A vehicle road domain generation method and device

Abstract

The application discloses a kind of vehicle road area generation method and device, it is related to the field of automatic driving technology.The application is according to the direction aggregation of various road boundary elements in preset direction interval, generates road main curve;Along road main curve generates several sampling centers, and constructs normal section perpendicular to the local tangent of the road main curve at sampling center, and then determines local road area and corresponding boundary mark pair according to normal section, to concatenate local road area with the same boundary mark pair along the extension direction of the road main curve, generates the candidate vehicle road area of current frame;Finally, based on the motion information of ego-vehicle, the historical vehicle road area is associated with the candidate vehicle road area Matching, generates the stable vehicle road area set of current frame.The application can adapt to road trend change, fusion multi-source heterogeneous boundary elements, and stably output the passable vehicle road area with continuous identity mark.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method and apparatus for generating vehicle-road domains. Background Technology

[0002] Vehicle-mounted perception systems typically output heterogeneous line features such as lane lines, road edges, and curbs as discrete control points, polylines, or polynomial curves. The road topology module needs to convert these line features into continuous candidate passable areas for use by downstream modules such as lane generation, localization, planning, and control.

[0003] However, existing methods for generating road domains typically set a fixed longitudinal position along the vehicle's forward direction and use a cross-section parallel to the vehicle's transverse axis to intersect the boundary. In scenarios where the road direction changes in segments, such as curves and ramps, the sampling cross-section cannot follow the local road trend, easily leading to skew width errors and boundary misordering. Although there are road domain generation schemes that select a single most reliable boundary as the global reference line, this scheme is extremely sensitive to missed detections, occlusions, insufficient length, and label replacements of the reference line. Once the reference line changes, the geometric position of the subsequent sampling center and the overall shape of the passable area will change drastically. In addition, existing schemes often process lane lines and road edges separately, making it difficult to maintain a uniform topology when there are missing unilateral boundaries, mixed heterogeneous boundaries, and sparse far-end observations. If regions are generated independently for each frame, detection jitter can cause regions to repeatedly appear, disappear, or have their labels replaced, affecting the downstream modules' stable perception of the passable area. Summary of the Invention

[0004] This invention provides a method and apparatus for generating a vehicle-road domain, which, in an autonomous driving perception scenario, utilizes multi-source heterogeneous road boundary elements to stably, continuously, and accurately generate a passable vehicle-road domain that adapts to changes in road direction.

[0005] To address the aforementioned technical problems, this invention provides a method for generating a vehicle-road domain, comprising: Acquire several types of road boundary features in the current frame, and aggregate them in a preset direction range according to the various types of road boundary features to generate the main road curve; Several sampling centers are generated along the main curve of the road at a preset step size, and a normal section perpendicular to the local tangent of the main curve of the road is constructed at each of the sampling centers. The intersection points of each of the normal cross sections and various road boundary elements are determined respectively. An ordered sequence of intersection points is calculated based on each intersection point and each sampling center. Adjacent intersection points within the same normal cross section are determined as local road domains and corresponding boundary marker pairs. The boundary marker pairs are constructed based on the road boundary elements of the intersection points. By concatenating the boundary markers of the same local road domains along the extension direction of the main curve of the road, candidate vehicle road domains for the current frame are generated. The historical road domains are obtained, and the historical road domains are associated and matched with the candidate road domains based on the vehicle's motion information. The region identifier of the candidate road domains is determined according to the matching result, and a set of stable road domains for the current frame is generated.

[0006] This invention obtains and aggregates multiple types of road boundary elements to generate the main road curve, thus eliminating the reliance on a single boundary for sampling references. This avoids overall changes in the sampling reference caused by missed detections, occlusions, or label replacements of a single reference line, enhancing the robustness of the sampling reference. Furthermore, by generating sampling centers along the main road curve in step increments and constructing normal sections perpendicular to their local tangents, the sampling sections always follow the local road orientation. This avoids width measurement errors and boundary misordering caused by the oblique cutting of fixed transverse sections in curves, improving geometric adaptability in scenarios with changing road directions, such as curves and ramps. Simultaneously, it calculates intersections based on various road boundary elements and obtains an ordered sequence of intersection points based on signed distance. Adjacent intersection points in the sequence are paired into local road domains, and their boundary label pairs are recorded. This method unifies heterogeneous line elements from lane lines, road edges, and curbs into structured local regions with clear boundary identities, achieving a unified topological expression of multi-source boundaries. This allows local road regions with the same boundary identifiers to be concatenated along the main curve extension direction, naturally aggregating local road regions belonging to the same left and right boundary pairs into complete and continuous candidate road regions. This ensures the topological continuity and boundary identity consistency of candidate regions. Finally, historical road regions are associated and matched with candidate road regions to determine region identifiers, enabling stable transmission of road region identifiers between adjacent frames. This reduces the phenomenon of repeated appearance, disappearance, or replacement of identifiers of regions caused by single-frame detection jitter, providing downstream positioning, planning, and control modules with temporally stable passable area information.

[0007] Furthermore, the step of acquiring several types of road boundary features in the current frame and performing directional aggregation based on these road boundary features within a preset directional range to generate the main road curve includes: Acquire several types of road boundary elements in the current frame of the vehicle; the road boundary elements include boundary identifier, boundary type, control point sequence, and effective longitudinal range; Divide the road in front of the vehicle into one or more longitudinal intervals, calculate the sum of the line segment lengths of each type of road boundary element in each longitudinal interval, and use the sum of the line segment lengths as the effective length of each type of road boundary element. For each of the longitudinal intervals, the start and end point directions of various road boundary elements falling within the longitudinal interval are obtained, and the start and end point directions are vector-weighted averaged using the effective length of various road boundary elements within the longitudinal interval as weights to obtain the representative direction of the longitudinal interval. Based on the differences between the representative directions of each of the longitudinal intervals, the slope variation coefficient of the main curve of the road is fitted. The main road curve is generated based on the initial slope of the main road curve at the origin of the preset vehicle coordinate system and the slope variation coefficient.

[0008] This invention divides the road ahead of the vehicle into one or more longitudinal intervals and calculates the sum of the line segment lengths of each boundary within each interval as the effective length. This allows road boundaries at different longitudinal positions to contribute to direction estimation according to their actual coverage within each interval, avoiding the direction deviation caused by mixing distant sparse boundaries with near dense boundaries with equal weight. Subsequently, representative directions for each interval are obtained by vector weighting the start and end point directions of boundary elements falling into each longitudinal interval with the effective length as the weight. This gives longer, more fully covered boundaries higher weight in direction estimation, improving the accuracy and noise resistance of direction aggregation. Furthermore, the slope variation coefficient of the master curve is fitted based on the differences between the representative directions of each longitudinal interval, enabling the master curve to reflect the segmented change trend of the road direction in the longitudinal direction. This provides a geometric benchmark that adapts to the road trend for normal sampling in subsequent curve and transition segment scenarios. Finally, the master curve is generated based on the initial slope and slope variation coefficient at the origin of the preset vehicle coordinate system, giving the master curve a clear mathematical expression and geometric computability. This provides a stable and differentiable curve equation for subsequent sampling center generation and normal section construction.

[0009] Furthermore, the step of generating a plurality of sampling centers along the main curve of the road at preset step sizes, and constructing a normal section perpendicular to the local tangent of the main curve of the road at each of the sampling centers, includes: Using a preset position on the main curve of the road as the sampling starting point, several sampling centers are generated on the main curve of the road at a preset step size, and the sampling center coordinates of each sampling center in the vehicle coordinate system are determined. Calculate the derivative value of the main curve of the road at each of the sampling centers, determine the unit tangent vector at each of the sampling centers based on the derivative value, and construct the unit normal vector at each of the sampling centers based on the unit tangent vector; Starting from each of the sampling centers, and extending along the unit normal vector direction and its opposite direction of the sampling center based on a preset horizontal range, a normal cross section is constructed at each of the sampling centers.

[0010] This invention uses a preset position on the main curve as the sampling starting point and generates sampling centers sequentially according to a preset step size, determining the coordinates of each center in the vehicle coordinate system. This ensures that the sampling points are distributed with equal arc lengths along the road's extension direction, providing a discrete sampling benchmark with a defined position and uniform distribution for subsequent cross-section construction and boundary intersection calculations. Based on this, the derivative value of the main curve at each sampling center is calculated, and the unit tangent vector is determined according to the derivative value. Then, a unit normal vector is constructed based on the tangent vector, ensuring that the direction of the normal vector is precisely perpendicular to the local road orientation. This establishes a spatial reference direction for subsequent lateral distance calculations and intersection point sorting. Furthermore, using each sampling center as the starting point and extending along the normal vector direction and its opposite direction based on a preset lateral range, normal cross-sections are constructed. These cross-sections cover all possible boundary elements in the lateral direction of the road, providing a sampling plane with a defined spatial range for subsequent intersection calculations with various boundaries.

