CN115435798A - Unmanned vehicle high-precision map road network generation system and method - Google Patents
Unmanned vehicle high-precision map road network generation system and method Download PDFInfo
- Publication number
- CN115435798A CN115435798A CN202210906460.5A CN202210906460A CN115435798A CN 115435798 A CN115435798 A CN 115435798A CN 202210906460 A CN202210906460 A CN 202210906460A CN 115435798 A CN115435798 A CN 115435798A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- map
- marking
- segmentation
- lane
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/28—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network with correlation of data from several navigational instruments
- G01C21/30—Map- or contour-matching
- G01C21/32—Structuring or formatting of map data
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/38—Electronic maps specially adapted for navigation; Updating thereof
- G01C21/3804—Creation or updating of map data
- G01C21/3807—Creation or updating of map data characterised by the type of data
- G01C21/3815—Road data
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
- G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
Abstract
The invention provides a system and a method for generating an unmanned vehicle high-precision map road network. According to the method, accurate track information provided by the unmanned vehicle positioning module, laser point cloud, image and other information are fully utilized, a complete tool chain for semantic point cloud generation, vectorization extraction and topological road network generation is formulated, a lane-level high-precision road network map can be generated rapidly and automatically, and core requirements of unmanned vehicle planning control and auxiliary positioning are met.
Description
Technical Field
The invention relates to the field of unmanned vehicle high-precision maps, in particular to a road network generating system and method of an unmanned vehicle high-precision map.
Background
With the rapid rise of the automatic driving industry, the exploration and the application of unmanned vehicles in various industries are greatly expanded. However, in an actual landing scenario, the requirements for the intelligence and safety of a single vehicle are high, and great risks and challenges still exist, and continuous technical exploration and iteration are required. The automatic driving high-precision map can help a bicycle provide abundant prior, helps the vehicle to better sense, position, plan and control, and improves the intelligence and safety level of the bicycle.
Under the addition of the high-precision map, the automobile can also move from the bicycle intelligence to the car networking intelligence through the intelligent network networking technology. The high-precision map can be accurate to the element granularity of a lane level, the lane can contain rich traffic rules and geometric features, a complete spatial topological relation is constructed between the lane and the lane, more dimensional information such as lane lines and ramp roads of the road is fed back to an electric control system of the automobile, the automobile can have data at any time during automatic driving, the over-the-horizon sensing is carried out, the limitation of automobile body sensing hardware is broken through, and therefore real high-order automatic driving is achieved.
A road network high-precision map of an unmanned vehicle is characterized in that global lane-level topological map data are quickly constructed by utilizing sensor data and track positioning information of the unmanned vehicle, and the map is directly applied to decision planning and auxiliary positioning of the unmanned vehicle. Therefore, the rapid construction of the high-precision road network map is a hot spot of research in the field of unmanned driving at present.
At present, the original data acquisition of a high-precision map mainly has two modes, one mode is a data acquisition vehicle utilizing a professional sensor, and the other mode is a mode of carrying out multi-batch crowd-sourcing mapping on the acquisition vehicle carrying a low-cost sensor. The professional collection vehicle is high in equipment cost and low in data collection efficiency, and the map collection vehicle is not related to the unmanned vehicle. Crowdsourcing collection is data collection by using low-cost sensors matched with vehicle ends, and because the data source precision is low, the requirement of single-batch mapping is difficult to form, a large amount of vehicle ends are required to be collected to upload data, a map with reliable precision and integrity is processed and generated, and a collection vehicle with the same data is a manned vehicle. The method utilizes the sensor equipment of the commercial unmanned vehicle, the equipment cost is between the two methods, and the high-precision sensor equipment and the central data processor which are perfectly matched with the unmanned vehicle can be used for carrying out single-batch rapid construction of the lane-level high-precision road network map through the complete algorithm tool chain processing of the patent. The sensor equipment refers to equipment such as laser radar, a camera, an IMU (inertial measurement unit), a GNSS (global navigation satellite system) and the like.
Disclosure of Invention
The invention provides a road network generating system and method of an unmanned vehicle high-precision map, which aims to solve the problems in the prior art, makes full use of accurate track information provided by an unmanned vehicle positioning module, information such as laser point cloud and images, and sets up a complete tool chain for semantic point cloud generation, vectorization extraction and topological road network generation, can quickly and automatically generate a lane-level high-precision road network map, and meets the core requirements of unmanned vehicle planning control and auxiliary positioning.
