Cohen-Sutherland algorithm, midpoint segmentation method, Liang-Barsky algorithm, etc.) for automatic recognition and cropping, or manual cropping can be performed through a visual operation interface, which is not limited in this embodiment.

[0073] Step S2: Compare the road images before and after the sequence. If there is a new area, start the map update; otherwise, end.

[0074] Since the need for map updates arises from road traffic events, including road maintenance, traffic accidents, and natural disasters, the frequency of map updates is actually very low. Therefore, this embodiment sets a map update detector to detect the need for map updates.

[0075] The map update detector compares the images before and after the time series and detects new changed areas in the image based on the linear change detection method. This is considered an abnormal situation, that is, there is a need for a map update. The next step is to start the map update calculation; otherwise, it returns to the previous step for detection.

[0076] Step S3: After starting the map update, extract the road traffic map elements in the road image in the current time sequence, and construct map element identifiers based on the road traffic map rules.

[0077] On the cropped map, road traffic map elements are extracted, including physical signs such as lane lines, parking areas, stop lines, zebra crossings, edge lines, lane markings, etc., and map element signs such as road center lines are constructed based on road traffic map rules.

[0078] like Figure 4 As shown, extracting road traffic map elements from the road image at the current time sequence includes:

[0079] Step 1: Use the image edge detection operator to extract linear features from the road image. Linear features are visible linear targets on the road surface, including edge lines on both sides of the road, lane lines on the road surface, zebra crossings, stop lines, lane markings, and parking areas.

[0080] Step 2: Based on the hidden Markov chain model, point elements are extracted from linear elements, and the observation space, state space, state transfer equation and output probability matrix are constructed to estimate the maximum likelihood road edge line combination. The edge line is then interpolated to form a complete road surface line, thereby determining the overall road surface width and lane line combination.

[0081] Among them, such as Figure 5 As shown in Figure 2, the construction methods of the five basic models are as follows:

[0082] Observation space: Linear features extracted from road images are used as observation space, and the data format is line segments.

[0083] State space: Through the statistical information of the original road, such as the traffic road area, the discrete point features in the road area are extracted as the state space. The data format is the point data in the image.

[0084] Initial state: manually pick up the starting position to form a reliable and accurate initial value.

[0085] State transfer equation: Based on the spatial distance between the point position and the linear feature, a state transfer equation is constructed. In this embodiment, when calculating the linear feature, the linear feature is discretized into points. The shortest distance between the points on the two linear features is calculated as the distance between the two linear features. The specific equation is as follows:

[0086]

[0087] Where a ij From state q i To state q j The probability of transition, q i With q j Refers to different states in the observation space, which are expressed as linear elements, and d is the linear element q i With the linear feature q j distance.

[0088] Output probability matrix: This embodiment uses the reciprocal of the distance between two points as the numerator, and the sum of the reciprocals of the distances between the point and multiple adjacent points as the denominator. Linear features are also discretized into points for calculation. The matrix is as follows:

[0089]

[0090] Where, P(R l |O t ) is in state R l Lower observation value O t The output probability, R l is the lth state, that is, the lth state value in the state space, expressed as a point element, O tis the observed value at time t, distance between R l to O t Indicates state R l With the observation value O t The distance between k is the kth state in the sum.

[0091] By constructing a hidden Markov chain model and then performing random estimation calculations, the road edge line combination can be obtained, thereby determining the overall road surface width and lane line combination.

[0092] Step 3: Based on the overall road width and lane line combination, the lane line combination is topologically associated according to the length and distance attributes of the lane lines, and continuous lane lines are output.

[0093] Step 4: Based on the linear features in the extracted road image, the lane lines and edge lines are removed from the linear features within the constraints of the lane lines output in Step 3. The resulting road block markers and their coordinate locations are output. Block markers include visible features such as zebra crossings, stop lines, lane markings (such as straight, left, and right signs), and parking areas. The resulting lane lines, edge lines, and block markers are ultimately used as the extracted road traffic map elements.

[0094] In addition, in order to improve the usability of the map, this embodiment constructs map element identifications, including road center lines, digital road network topology associations, etc., based on the road traffic map rules formulated in the current existing technology.

[0095] Step S4: Integrate the road traffic map elements and map element identifiers with the latest map to complete the incremental update of the high-precision map.