[0011] Furthermore, the step of determining the intersection points of each of the normal sections and various road boundary elements, calculating an ordered sequence of intersection points based on each intersection point and each sampling center, and determining adjacent intersection points within the same normal section as local road domains and corresponding boundary marker pairs includes: Traverse each of the normal cross sections, and sequentially perform intersection operations between the current normal cross section and each of the road boundary elements to obtain the initial intersection points between the current normal cross section and each of the road boundary elements; The initial intersection point is corrected to obtain the final intersection point between the current normal section and each of the road boundary elements; Calculate the signed lateral distance of each of the final intersection points relative to the corresponding sampling center. The signed lateral distance is the projection length of the line vector connecting the final intersection point and the sampling center in the direction of the unit normal vector of the normal section. Based on the signed lateral distance, all final intersection points on each normal section are sorted to generate an ordered sequence of intersection points for each normal section; In the ordered sequence of intersections of each of the normal sections, two adjacent final intersections are paired as a local road domain, and the boundary identifiers of the road boundary elements to which the two final intersections belong are used as the boundary identifier pairs of the local road domains.

[0012] This invention obtains initial intersection points by performing intersection operations between each normal cross section and various road boundary elements. Then, it obtains final intersection points by correcting these initial points. Next, it calculates the signed lateral distance of each final intersection point relative to the sampling center, quantifying the spatial position of the intersection points on the cross section into one-dimensional coordinates along the normal vector direction, providing a unified numerical metric for the spatial ordering of the intersection point sequence. Furthermore, based on the signed lateral distance, it sorts all final intersection points on each cross section to generate an ordered intersection point sequence, reorganizing the unordered intersection point set on the cross section according to a left-to-right spatial order, providing a definite order basis for the pairing of subsequent adjacent intersection points. Finally, in the ordered intersection point sequence, two adjacent final intersection points are paired into a local road domain, and their respective boundary identifiers are used as boundary identifier pairs. Two spatially adjacent boundary intersection points and the area between them are clearly defined as a candidate passable unit. Simultaneously, the boundary identifier pairs record the identities of the left and right boundaries of this unit, providing matching keys for connecting the same area across cross sections.

[0013] Furthermore, the step of correcting the initial intersection point to obtain the final intersection point between the current normal section and each of the road boundary elements includes: Each initial intersection point is checked for validity, and initial intersection points located outside the valid longitudinal range of the road boundary element or outside the preset lateral range of the normal section are removed, thus obtaining the valid intersection points of each normal section. When the same road boundary element has multiple valid intersection points with the same normal section, the valid intersection point closest to the sampling center is taken as the final intersection point of the road boundary element on the normal section.

[0014] This invention ensures that each intersection point participating in subsequent pairing and concatenation is within the actual observation range of the boundary and within the preset scanning width of the cross section by checking the validity of the initial intersection points and eliminating those located outside the effective longitudinal range of the boundary or outside the preset transverse range of the cross section. This avoids the generation of false road regions caused by extrapolation of the boundary far end or abnormal points outside the cross section range. At the same time, when there are multiple valid intersection points between the same road boundary element and the same normal cross section, the intersection point closest to the sampling center is selected as the final intersection point, so that each boundary contributes at most one representative intersection point at each cross section. This eliminates redundant intersection points caused by multiple intersections of spline curves and straight lines, ensures a one-to-one correspondence between intersection points on the cross section and the boundary, and clears up ambiguities for the accurate sorting of the ordered intersection point sequence and the correct pairing of adjacent intersection points.

[0015] Furthermore, the step of concatenating boundary identifiers for the same local road domains along the extension direction of the main road curve to generate candidate vehicle road domains for the current frame includes: The normal sections are arranged sequentially along the extension direction of the main curve of the road to obtain a sequence of normal sections; For any two adjacent normal sections in the normal section sequence, the boundary markers of the two adjacent normal sections are concatenated with the same local road domain to generate an initial candidate vehicle road domain set. The initial candidate vehicle road domain set is filtered for validity based on preset filtering conditions to obtain the candidate vehicle road domain for the current frame.

[0016] This invention obtains a cross-section sequence by arranging the normal cross-sections sequentially along the extension direction of the main curve. This establishes a definite spatial order relationship between the originally discrete cross-sections according to the road direction, providing a traversal direction and sequential logic for the longitudinal connection of local road domains. Furthermore, the local road domains with the same upper boundary markers of adjacent cross-sections in the cross-section sequence are connected, allowing local road domains belonging to the same left and right boundary pairs to naturally aggregate in the road extension direction, forming a complete and continuous lane area. This ensures the longitudinal consistency and topological continuity of the boundary identities of candidate vehicle road domains. In addition, the initial candidate vehicle road domain set is effectively screened based on preset filtering conditions, ensuring the quality and reliability of the output candidate regions and preventing downstream modules from receiving inexplicable abnormal road domains.

[0017] Furthermore, the acquisition of historical road domains involves associating and matching the historical road domains with the candidate road domains based on vehicle motion information, determining the region identifier of the candidate road domains according to the matching results, and generating a stable road domain set for the current frame, including: Obtain historical road domains, which include historical area identifiers and historical centerline information; Based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domain in the current frame, a correlation cost matrix between the historical vehicle road domain and the candidate vehicle road domain is constructed. The correlation cost matrix is ​​globally matched and solved to obtain a one-to-one matching relationship between each candidate road domain and each historical road domain. Based on the matching relationship, the region identifier of the candidate vehicle road domain is determined, and a stable vehicle road domain set for the current frame is generated.

[0018] This invention preserves traceable spatiotemporal context information for the inheritance of regional identities in the current frame by acquiring historical road domains containing historical region identifiers and historical centerline information. This provides a clear historical reference for cross-frame association of road domains. Subsequently, an association cost matrix is ​​constructed based on vehicle motion information, historical centerline information, and candidate road domain centerlines. The coordinate system changes caused by vehicle movement are incorporated into the pose compensation of historical regions, making historical regions and current candidate regions comparable under the same spatiotemporal reference. Simultaneously, geometric similarity and spatial overlap are quantified into calculable cost values, providing a quantitative basis for globally optimal matching. A one-to-one matching relationship is obtained by globally matching the association cost matrix. Compared to traditional pairwise greedy matching, which avoids matching conflicts caused by local optima, this invention ensures that the matching results between all candidate regions and historical regions are globally optimal. Finally, the regional identifier of the candidate road domain is determined based on the matching relationship. Successfully matched regions inherit historical identifiers without reallocation, while unmatched qualified regions receive new identifiers. This achieves unified identifier management for regions from different identity sources in the output set, ensuring that each region in the stable road domain set output in each frame has a clear and continuous identity.

[0019] Furthermore, the step of determining the region identifier of the candidate vehicle road domain based on the matching relationship and generating a stable vehicle road domain set for the current frame includes: If there is a one-to-one matching relationship between the candidate road domains, the regional identifier of the candidate road domain inherits the historical regional identifier of the historical road domain it matches, and a first road domain is generated. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains meet the preset area threshold, then a new area identifier is assigned to the candidate vehicle road domains to generate a second vehicle road domain. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains do not meet the preset area threshold, then the candidate vehicle road domains are eliminated. A stable set of vehicle road domains for the current frame is generated based on the first and second vehicle road domains.

[0020] This invention ensures that existing stable road domains maintain the same identity across multiple consecutive frames by inheriting the historical region identifier when a matching relationship exists between candidate road domains, thus preventing downstream modules from re-initializing their states due to frequent changes in region identifiers. By assigning a new region identifier to candidate road domains that do not have a matching relationship but meet a preset region threshold, newly emerging passable areas can be promptly included in the stable set and given independent identities, ensuring that the system's ability to perceive road topology changes is not suppressed. Furthermore, by eliminating candidate road domains that do not have a matching relationship and do not meet the preset region threshold, false regions caused by short-term noise or perception anomalies are prevented from being assigned identities and entering the stable set, ensuring that each region in the output set has sufficient time and spatial confidence. This results in a stable road domain set containing both historically established mature regions and newly discovered candidate regions, ensuring that downstream modules always obtain complete road domain information including all valid passable areas.