The invention provides an unmanned vehicle high-precision map road network generation system which comprises a data receiving module, a semantic map module, a vectorization module and a topological map module, wherein the data receiving module, the semantic map module, the vectorization module and the topological map module are sequentially connected;
the data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module;
the semantic map module is used for performing road surface point cloud segmentation and marking point cloud segmentation according to input data to generate a point cloud map with semantic labels;
the vectorization module is used for extracting vector skeleton lines from the point cloud data segmented from the semantic map to generate a reticle line string containing semantic category attributes;
and the topological map module is used for constructing lanes of the vector marking lines, completing virtual lanes, constructing a spatial inclusion relationship between the marking lines and the lanes and an adjacent relationship between the lanes to form a global and complete lane network map.
The invention also provides a method for generating the unmanned vehicle high-precision map road network, which comprises the following steps:
1) The data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module;
2) Preprocessing the input data by using a semantic map generation module, and performing pavement segmentation and marking segmentation of laser point cloud;
3) Utilizing a marking vectorization generation module to perform longitudinal differential sectioning on the sectioned marking point cloud, clustering the point cloud blocks in each sectioning, extracting key points and a main direction of each clustering cluster, connecting skeleton lines by utilizing the key points, and fitting and optimizing the skeleton lines to generate high-precision smooth vector markings;
4) The method comprises the steps of calculating a mapping area and constructing a global map grid by using a topological map generation module, activating the grid by using tracks, clustering the conditions of multiple tracks on the same road section, constructing an undirected graph connected domain according to the information of track grid units, judging an intersection area and a non-intersection area according to the characteristics of single connection and multi-connection of undirected graph nodes, constructing a topological relation between longitudinal sections and transverse lanes in the sections of the single connected domain of the non-intersection, matching and connecting homonymous lanes between the longitudinal sections and constructing a virtual lane of a lane change point, constructing and connecting virtual lanes on the intersection area according to lane steering information, and finally generating a global lane-level topological road network high-precision map.
In the step 1), the track information comprises time information, three-dimensional coordinate information, attitude information and position translation information of each frame of data at the image acquisition moment;
the laser point cloud map is complete global point cloud data spliced by adopting all frames of laser radar data;
the semantic segmentation image is to segment an image target according to original color image data acquired by image acquisition equipment and output a mask image with a class label.
In the step 2), the preprocessing of the input data refers to the segmentation and ordered organization of the laser point cloud map, the semantic image and the track data;
the input data refers to track data, laser point cloud data and semantic image data; the ordered organization means that the image sequences and the corresponding tracks are ordered according to the time sequence of the tracks, a unique ID is distributed to each segment, each image sequence and the corresponding track are also ordered according to the time sequence in each segment, and ID names are distributed; the segment length takes into account the actual road situation.
The specific implementation mode of the road surface segmentation of the segmented point cloud is that the segmented point cloud is subjected to bilateral filtering noise reduction treatment, uniform sampling interpolation is carried out on the segmented point cloud, ground point cloud near the segmentation track of the unmanned vehicle is used as an initial road surface seed surface patch, surface patch diffusion is carried out according to the normal vector consistency, the intensity consistency, the elevation consistency and the density consistency of the surface patch, and the road surface point cloud is extracted; sequentially processing the point clouds of all the segments to complete the road surface segmentation of the global map;
the specific implementation mode of the marking segmentation is that segmented pavement point clouds are projected to a semantic image plane, multi-frame voting is carried out by utilizing the relation between the projected point clouds and marking semantic blocks, probability statistics is carried out according to the number of times of voting, marking of semantic labels is carried out on the point clouds with voting probability meeting the requirements, then Euclidean distance clustering is carried out on the point cloud blocks in the segments, point cloud intensity mean value statistics is carried out on each cluster, the cluster clusters with large intensity deviation are filtered according to the reflection intensity prior value of the marking point clouds and the statistical intensity mean value, and therefore point cloud clusters which are wrongly classified are screened out; and sequentially processing the point cloud blocks of each segment to finish the extraction of the marking point cloud of the global map.