[0096] Based on the elements extracted in the above steps, the road traffic map is reconstructed and reorganized, and then integrated with the map at the reference time to update the map data. In order to improve the efficiency of map updating, this embodiment adopts the method of reconstructing the map elements of the current observation area in a refined manner in a local environment; in a global environment, a corrective digital expression is performed on the attributes that do not affect the overall performance of the road. Specifically, Figure 6 As shown, the road traffic map elements and map element identifiers are integrated with the latest map, including:

[0097] Step 1: Correlate and match the extracted and constructed road traffic map elements and map element identifiers with the latest map. This matching includes explicit elements such as road edge lines, lane markings, parallel lines, zebra crossings, stop lines, lane markings, and other elements, as well as constructed elements such as the road centerline. Matching constraints include geographic location, road attributes, and digital road network topology.

[0098] The second step, feature attribute estimation, comprehensively optimizes the matching results and constructs a maximum likelihood mapping relationship for the matched features. This primarily involves calculating the geographic attributes of feature instances, such as location, orientation, and size. Because the actual size, orientation, and relative positions of map features must conform to road traffic construction standards, the feature attribute estimation calculation primarily utilizes maximum likelihood estimation. The value with the highest probability meeting multiple constraints is accepted as the true value.

[0099] Step 3: Filter abnormal elements, treat the unmatched elements as abnormal elements, and obtain the abnormal element instances with the greatest similarity.

[0100] Step 4: Characterize the abnormal elements, that is, express the location area, orientation, type, other map elements affected, and other features of these newly appeared elements, and submit them to the system for manual review and data memorandum.

[0101] Step 5: In the local environment, map elements within the current observation station are reconstructed based on the maximum likelihood mapping relationship of map elements and the characteristic expression of abnormal elements; in the global environment, map elements in the global scope are updated based on the characteristic expression of abnormal elements.

[0102] In addition, in order to improve the accuracy of map updates, this embodiment performs a logical consistency check after map fusion, including geographic spatial location, geographic attributes of map elements, digital road network topology association, traffic rule constraints, map database construction rules, etc.

[0103] It should be noted that the geographic attributes of map elements, road network topology constraints, and map database construction rules are all publicly available and consistent with reality. This prevents the appearance of non-realistic elements in the map, thereby optimizing the map and improving the accuracy of map expression. During logical consistency verification, if non-compliant elements appear, they can be directly corrected according to the rules or reported as an exception for manual inspection and correction.

[0104] This invention has created a new theory and method for lane-level high-precision road crowdsourcing mapping based on roadside cameras, breaking through technologies such as high-precision cleaning of trajectory big data and lane-level road updating, and forming a new high-precision road crowdsourcing mapping technology that is easy to collect, low-cost, fast to update, and can be applied globally. It supports global cutting-edge applications such as lane-level high-precision navigation and unmanned vehicles, realizes a technological leap from professional road mapping to ubiquitous mapping, and promotes the development of surveying and mapping geographic information technology.

[0105] Based on image data from roadside cameras, this method performs image stitching, road traffic map feature extraction, feature correlation matching, and abnormal feature screening, reducing the cost of high-precision map updates and increasing the frequency of map updates. Based on a hidden Markov chain model, it performs global matching on initially identified map features, improving the accuracy and efficiency of map feature recognition. When establishing features, it first identifies road edges, then lane lines, and then constructs road centerlines based on the results of the former. This significantly improves the accuracy of feature recognition and enhances the accuracy of road sign understanding. This method achieves automated updates of road traffic maps with an update frequency of 1 second and enables integrated map stitching of the entire road network, resulting in high data accuracy and rich map semantics.

[0106] In another embodiment, a high-precision map incremental update device based on a roadside camera is also provided, comprising a processor and a memory storing a plurality of computer instructions. When the computer instructions are executed by the processor, the steps of the high-precision map incremental update method based on a roadside camera are implemented.

[0107] For the specific limitations of the high-precision map incremental update device based on roadside cameras, please refer to the limitations of the high-precision map incremental update method based on roadside cameras above, which will not be repeated here.

[0108] like Figure 7 As shown, in specific applications, a high-performance computing board can be used as a processor to execute the update method, and a high-speed data server can be used as a storage to design the data structure, build a data database, and index and store the map data. In addition, a computing hardware platform can be formed in conjunction with an IO kit, a GPU and other graphics cards, and a visible light high-resolution camera can be used as the sensor hardware to form a complete hardware platform.

[0109] The memory and processor of this embodiment are electrically connected, directly or indirectly, to enable data transmission or interaction. For example, these components may be electrically connected to each other via one or more communication buses or signal lines. The memory stores a computer program executable on the processor, and the processor implements the method of the embodiment of the present invention by executing the computer program stored in the memory.