[0021] Furthermore, the construction of the association cost matrix between the historical vehicle road domain and the candidate vehicle road domain based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domain in the current frame includes: Based on the vehicle motion information, the historical centerline information of the historical road domain is used to predict the vehicle motion, and the predicted centerline of the historical road domain in the vehicle coordinate system of the current frame is obtained. The centerline distance and longitudinal overlap rate between the predicted centerline and the centerline of the candidate vehicle road domain are statistically analyzed. Based on the centerline distance and longitudinal range overlap rate, an association cost matrix is ​​constructed between the historical vehicle road domain and the candidate vehicle road domain.

[0022] This invention predicts the historical centerline of the historical road domain based on vehicle motion information to obtain the predicted centerline in the current frame coordinate system. It compensates the geometric position of the region at the previous moment into the vehicle coordinate system at the current moment, eliminating the relative displacement and rotation of the coordinate system caused by vehicle movement. This allows the historical region and the current candidate region to be geometrically compared under the same spatial reference. It also calculates the centerline distance and longitudinal range overlap rate between the predicted centerline and the candidate road domain centerline, characterizing the spatial association strength between the historical region and the candidate region from two dimensions: lateral deviation and longitudinal coverage overlap. This avoids mismatches or missed matches caused by relying on a single geometric indicator. Thus, it constructs an association cost matrix by using the centerline distance and longitudinal range overlap rate, integrating two different dimensions of spatial similarity measurement into a unified cost value, providing quantitative and sortable input data for subsequent global matching solutions.

[0023] In a second aspect, the present invention provides a road domain generation device, comprising: a main curve determination module, a normal section construction module, a local road domain determination module, a candidate road domain determination module, and a stable road domain determination module; The main curve determination module is used to acquire several types of road boundary elements in the current frame, and perform directional aggregation within a preset directional range based on the various types of road boundary elements to generate the main road curve. The normal section construction module is used to generate a number of sampling centers along the main curve of the road at a preset step size, and to construct a normal section perpendicular to the local tangent of the main curve of the road at each of the sampling centers. The local road domain determination module is used to determine the intersection points of each of the normal sections and various types of road boundary elements, calculate an ordered sequence of intersection points based on each intersection point and each sampling center, and determine adjacent intersection points within the same normal section as local road domains and corresponding boundary identifier pairs; the boundary identifier pairs are constructed based on the road boundary elements of the intersection points; The candidate road domain determination module is used to concatenate local road domains with the same boundary markers along the extension direction of the main road curve to generate candidate vehicle road domains for the current frame. The stable road domain determination module is used to acquire historical road domains, associate and match the historical road domains with the candidate road domains based on vehicle motion information, determine the region identifier of the candidate road domains according to the matching results, and generate a set of stable road domains for the current frame. Attached Figure Description

[0024] Figure 1 This is a schematic flowchart of a method for generating a vehicle-road domain according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a vehicle-road domain generation device provided in an embodiment of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0026] The terms "first" and "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] Example 1 See Figure 1 , Figure 1 This is a schematic flowchart illustrating a method for generating a vehicle-road domain according to an embodiment of the present invention. The embodiment of the present invention provides a method for generating a vehicle-road domain, including steps 101 to 105, as detailed below: Step 101: Obtain several types of road boundary features in the current frame, and perform directional aggregation within a preset directional range based on the various types of road boundary features to generate the main road curve; In this embodiment, the step of acquiring several types of road boundary features in the current frame and performing directional aggregation within a preset directional range based on these road boundary features to generate the main road curve includes: Acquire several types of road boundary elements in the current frame of the vehicle; the road boundary elements include boundary identifier, boundary type, control point sequence, and effective longitudinal range; Divide the road in front of the vehicle into one or more longitudinal intervals, calculate the sum of the line segment lengths of each type of road boundary element in each longitudinal interval, and use the sum of the line segment lengths as the effective length of each type of road boundary element. For each of the longitudinal intervals, the start and end point directions of various road boundary elements falling within the longitudinal interval are obtained, and the start and end point directions are vector-weighted averaged using the effective length of various road boundary elements within the longitudinal interval as weights to obtain the representative direction of the longitudinal interval. Based on the differences between the representative directions of each of the longitudinal intervals, the slope variation coefficient of the main curve of the road is fitted. The main road curve is generated based on the initial slope of the main road curve at the origin of the preset vehicle coordinate system and the slope variation coefficient.

[0029] In this embodiment, the vehicle's onboard perception system (such as camera, lidar, etc.) outputs multiple heterogeneous road boundary elements of the current frame in real time. These elements constitute the input data of this embodiment, namely, several types of road boundary elements.

[0030] In this embodiment, several types of road boundary features of the vehicle in the current frame are obtained. Each element in the set This refers to the aforementioned type of road boundary element. Each boundary... Includes boundary markers Boundary type, control point sequence And the effective longitudinal range. Among them, boundary markings. Used to uniquely identify the road boundary element; boundary type, used to indicate whether the boundary belongs to a lane line, road edge, or curb, etc.; control point sequence. ,in The total number of control points for this boundary; the effective longitudinal range, indicating the longitudinal intervals within which this boundary can be used for evaluation or intersection.

[0031] In an optional embodiment, each road boundary feature may also include information such as point-by-point confidence, occlusion status, and virtual / real attributes.

[0032] In one alternative embodiment, before generating the master curve, the input original road boundary features need to be filtered for validity to ensure the stability and accuracy of subsequent calculations. Specifically, the ordered control point sequence of each road boundary feature is checked, and different filtering strategies are executed according to its boundary type.

[0033] For example, for road boundary features of lane line type, if the number of control points is less than 4, the geometric information of the road boundary feature is insufficient, and the road boundary feature is skipped; for road boundary features of road edge or curb type, if the number of control points is less than 8, the boundary is skipped.

[0034] Furthermore, road boundary features with an overall length of less than 10 meters, or an effective chord length of less than 2 meters within the target longitudinal interval, will not be used for subsequent direction estimation to avoid interference from short or abnormal boundaries on the direction of the main curve. The road boundary features retained after validity filtering will be fitted into spline curves (e.g., FlushSplineLine) to enable accurate intersection calculations with any given cross-section.

[0035] In this embodiment, when generating the main road curve, the road in front of the vehicle is first divided into one or more longitudinal sections along the vehicle's forward direction (x-axis). In one specific embodiment, three-segment modes [-90,-5], [-5,5], [5,145], two-segment modes [-90,0], [0,145], and a single-segment mode [-90,145] are supported. The range of these intervals can be configured according to the origin position of the vehicle coordinate system and the sensing distance.

[0036] Then, the effective lengths of each type of road boundary element within each longitudinal interval are calculated. In one embodiment, for road boundary elements... In the vertical interval Effective length within Defined as the sum of the lengths of adjacent line segments within the interval from which the control point sequence of the boundary falls, specifically: (1) Among them, the summation symbol Only includes those that fall completely or partially within the vertical range All adjacent line segments within, and Road boundary elements The l-th and (l+1)-th control points in the control point sequence. The sum of the calculated line segment lengths. This refers to the effective length of the road boundary element within the corresponding longitudinal interval, reflecting the degree of coverage of the boundary within that interval.

[0037] In this embodiment, for each of the longitudinal intervals This involves obtaining the start and end points of various road boundary features falling within the longitudinal interval. For each valid road boundary feature... Take its start and end control points within the interval and calculate its direction vector. =( , ), and thus obtain the direction angle. =atan2( , ).

[0038] In one embodiment, the orientation angle is clipped according to the maximum slope threshold before entering the aggregation process to prevent extreme abnormal boundaries from dominating the aggregation result.

[0039] Subsequently, the effective length of various road boundary elements within the longitudinal interval is used. For weight A vector-weighted average is then applied to the directions of the start and end points to obtain the representative direction of the longitudinal interval. Representative direction angle. for: (2) in, This refers to the directional aggregation weight. In this embodiment, the directional aggregation weight is directly taken as the effective length, i.e. = In another alternative embodiment, to further improve the robustness of the orientation estimation, the weights... It can also incorporate more factors, and its expanded form is as follows: = .in, Road boundary elements In the interval The average confidence level within, This represents the reliability coefficient for boundary types (e.g., lane lines may have higher reliability than curbs). These extended weights are optional enhancements to further improve performance in complex scenarios.