[0110] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A high-precision map incremental update method based on roadside cameras, characterized in that: The method for incrementally updating a high-precision map based on a roadside camera includes: Acquire image data from multiple cameras for image stitching, and crop the stitched image to obtain a road image that only retains the road area; Compare the road images before and after the time series. If there is a new area, start the map update; otherwise, end it. After starting the map update, extract the road traffic map elements in the road image at the current time sequence, and construct the map element identification based on the road traffic map rules; Merging the road traffic map elements and map element identifiers with the latest map to complete the incremental update of the high-precision map; The step of extracting road traffic map elements from the road image at the current time sequence includes: Using image edge detection operators, linear features are extracted from road images. Linear features are visible linear targets on the road surface, including road edge lines, lane lines, zebra crossings, stop lines, lane markings, and parking area edge lines. Based on the hidden Markov chain model, point features are extracted from linear features, and the observation space, state space, state transition equation and output probability matrix are constructed to estimate the maximum likelihood road edge line combination, thereby determining the overall road surface width and lane line combination; Based on the overall road width and lane line combination, the lane line combination is topologically associated according to the length and distance attributes of the lane lines, and continuous lane lines are output; According to the linear elements in the extracted road image, lane lines and edge lines in the linear elements are removed, and block marks of the road surface are output, wherein the lane lines, edge lines and block marks serve as the extracted road traffic map elements; The method of extracting point elements from linear elements based on the hidden Markov chain model and constructing the observation space, state space, state transfer equation and output probability matrix includes: Observation space: The linear features in the road image are extracted as the observation space, and the data format is line segments; State space: extract discrete point features in the road area as state space, and the data format is point; The state transfer equation is constructed as follows: ； Where, From the state To status The probability of transfer, Refers to different states in the observation space, which are expressed as linear elements. For linear features With linear features distance; The output probability matrix is constructed as follows: ； Where, In state Lower observation value The output probability of For the state, i.e. the first state in the state space A state value, represented as a point feature, for The observed value at time, Indicates status With the observed value The distance between For the sum of status.

2. The method for incrementally updating a high-precision map based on a roadside camera according to claim 1, wherein: The arrangement rules of the cameras are as follows: Observation stations are set up at intervals along the road. Two cameras facing opposite directions are set up at each observation station at a preset height above the road. The main axes of the two cameras are consistent with the radial direction of the road, and the angle between them and the plane where the road is located is , providing image data by observation station when the map is updated.

3. The method for incrementally updating a high-precision map based on a roadside camera according to claim 1, wherein: Before acquiring image data from multiple cameras, the multiple cameras are jointly calibrated. The joint calibration includes: calibrating the multiple cameras based on the latest map to obtain camera distortion parameters.

4. The method for incrementally updating a high-precision map based on a roadside camera according to claim 1, wherein: The camera is a visible light high-resolution camera.

5. The method for incrementally updating a high-precision map based on a roadside camera according to claim 1, wherein: The step of acquiring image data from multiple cameras and performing image stitching includes: Based on the camera distortion parameters obtained by camera calibration, the image data is rectified to restore the spatial observation value of the image data; Remove moving target areas from image data through linear change detection; Based on corner detection, feature corners are extracted from image data to establish the topological features of the image; Extract lane lines, stop lines, and lane markings from the latest map, perform feature point generalization calculations, and construct a feature point matrix. The constructed feature point matrix is used as the main observation value, and the topological features of the established image are sequentially fused at multiple scales to complete the image stitching.

6. The method for incrementally updating a high-precision map based on a roadside camera according to claim 2, wherein: The fusion of road traffic map elements and map element identifiers with the latest map includes: The extracted and constructed road traffic map elements and map element identifiers are matched with the latest map. The maximum likelihood map element mapping relationship is constructed for the elements that can be matched. The elements that cannot be matched are regarded as abnormal elements and their features are expressed. In the local environment, the map elements within the current observation station are reconstructed based on the maximum likelihood mapping relationship of map elements and the characteristic expression of abnormal elements; in the global environment, the map elements in the global scope are updated based on the characteristic expression of abnormal elements.

7. The method for incrementally updating a high-precision map based on a roadside camera according to claim 1, wherein: The high-precision map incremental update method based on roadside cameras also includes performing a logical consistency check after the map is fused. The rules based on which the logical consistency check is based include the geographical attributes of map elements, digital road network topology associations, and map database construction rules.

8. A high-precision map incremental update device based on a roadside camera, comprising a processor and a memory storing a plurality of computer instructions, characterized in that: When the computer instructions are executed by the processor, the steps of the high-precision map incremental update method based on roadside cameras as described in any one of claims 1 to 7 are implemented.

Images