[0040] In this embodiment, the slope variation coefficient of the main road curve is fitted according to the representative direction of each of the longitudinal intervals. The main road curve is expressed in quadratic polynomial form, and its mathematical form is: (3) Where x is the longitudinal coordinate in the vehicle coordinate system, and y(x) is the longitudinal position of the principal curve. x The horizontal coordinate at that location. This is the lateral intercept of the main curve of the road at x=0. Since this embodiment sets the sampling origin at the origin of the vehicle coordinate system, therefore... . It is the initial slope of the main curve of the road at x=0. It is the slope variation coefficient that describes the trend of slope change with longitudinal position x.

[0041] In a specific embodiment employing a bipartite principal curve, the representative slopes of the near-end interval I1 and the far-end interval I2 are respectively denoted as... and The corresponding effective lengths are respectively and Total effective length W = + The initial slope and slope variation coefficient Specifically: (4) (5) (6) (7) in, This is the curvature trend scaling factor, used to control the magnitude of the curvature of the main curve, and its value can be 0.05.

[0042] In an optional embodiment, if the total effective length W is less than a preset threshold, a default horizontal curve is returned (e.g., = =0), and subsequent processes will be handled according to the rollback strategy.

[0043] When a certain vertical interval (such as the far-end interval) does not have enough effective boundaries, the representative direction of the adjacent interval can be used for extrapolation, the effective principal curve coefficients of the previous frame can be used for prediction, or the weight of the interval can be directly reduced to ensure that the principal curve can still provide a reasonable sampling benchmark when some boundaries are missing.

[0044] In this embodiment, the road ahead of the vehicle is divided into one or more longitudinal intervals, and the sum of the line segment lengths of each boundary within each interval is calculated as the effective length. This allows road boundaries at different longitudinal positions to contribute to the direction estimation according to their actual coverage within each interval, avoiding the direction deviation caused by mixing distant sparse boundaries with near dense boundaries with equal weight. Subsequently, the representative direction of each interval is obtained by vector weighting the start and end point directions of the boundary elements falling into each longitudinal interval with the effective length as the weight. This gives longer and more fully covered boundaries a higher weight in the direction estimation, improving the accuracy and noise resistance of direction aggregation. Furthermore, the slope variation coefficient of the master curve is fitted based on the differences between the representative directions of each longitudinal interval, enabling the master curve to reflect the segmented change trend of the road direction in the longitudinal direction. This provides a geometric benchmark that adapts to the road trend for normal sampling in subsequent curve and transition segment scenarios. Finally, the master curve is generated based on the initial slope and slope variation coefficient at the origin of the preset vehicle coordinate system, giving the master curve a clear mathematical expression and geometric computability. This provides a stable and differentiable curve equation for subsequent sampling center generation and normal section construction.

[0045] In this embodiment, after obtaining the main curve of the road in the current frame, in order to suppress the inter-frame impact of perceived jitter on the geometry of the main curve and to perform cross-frame phase maintenance on the sampling center, it is necessary to perform temporal smoothing and prediction updates on the coefficients of the main curve of the road and the state of the sampling center.

[0046] In this embodiment, the initial slope of each piecewise principal curve is maintained. and slope variation coefficient A recursive smoother.

[0047] For the first frame, use the observations from the current frame directly. , The smoothing state is initialized; for subsequent frames, the prediction value from the previous frame is used to recursively smooth and update the current observation. The abstract mathematical expression of this smoothing process is: (8) in, Indicates the first t Frame time The observed value (i.e., the initial slope or slope change coefficient calculated in the current frame). This represents the prior estimate of the current frame obtained based on the smoothed state prediction of the previous frame. This is the final smoothed value of the current frame after recursive smoothing. This is the smoothing gain coefficient, which takes a value between 0 and 1 and is used to balance the confidence levels of observed and predicted values.

[0048] In an optional embodiment, the recursive update is performed by a Kalman filter (Kalman Float), and the filter's process noise and observation noise parameters are determined by engineering configuration; the above formula is only used to illustrate the basic principle of state fusion and does not constitute a limitation on the specific filter implementation.

[0049] Meanwhile, the sampling center state is maintained by the sampling center tracking module (e.g., TrackSamplingOrigin) to ensure the phase continuity of the sampling grid between adjacent frames. If the historical center state is empty (e.g., in the first frame), the sampling center is initialized to the origin (0,0) of the vehicle coordinate system; otherwise, motion prediction is performed on the historical sampling center based on the vehicle motion information, compensating it from the vehicle coordinate system of the previous frame to the vehicle coordinate system of the current frame. This prediction process is as follows: (9) in, This is the predicted position of the sampling center in the vehicle coordinate system of the current frame. Let v be the historical position of the sampling center in the vehicle coordinate system of the previous frame, and v be the vehicle speed. The time interval between two adjacent frames. This is the change in yaw angle of the vehicle within the time interval Δt (when the odometer time interval is valid, it can be calculated from the angular velocity). The rotation matrix, when applied to a vector, can achieve coordinate system rotation compensation. The physical meaning of this prediction process is: first, the historical sampling center is transformed from the vehicle coordinate system of the previous frame to the world coordinate system, and then transformed to the vehicle coordinate system of the current frame, thereby eliminating the coordinate changes caused by the vehicle's motion.

[0050] After completing motion prediction, when the latest sampling center has moved to the vicinity of the starting end of the main curve, the system advances the sampling center position along the target main curve at sampling intervals Δs and deletes expired historical center states, thereby maintaining the phase continuity of the sampling grid of adjacent frames.

[0051] In an optional embodiment, for devices without vehicle motion information input (such as those lacking an IMU or wheel speed odometer), a static origin mode can be used, i.e., motion prediction is not performed on the sampling center, or only a single-frame geometry mode can be executed, i.e., the master curve and sampling grid are generated independently for each frame without relying on historical states. After the above smoothing and prediction processing, the system passes the updated master curve smoothing coefficient and sampling center state to the subsequent normal section construction step, providing a temporally consistent geometric reference for stable sampling across frames.

[0052] Step 102: Generate several sampling centers along the main curve of the road at a preset step size, and construct a normal section perpendicular to the local tangent of the main curve of the road at each sampling center; In this embodiment, generating a plurality of sampling centers along the main curve of the road at preset step sizes, and constructing a normal section perpendicular to the local tangent of the main curve of the road at each sampling center, includes: Using a preset position on the main curve of the road as the sampling starting point, several sampling centers are generated on the main curve of the road at a preset step size, and the sampling center coordinates of each sampling center in the vehicle coordinate system are determined. Calculate the derivative value of the main curve of the road at each of the sampling centers, determine the unit tangent vector at each of the sampling centers based on the derivative value, and construct the unit normal vector at each of the sampling centers based on the unit tangent vector; Starting from each of the sampling centers, and extending along the unit normal vector direction and its opposite direction of the sampling center based on a preset horizontal range, a normal cross section is constructed at each of the sampling centers.

[0053] In this embodiment, after obtaining a continuous and smooth main road curve y(x), in order to discretely generate a series of sampling points on it and construct a normal cross section for each point to intersect with the road boundary, this embodiment takes a preset position on the main road curve as the sampling starting point, generates several sampling centers on the main road curve along the main road curve at a preset step size, and determines the sampling center coordinates of each sampling center in the vehicle coordinate system.

[0054] In one embodiment, the main road curve is parameterized as follows: ,in This represents the arc length along the main curve. Starting from the sampling origin, the arc is calculated according to a preset step size. Sampling centers are generated sequentially. : (10) in, Where N is the starting arc length position, and N is the total number of sampling centers. The preset step size (arc length) (i.e., FLAGS_sampling_interval) can be determined by the project configuration; the default value is 2 meters. Sampling Center The coordinates are .

[0055] In some alternative embodiments, the sampling step size can be adaptive. For example, in regions with high curvature of the main curve or high road boundary density, the sampling step size can be reduced to increase sampling resolution; in regions with low curvature or sparse boundaries, the sampling step size can be increased to reduce computational load.

[0056] Then, the derivative values ​​of the main curve of the road at each of the sampling centers are calculated respectively, the unit tangent vector at each of the sampling centers is determined according to the derivative values, and the unit normal vector at each of the sampling centers is constructed based on the unit tangent vector.

[0057] For sampling center Its derivative on the principal curve y(x) is Then the unit tangent vector at the sampling center and unit normal vector for: (11) (12) in, The direction is consistent with the local trend of the main curve. yes The unit normal vector obtained by rotating 90 degrees points in the lateral direction of the road.

[0058] Next, taking each of the sampling centers as the starting point, and extending along the unit normal vector direction and its opposite direction of the sampling center based on a preset horizontal range, a normal cross section at each of the sampling centers is constructed.

[0059] In this embodiment, the normal section Constructed as a line passing through the sampling center Along the normal vector Finite line segments extending in a direction: (13) Where λ is a one-dimensional coordinate along the direction of the normal vector. The preset lateral range (maximum half-width) should be determined so that it covers the maximum lateral range of the road that the vehicle may travel on (e.g., set to 20 meters or more based on road design specifications and perception capabilities). In this embodiment, when the main curve direction is close to vertical, the parametric formula of the above-mentioned normal section remains valid, avoiding the division-to-zero problem that may occur when using the inverse of the slope.

[0060] In this embodiment, by using a preset position on the main curve as the sampling starting point and generating sampling centers sequentially according to a preset step size, and determining the coordinates of each center in the vehicle coordinate system, the sampling points are distributed with equal arc lengths along the road extension direction, providing a discrete sampling reference with a fixed position and uniform distribution for subsequent cross-section construction and boundary intersection calculation. Based on this, the derivative value of the main curve at each sampling center is calculated, and the unit tangent vector is determined according to the derivative value. Then, a unit normal vector is constructed based on the tangent vector, ensuring that the direction of the normal vector is precisely perpendicular to the local road orientation. This establishes a spatial reference direction for subsequent lateral distance calculation and intersection point sorting. Furthermore, using each sampling center as the starting point and extending along the normal vector direction and its opposite direction based on a preset lateral range, a normal cross-section is constructed, ensuring that the cross-section covers all possible boundary elements in the lateral direction of the road. This provides a sampling plane with a fixed spatial range for subsequent intersection calculations with various boundaries.

[0061] Step 103: Determine the intersection points of each of the normal sections and the various types of road boundary elements respectively; calculate an ordered sequence of intersection points based on each intersection point and each sampling center; and determine adjacent intersection points within the same normal section as local road domains and corresponding boundary marker pairs; the boundary marker pairs are constructed based on the road boundary elements of the intersection points; In this embodiment, the step of determining the intersection points of each of the normal sections and various road boundary elements, calculating an ordered sequence of intersection points based on each intersection point and each sampling center, and determining adjacent intersection points within the same normal section as local road domains and corresponding boundary marker pairs includes: Traverse each of the normal cross sections, and sequentially perform intersection operations between the current normal cross section and each of the road boundary elements to obtain the initial intersection points between the current normal cross section and each of the road boundary elements; The initial intersection point is corrected to obtain the final intersection point between the current normal section and each of the road boundary elements; Calculate the signed lateral distance of each of the final intersection points relative to the corresponding sampling center. The signed lateral distance is the projection length of the line vector connecting the final intersection point and the sampling center in the direction of the unit normal vector of the normal section. Based on the signed lateral distance, all final intersection points on each normal section are sorted to generate an ordered sequence of intersection points for each normal section; In the ordered sequence of intersections of each of the normal sections, two adjacent final intersections are paired as a local road domain, and the boundary identifiers of the road boundary elements to which the two final intersections belong are used as the boundary identifier pairs of the local road domains.

[0062] In this embodiment, after all normal sections are constructed, the intersection of the sections and the boundary curves is calculated, and the spatial intersections are organized into a structured local road domain.

[0063] In this embodiment, each of the normal cross sections is traversed, and the intersection operation between the current normal cross section and each of the road boundary elements is performed sequentially to obtain the initial intersection points between the current normal cross section and each of the road boundary elements. The intersection operation can be implemented using various methods such as analytical root finding, iterative root finding, piecewise linear interpolation, or distance field-based methods.

[0064] In this embodiment, since the road boundary features have been prepared as cubic spline curves, the mathematical intersection of the cross-section line and the cubic spline can be directly performed. After obtaining the initial intersection point, the initial intersection point is corrected to obtain the final intersection point between the current normal cross-section and each of the road boundary features.

[0065] In this embodiment, the step of correcting the initial intersection point to obtain the final intersection point between the current normal section and each of the road boundary elements includes: Each initial intersection point is checked for validity, and initial intersection points located outside the valid longitudinal range of the road boundary element or outside the preset lateral range of the normal section are removed, thus obtaining the valid intersection points of each normal section. When the same road boundary element has multiple valid intersection points with the same normal section, the valid intersection point closest to the sampling center is taken as the final intersection point of the road boundary element on the normal section.

[0066] In this embodiment, firstly, the validity of each initial intersection point is checked. Initial intersection points located outside the valid longitudinal range of the road boundary feature, or outside the preset lateral range (i.e., [-Wmax, Wmax]) of the normal section, and intersection points with sentinel values ​​(indicating invalid special values) are eliminated, thereby obtaining the valid intersection points of each normal section. Then, when the same road boundary feature has multiple valid intersection points with the same normal section, for example, a spline may intersect a straight line twice at a curve, the valid intersection point closest to the sampling center is taken as the final intersection point of the road boundary feature on that normal section. This ensures that each road boundary feature contributes at most one intersection point at each section, eliminating ambiguity caused by multiple solutions.

[0067] In another alternative embodiment, approximately parallel numerical outliers can also be eliminated based on the local directional angle between the cross section and the boundary.

[0068] In this embodiment, after obtaining the final intersection points of each normal section, the signed lateral distance of each final intersection point relative to the corresponding sampling center is calculated. For the valid intersection point of the k-th normal section and the j-th road boundary element... Its signed horizontal distance Defined as: (14) in, The sampling center of this normal section, This is the unit normal vector for this cross section. The symbol represents the lateral distance. The geometric meaning of is the intersection point. to sampling center The connecting vectors are in the direction of the normal vector. The projection length on the surface indicates whether the intersection point is located to the left or right of the normal vector.

[0069] In this embodiment, based on the signed lateral distance Sort all final intersection points on each of the aforementioned normal sections to generate an ordered sequence of intersection points for each normal section. Specifically, for the k-th normal section, sort all its final intersection points according to... Arrange the points in ascending order (i.e., from left to right) to obtain an ordered sequence of intersection points. ,in This represents the number of valid intersection points on the normal section k. Each intersection point... After sorting, the boundary identifiers of the source road boundary elements are retained. And boundary type and other information. If the number of valid intersection points of a certain section | If |<2, then the cross section cannot form any local road domain, does not generate empty left and right boundaries, and does not forge any intersection information.

[0070] In this embodiment, in the ordered intersection sequence of each normal section, two adjacent final intersections are paired as a local road domain, and the boundary identifiers of the road boundary elements to which the two final intersections belong are used as the boundary identifier pairs of the local road domains. Specifically, in the ordered intersection sequence... In the middle, the adjacent intersection points and Each pair is considered a candidate local road domain. The ordered boundary identifiers of this local road domain are defined as follows: (15) in, Indicates the intersection point Boundary markers for the road boundary elements. Indicates the intersection point Boundary identifiers for the road boundary elements.

[0071] In this embodiment, candidate local road domain pairings also need to undergo access control checks to ensure their validity, including: Check the distance between the two intersections, i.e. whether the width of the local road area is within the preset minimum and maximum width range; Check whether the effective longitudinal extents of the two boundary features overlap at the current sampling position; Check whether the combination of boundary types allows the formation of candidate passable zones, such as lane line-lane line, lane line-curb, etc. Check if the intersection point itself is a valid value (not a sentinel); Any pairing that does not meet the above access control conditions will be discarded.

[0072] In this embodiment, by checking the validity of the initial intersection points and removing those located outside the effective longitudinal range of the boundary or outside the preset transverse range of the cross section, it is ensured that each intersection point participating in subsequent pairing and concatenation is within the actual observation range of the boundary and within the preset scanning width of the cross section. This avoids the generation of false road regions caused by extrapolation of the boundary far end or abnormal points outside the cross section range. At the same time, when there are multiple valid intersection points between the same road boundary element and the same normal cross section, the intersection point closest to the sampling center is selected as the final intersection point. This ensures that each boundary contributes at most one representative intersection point at each cross section, eliminating redundant intersection points caused by multiple intersections of spline curves and straight lines. This ensures a one-to-one correspondence between intersection points on the cross section and the boundary, clearing up ambiguities for the accurate sorting of the ordered intersection point sequence and the correct pairing of adjacent intersection points.

[0073] In this embodiment, initial intersection points are obtained by performing intersection operations between each normal cross section and various road boundary elements. Then, final intersection points are obtained by correcting the initial intersection points. Next, the signed lateral distance of each final intersection point relative to the sampling center is calculated, quantifying the spatial position of the intersection point on the cross section into a one-dimensional coordinate along the normal vector direction, providing a unified numerical metric for the spatial sorting of the intersection point sequence. Furthermore, based on the signed lateral distance, all final intersection points on each cross section are sorted to generate an ordered intersection point sequence. The unordered set of intersection points on the cross section is reorganized according to a left-to-right spatial order, providing a definite order basis for the pairing of subsequent adjacent intersection points. Finally, in the ordered intersection point sequence, two adjacent final intersection points are paired into a local road domain, and their respective boundary identifiers are used as boundary identifier pairs. Two spatially adjacent boundary intersection points and the area between them are clearly defined as a candidate passable unit. Simultaneously, the identity of the left and right boundaries of this unit is recorded through the boundary identifier pairs, providing matching keys for connecting the same area across cross sections.

[0074] Step 104: Connect the boundary markers of the same local road domains along the extension direction of the main curve of the road to generate the candidate vehicle road domains of the current frame; In this embodiment, the step of concatenating boundary markers for the same local road domains along the extension direction of the main road curve to generate candidate vehicle road domains for the current frame includes: The normal sections are arranged sequentially along the extension direction of the main curve of the road to obtain a sequence of normal sections; For any two adjacent normal sections in the normal section sequence, the boundary markers of the two adjacent normal sections are concatenated with the same local road domain to generate an initial candidate vehicle road domain set. The initial candidate vehicle road domain set is filtered for validity based on preset filtering conditions to obtain the candidate vehicle road domain for the current frame.

[0075] In this embodiment, firstly, the normal cross-sections are arranged sequentially along the extension direction of the main curve of the road (i.e., the direction in which the sampling number k increases) to obtain a normal cross-section sequence. Then, for any two adjacent normal cross-sections in the normal cross-section sequence, the boundary markers of the two adjacent normal cross-sections are concatenated to the same local road domain to generate an initial candidate vehicle road domain set.

[0076] In one specific embodiment, a historical local lane list (history_lanes) and a current cross-section local lane list (now_lanes) are maintained. When a pair of lanes with the same ordered boundary identifiers is detected on adjacent cross-sections... When a local road domain is generated in the current section, the local road domain information of the next section is appended to the candidate road domain generated in the previous section. If a specific boundary marker disappears in the current section, or the width exceeds the limit, the extension of the current candidate road domain ends, and the process proceeds to the finalization and post-processing.

[0077] In one optional embodiment, when a boundary is temporarily missing on one side, or when the boundary ID undergoes an interpretable switch (such as a change in the identifier of a physical boundary or geometric continuity of adjacent center points), the system can retain the region state and interpolate the gap within the thresholds for changes in the gap length and width. This achieves a certain degree of fault tolerance and avoids candidate region breakage due to short-term perception loss. If the gap exceeds a preset length, it is not forcibly extended, and the final result is maintained.

[0078] In this embodiment, after obtaining the initial candidate road area set, the initial candidate road area set is filtered for effectiveness based on preset filtering conditions to obtain the candidate road areas for the current frame. The post-processing process includes: deleting regions with fewer than a preset threshold (e.g., min_roadarea_sample_times) to eliminate excessively short noise regions; filtering regions with excessively short lengths or average widths exceeding the normal road width range; merging longitudinally adjacent segments with the same boundary marker pairs; and unifying the direction of the left and right endpoints to ensure geometric consistency of the regions. After processing, each candidate road area contains complete closed-loop fields such as left and right boundary point sequences, cross-section message sequences, center point sequences, boundary marker pairs, effective longitudinal range, and enable status.

[0079] In this embodiment, a cross-section sequence is obtained by arranging the normal cross-sections sequentially along the extension direction of the main curve. This establishes a definite spatial order relationship between the originally discrete cross-sections according to the road direction, providing a traversal direction and sequential logic for the longitudinal connection of local road domains. Furthermore, local road domains with the same upper boundary markers on adjacent cross-sections in the cross-section sequence are connected, allowing local road domains belonging to the same left and right boundary pairs to naturally aggregate in the road extension direction, forming a complete and continuous lane area. This ensures the longitudinal consistency and topological continuity of the boundary identities of candidate vehicle road domains. In addition, the initial candidate vehicle road domain set is effectively screened based on preset filtering conditions, ensuring the quality and reliability of the output candidate regions and preventing downstream modules from receiving inexplicable abnormal road domains.

[0080] Step 105: Obtain historical road domains, associate and match the historical road domains with the candidate road domains based on the vehicle motion information, determine the region identifier of the candidate road domains according to the matching results, and generate a set of stable road domains for the current frame.

[0081] In this embodiment, the step of acquiring historical road domains involves associating and matching the historical road domains with the candidate road domains based on vehicle motion information, determining the region identifier of the candidate road domains according to the matching results, and generating a stable road domain set for the current frame, including: Obtain historical road domains, which include historical area identifiers and historical centerline information; Based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domain in the current frame, a correlation cost matrix between the historical vehicle road domain and the candidate vehicle road domain is constructed. The correlation cost matrix is ​​globally matched and solved to obtain a one-to-one matching relationship between each candidate road domain and each historical road domain. Based on the matching relationship, the region identifier of the candidate vehicle road domain is determined, and a stable vehicle road domain set for the current frame is generated.

[0082] In this embodiment, by cross-frame tracking, stable and continuous region identifiers are assigned to the aforementioned single-frame candidate vehicle road domains, thereby suppressing region identity jumps caused by single-frame detection jitter. First, historical vehicle road domains are acquired. The historical vehicle road domains are a set of stable vehicle road domains maintained by the system from the previous frame or earlier, which includes historical region identifiers (track_id), historical centerline information, and continuous unmatched counts (miss_cnt) for tracking. Then, based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domains in the current frame, an association cost matrix between historical vehicle road domains and candidate vehicle road domains is constructed.

[0083] In this embodiment, the step of constructing the association cost matrix between the historical vehicle road domain and the candidate vehicle road domain based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domain in the current frame includes: Based on the vehicle motion information, the historical centerline information of the historical road domain is used to predict the vehicle motion, and the predicted centerline of the historical road domain in the vehicle coordinate system of the current frame is obtained. The centerline distance and longitudinal overlap rate between the predicted centerline and the centerline of the candidate vehicle road domain are statistically analyzed. Based on the centerline distance and longitudinal range overlap rate, an association cost matrix is ​​constructed between the historical vehicle road domain and the candidate vehicle road domain.

[0084] In this embodiment, before constructing the cost matrix, vehicle motion prediction is first performed on the historical centerline information of the historical road domain based on the vehicle motion information to obtain the predicted centerline of the historical road domain in the vehicle coordinate system of the current frame. If the historical centerline state is empty, it is initialized to the origin (0,0); otherwise, it is determined based on the vehicle speed v and yaw change. And the time interval Δt, the historical centerline control point Predicted in the current frame coordinate system: (16) in, This represents the change in yaw angle of the vehicle within the time interval Δt. As the corresponding rotation matrix, this prediction process compensates the historical region's position to the current vehicle coordinate system, eliminating coordinate changes caused by the vehicle's motion.

[0085] In this embodiment, the centerline distance and longitudinal range overlap rate between the predicted centerline and the centerline of the candidate vehicle road domain are statistically analyzed. For matchable pairs, matching is only allowed if their longitudinal (x-direction) ranges intersect; otherwise, matching is prohibited. The centerline distance can be calculated using the average nearest distance method, which involves finding the distance between the nearest point on the current candidate centerline and multiple points on the historical predicted centerline, and then averaging the distances. Subsequently, an association cost matrix between the historical vehicle road domain and the candidate vehicle road domain is constructed based on the centerline distance and the longitudinal range overlap rate.

[0086] In this embodiment, the association cost The nearest distance is directly averaged using the centerline. The primary constraint is a threshold constraint. (17) in, It is the set of predicted centerline points for the i-th historical road domain. It is the set of centerline points of the j-th current candidate road domain. During the calculation, only candidate point pairs with longitudinal distance difference |Δx|≤5m and lateral distance difference |Δy|≤1m are calculated. If the number of matching points is less than 3, a preset large cost indicating that the match is not possible is returned, for example, 1000.

[0087] In another alternative embodiment, to make the association more robust, a joint cost function incorporating multi-dimensional information can be used: (18) in, and These represent the overlap rates in the longitudinal and lateral directions, respectively. and These are the normalized width and length differences. Penalty for inconsistent left and right boundary markings. , , , , , All are non-negative weighting coefficients used to balance the importance of various indicators. Any pair of objects whose individual cost or total cost exceeds a preset threshold is set as unmatchable.

[0088] In this embodiment, a complete cost matrix is ​​constructed based on the association cost between each historical road domain and each candidate road domain. The association cost matrix is ​​then globally matched and solved to obtain a one-to-one matching relationship between each candidate road domain and each historical road domain.

[0089] In this embodiment, the historical centerline of the historical road domain is predicted based on the vehicle's motion information to obtain the predicted centerline in the current frame coordinate system. The geometric position of the region at the previous moment is compensated into the vehicle coordinate system at the current moment, eliminating the relative displacement and rotation of the coordinate system caused by the vehicle's movement. This allows the historical region and the current candidate region to be geometrically compared under the same spatial reference. The centerline distance and longitudinal range overlap rate between the predicted centerline and the candidate road domain centerline are statistically analyzed. The spatial association strength between the historical region and the candidate region is characterized from two dimensions: the degree of lateral deviation and the degree of longitudinal coverage overlap. This avoids mismatches or missed matches caused by relying on only a single geometric indicator. Thus, an association cost matrix is ​​constructed by the centerline distance and the longitudinal range overlap rate, which integrates two different dimensions of spatial similarity measurement into a unified cost value, providing quantitative and sortable input data for subsequent global matching solutions.

[0090] In this embodiment, the Kuhn-Munkres (KM) algorithm is used to solve the problem. This algorithm can find the globally optimal one-to-one match on the cost matrix, avoiding the local optimal conflicts that may occur with greedy algorithms.

[0091] In this embodiment, by acquiring historical road domains containing historical region identifiers and historical centerline information, traceable spatiotemporal context information is preserved for the regional identity inheritance of the current frame, providing a clear historical reference for cross-frame association of road domains. Subsequently, an association cost matrix is ​​constructed based on vehicle motion information, historical centerline information, and candidate road domain centerlines. The coordinate system changes caused by vehicle movement are incorporated into the pose compensation of historical regions, making historical regions and current candidate regions comparable under the same spatiotemporal reference. At the same time, geometric similarity and spatial overlap are quantified into calculable cost values, providing a quantitative basis for global optimal matching. One-to-one matching relationships are obtained by globally matching the association cost matrix. Compared with traditional pairwise greedy matching, this embodiment avoids matching conflicts caused by local optima, ensuring that the matching results between all candidate regions and historical regions are optimal in a global sense. Finally, the regional identifiers of candidate vehicle road domains are determined based on the matching relationship. Regions that successfully match inherit historical identifiers without needing to be reassigned, while unmatched qualified regions obtain new identifiers. This achieves unified identifier management for regions from different identity sources in the output set, ensuring that each region in the stable vehicle road domain set output in each frame has a clear and continuous identity.

[0092] In this embodiment, determining the region identifier of the candidate vehicle road domain based on the matching relationship and generating a stable vehicle road domain set for the current frame includes: If there is a one-to-one matching relationship between the candidate road domains, the regional identifier of the candidate road domain inherits the historical regional identifier of the historical road domain it matches, and a first road domain is generated. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains meet the preset area threshold, then a new area identifier is assigned to the candidate vehicle road domains to generate a second vehicle road domain. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains do not meet the preset area threshold, then the candidate vehicle road domains are eliminated. A stable set of vehicle road domains for the current frame is generated based on the first and second vehicle road domains.

[0093] In this embodiment, the region identifier of the candidate vehicle road domain is determined based on the matching relationship, and a stable vehicle road domain set for the current frame is generated: If there is a one-to-one matching relationship between the candidate road domains, the regional identifier of the candidate road domain inherits the historical regional identifier of the historical road domain it matches, and the unmatched count miss_cnt of the historical region is cleared to zero, thus generating the first road domain.

[0094] If there is no one-to-one matching relationship between the candidate vehicle road domains, and the candidate vehicle road domains meet the preset area thresholds (such as length, width, number of cross sections, and other tracking thresholds), then a new globally incrementing area identifier is assigned to the candidate vehicle road domains to generate a second vehicle road domain.

[0095] If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domain does not meet the preset area threshold, then the candidate vehicle road domain is removed and no identifier is assigned to it.

[0096] In this embodiment, for a historical road domain, if it fails to match in the current frame, its miss_cnt is incremented by 1, and it continues to be retained for a certain number of frames (for example, in the current implementation, if miss_cnt>1, it is deleted, that is, one frame is retained); if the number of consecutive unmatched times exceeds the preset retention frame threshold, the historical road domain is deleted from the historical road domain set.

[0097] Finally, based on the first and second road domains, a stable set of road domains for the current frame is generated. Each road domain in this set has a stable, frame-continuous region identifier (track_id), which can be directly used by downstream modules such as Lane generation, vehicle localization, path planning, or vehicle control. The output of the entire method is a stable set of road domains for the current frame, rather than the original boundary features or intermediate intersection sets, which provides a clear, stable, and structured interface for downstream modules.

[0098] In this embodiment, when a candidate road domain has a matching relationship, its region identifier inherits the historical region identifier of the matched region. This ensures that existing stable road domains maintain the same identity across multiple consecutive frames, preventing downstream modules from re-initializing their states due to frequent changes in region identifiers. When a candidate road domain has no matching relationship but meets a preset region threshold, a new region identifier is assigned to it. This ensures that newly emerging passable areas are promptly included in the stable set and given independent identities, guaranteeing that the system's ability to perceive changes in road topology is not suppressed. When a candidate road domain has no matching relationship and does not meet the preset region threshold, it is eliminated. This prevents false regions caused by short-term noise or perception anomalies from being assigned identities and entering the stable set, ensuring that each region in the output set has sufficient time and spatial confidence. This results in a stable road domain set containing both historically established mature regions and newly discovered candidate regions, ensuring that downstream modules always obtain complete road domain information including all valid passable areas.

[0099] Please refer to Figure 2 , Figure 2 A schematic diagram of a road domain generation device provided in an embodiment of the present invention includes: a main curve determination module 201, a normal section construction module 202, a local road domain determination module 203, a candidate road domain determination module 204, and a stable road domain determination module 205; The main curve determination module 201 is used to obtain several types of road boundary elements in the current frame, and perform directional aggregation within a preset directional range based on the various types of road boundary elements to generate the main road curve. The normal section construction module 202 is used to generate a number of sampling centers along the main curve of the road at a preset step size, and to construct a normal section perpendicular to the local tangent of the main curve of the road at each of the sampling centers. The local road domain determination module 203 is used to determine the intersection points of each of the normal sections and various types of road boundary elements, calculate an ordered sequence of intersection points based on each intersection point and each sampling center, and determine adjacent intersection points within the same normal section as local road domains and corresponding boundary identifier pairs; the boundary identifier pairs are constructed based on the road boundary elements of the intersection points; The candidate road domain determination module 204 is used to concatenate local road domains with the same boundary markers along the extension direction of the main curve of the road to generate candidate vehicle road domains for the current frame. The stable road domain determination module 205 is used to acquire historical road domains, associate and match the historical road domains with the candidate road domains based on the vehicle's motion information, determine the region identifier of the candidate road domains according to the matching result, and generate a set of stable road domains for the current frame.

[0100] This invention also provides a vehicle, which generates a vehicle-road domain based on the above-described vehicle-road domain generation method.

[0101] This invention also provides an autonomous driving device, which generates a vehicle-road domain based on the above-described vehicle-road domain generation method and realizes autonomous driving based on the generated vehicle-road domain.

[0102] In this embodiment of the invention, a processing device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described vehicle-road domain generation method.

[0103] In this embodiment of the invention, a computer-readable storage medium is also provided, which includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the above-described vehicle-road domain generation method when it is running.

[0104] For example, a computer program can be divided into one or more modules, one or more of which are stored in memory and executed by a processor to perform the present invention. The one or more modules can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a processing device.

[0105] The processing device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The processing device may include, but is not limited to, a processor, memory, and a display. Those skilled in the art will understand that the above components are merely examples of the processing device and do not constitute a limitation on the processing device. It may include more or fewer components than the specified components, or a combination of certain components, or different components. For example, the processing device may also include input / output devices, network access devices, buses, etc.

[0106] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the processing device, connecting all parts of the processing device through various interfaces and lines.

[0107] Memory can be used to store computer programs and / or modules. The processor performs various functions of the processing device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function (such as sound playback, text conversion, etc.), etc.; the data storage area can store data created based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0108] In this invention, modules based on the generation of vehicle-road domains, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. Those skilled in the art can understand and implement this invention without any inventive effort.

[0109] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for generating a vehicle-road domain, characterized in that, include: Acquire several types of road boundary features in the current frame, and aggregate them in a preset direction range according to the various types of road boundary features to generate the main road curve; Several sampling centers are generated along the main curve of the road at a preset step size, and a normal section perpendicular to the local tangent of the main curve of the road is constructed at each of the sampling centers. The intersection points of each of the normal cross sections and various road boundary elements are determined respectively. An ordered sequence of intersection points is calculated based on each intersection point and each sampling center. Adjacent intersection points within the same normal cross section are determined as local road domains and corresponding boundary marker pairs. The boundary marker pairs are constructed based on the road boundary elements of the intersection points. By concatenating the boundary markers of the same local road domains along the extension direction of the main curve of the road, candidate vehicle road domains for the current frame are generated. The historical road domains are obtained, and the historical road domains are associated and matched with the candidate road domains based on the vehicle's motion information. The region identifier of the candidate road domains is determined according to the matching result, and a set of stable road domains for the current frame is generated.

2. The method for generating a vehicle-road domain as described in claim 1, characterized in that, The step of acquiring several types of road boundary features in the current frame and performing directional aggregation based on these road boundary features within a preset directional range to generate the main road curve includes: Acquire several types of road boundary elements in the current frame of the vehicle; the road boundary elements include boundary identifier, boundary type, control point sequence, and effective longitudinal range; Divide the road in front of the vehicle into one or more longitudinal intervals, calculate the sum of the line segment lengths of each type of road boundary element in each longitudinal interval, and use the sum of the line segment lengths as the effective length of each type of road boundary element. For each of the longitudinal intervals, the start and end point directions of various road boundary elements falling within the longitudinal interval are obtained, and the start and end point directions are vector-weighted averaged using the effective length of various road boundary elements within the longitudinal interval as weights to obtain the representative direction of the longitudinal interval. Based on the differences between the representative directions of each of the longitudinal intervals, the slope variation coefficient of the main curve of the road is fitted. The main road curve is generated based on the initial slope of the main road curve at the origin of the preset vehicle coordinate system and the slope variation coefficient.

3. The method for generating a vehicle-road domain as described in claim 2, characterized in that, The step of generating a plurality of sampling centers along the main curve of the road at a preset step size, and constructing a normal section perpendicular to the local tangent of the main curve of the road at each of the sampling centers, includes: Using a preset position on the main curve of the road as the sampling starting point, several sampling centers are generated on the main curve of the road at a preset step size, and the sampling center coordinates of each sampling center in the vehicle coordinate system are determined. Calculate the derivative value of the main curve of the road at each of the sampling centers, determine the unit tangent vector at each of the sampling centers based on the derivative value, and construct the unit normal vector at each of the sampling centers based on the unit tangent vector; Starting from each of the sampling centers, and extending along the unit normal vector direction and its opposite direction of the sampling center based on a preset horizontal range, a normal cross section is constructed at each of the sampling centers.

4. The method for generating a vehicle-road domain as described in claim 3, characterized in that, The process of determining the intersection points of each of the normal cross sections and various road boundary elements, calculating an ordered sequence of intersection points based on each intersection point and each sampling center, and identifying adjacent intersection points within the same normal cross section as local road domains and corresponding boundary marker pairs includes: Traverse each of the normal cross sections, and sequentially perform intersection operations between the current normal cross section and each of the road boundary elements to obtain the initial intersection points between the current normal cross section and each of the road boundary elements; The initial intersection point is corrected to obtain the final intersection point between the current normal section and each of the road boundary elements; Calculate the signed lateral distance of each of the final intersection points relative to the corresponding sampling center. The signed lateral distance is the projection length of the line vector connecting the final intersection point and the sampling center in the direction of the unit normal vector of the normal section. Based on the signed lateral distance, all final intersection points on each normal section are sorted to generate an ordered sequence of intersection points for each normal section; In the ordered sequence of intersections of each of the normal sections, two adjacent final intersections are paired as a local road domain, and the boundary identifiers of the road boundary elements to which the two final intersections belong are used as the boundary identifier pairs of the local road domains.

5. The method for generating a vehicle-road domain as described in claim 4, characterized in that, The step of correcting the initial intersection point to obtain the final intersection point between the current normal section and each of the road boundary elements includes: Each initial intersection point is checked for validity, and initial intersection points located outside the valid longitudinal range of the road boundary element or outside the preset lateral range of the normal section are removed, thus obtaining the valid intersection points of each normal section. When the same road boundary element has multiple valid intersection points with the same normal section, the valid intersection point closest to the sampling center is taken as the final intersection point of the road boundary element on the normal section.

6. The method for generating a vehicle-road domain as described in claim 5, characterized in that, The step of concatenating boundary markers for the same local road domains along the extension direction of the main road curve to generate candidate vehicle road domains for the current frame includes: The normal sections are arranged sequentially along the extension direction of the main curve of the road to obtain a sequence of normal sections; For any two adjacent normal sections in the normal section sequence, the boundary markers of the two adjacent normal sections are concatenated with the same local road domain to generate an initial candidate vehicle road domain set. The initial candidate vehicle road domain set is filtered for validity based on preset filtering conditions to obtain the candidate vehicle road domain for the current frame.

7. The method for generating a vehicle-road domain as described in claim 6, characterized in that, The process of acquiring historical road domains involves associating and matching the historical road domains with the candidate road domains based on vehicle motion information, determining the region identifiers of the candidate road domains based on the matching results, and generating a set of stable road domains for the current frame, including: Obtain historical road domains, which include historical area identifiers and historical centerline information; Based on the vehicle motion information, historical centerline information, and the centerline of the candidate vehicle road domain in the current frame, a correlation cost matrix between the historical vehicle road domain and the candidate vehicle road domain is constructed. The correlation cost matrix is ​​globally matched and solved to obtain a one-to-one matching relationship between each candidate road domain and each historical road domain. Based on the matching relationship, the region identifier of the candidate vehicle road domain is determined, and a stable vehicle road domain set for the current frame is generated.

8. The method for generating a vehicle-road domain as described in claim 7, characterized in that, The step of determining the region identifier of the candidate vehicle road domain based on the matching relationship and generating a stable vehicle road domain set for the current frame includes: If there is a one-to-one matching relationship between the candidate road domains, the regional identifier of the candidate road domain inherits the historical regional identifier of the historical road domain it matches, and a first road domain is generated. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains meet the preset area threshold, then a new area identifier is assigned to the candidate vehicle road domains to generate a second vehicle road domain. If there is no one-to-one matching relationship between the candidate vehicle road domains and the candidate vehicle road domains do not meet the preset area threshold, then the candidate vehicle road domains are eliminated. A stable set of vehicle road domains for the current frame is generated based on the first and second vehicle road domains.

9. The method for generating a vehicle-road domain as described in claim 7, characterized in that, The process of constructing a correlation cost matrix between the historical road domain and the candidate road domain based on the vehicle motion information, historical centerline information, and the centerline of the candidate road domain in the current frame includes: Based on the vehicle motion information, the historical centerline information of the historical road domain is used to predict the vehicle motion, and the predicted centerline of the historical road domain in the vehicle coordinate system of the current frame is obtained. The centerline distance and longitudinal overlap rate between the predicted centerline and the centerline of the candidate vehicle road domain are statistically analyzed. Based on the centerline distance and longitudinal range overlap rate, an association cost matrix is ​​constructed between the historical vehicle road domain and the candidate vehicle road domain.

10. A vehicle-road domain generation device, characterized in that, include: The module includes: main curve determination module, normal section construction module, local road zone determination module, candidate road zone determination module, and stable road zone determination module. The main curve determination module is used to acquire several types of road boundary elements in the current frame, and perform directional aggregation within a preset directional range based on the various types of road boundary elements to generate the main road curve. The normal section construction module is used to generate a number of sampling centers along the main curve of the road at a preset step size, and to construct a normal section perpendicular to the local tangent of the main curve of the road at each of the sampling centers. The local road domain determination module is used to determine the intersection points of each of the normal sections and various types of road boundary elements, calculate an ordered sequence of intersection points based on each intersection point and each sampling center, and determine adjacent intersection points within the same normal section as local road domains and corresponding boundary identifier pairs; the boundary identifier pairs are constructed based on the road boundary elements of the intersection points; The candidate road domain determination module is used to concatenate local road domains with the same boundary markers along the extension direction of the main road curve to generate candidate vehicle road domains for the current frame. The stable road domain determination module is used to acquire historical road domains, associate and match the historical road domains with the candidate road domains based on vehicle motion information, determine the region identifier of the candidate road domains according to the matching results, and generate a set of stable road domains for the current frame.

Images