In the step 3), the longitudinal differential sectioning is carried out on the marked line point cloud, namely according to the track advancing direction and according to smaller intervals;
extracting the key points, namely clustering the mark line point clouds in the segments according to Euclidean distances, and calculating gravity center points of point cloud blocks of the same cluster as the key points of the cluster;
the main direction extraction means that for the cluster of each point cloud block, a direction vector with the largest variance is calculated in a PCA mode to serve as the main direction of the cluster;
the connection of the skeleton lines refers to the connection of key points according to the time sequence, the spatial correlation and the main direction consistency of the cluster;
<xnotran> , (v 1, v2,..., vn) (lane _ list _1,lane_list_2,..., lane _ list _ n), ; </xnotran> Entering the next segmentation, checking whether the cluster m of the current segmentation is overlapped with the cluster n of the previous segmentation, if so, pressing the key point vm of the cluster into a list lane _ list _ n of the corresponding vector marking, and completing the connection of the key point vm of the cluster m of the current segmentation; if the distance between the cluster m and the cluster n is not overlapped, but the distance and the main direction of the cluster m and the cluster n are smaller than the set threshold value, the key point vm is pressed into a list lane _ list _ n of the corresponding vector marking line to complete connection; if the cluster m in the current segmentation does not have the overlapping degree with all clusters in the previous segmentation, and the distance threshold or the main direction angle threshold is over, the key point of the cluster is considered as the vertex of a new vector marking, a vector marking vertex list Lane _ list _ n +1 is newly built, and the key point vm is pressed; the connection of all key points of the cutting is finished by analogy in sequence;
the fitting and optimization of the skeleton line refers to fitting the skeleton line according to a quadratic B spline curve function, and simplifying the fitted curve by using a Douglas Pock method.
In the step 4), the grid division is performed on the mapping region, namely a regular grid is created by taking a bounding box of the mapping region as a grid range and taking the actual pavement width as the size of the grid;
the activation of the grid unit by using the driving track means that traversing track information marks a unit of which a track point falls into a grid, if a plurality of tracks exist in the same road section, clustering or redundancy removal processing needs to be carried out on the tracks at first, and the same road section is ensured to keep a complete track;
the method for constructing the undirected graph by using the trajectory grid is characterized in that the undirected graph G = (V, E) is defined, the vertex Vi of a trajectory cluster in a grid is calculated for an activated grid unit, the vertex of a grid unit in an adjacent field is constructed with an edge connection Ei, and an adjacency matrix of the undirected graph is generated;
judging a single connected road section and an intersection road section by the undirected graph, namely determining whether more than two sides are connected with one vertex Vi according to the relation between the vertex and the sides of the constructed undirected graph, if more than two sides are connected, considering the vertex as the intersection vertex, otherwise, considering the vertex as the vertex of the single connected road section area;
the method comprises the steps of constructing a connection relation of an intersection according to an undirected graph, namely determining a topological connection relation between side connections Ei1 and Ei2, ei1 and Ei3, ei2 and Ei3,. According to a topological relation between an intersection vertex Vi and a plurality of sides (Ei 1, ei2, ei3,. Eta.);
the connection relation of the single connected road section is established according to the undirected graph, namely the relation between the connection of the edges of the single connected domain road section (Ej 1, ej2, ej3,.. Multidot.Ejn) is determined according to the vertex Vj of the undirected graph, the front adjacent edge Ej1 and Ej2, the vertex Vj +1, the front adjacent edge Ej2 and Ej 3;
the method comprises the following steps of firstly setting a maximum longitudinal segmentation threshold DL, detecting a change point of a vector marking along the road advancing direction from the initial position of the single connected road section, immediately starting longitudinal cutting for segmentation if the change point is detected to generate a longitudinal segment, and starting cutting at the maximum longitudinal segmentation threshold DL to generate the segment if the change point is not detected; where i represents the number of a single connected domain and m represents the number of a segment within a single connected domain.
The change points of the vector marked lines refer to head and tail points of the vector marked lines, an intersection point of a plurality of vector marked lines, and change points of a dotted line and a solid line of the same marked line;
the method for constructing the horizontal topological relation of the vector marking lines in the segments comprises the steps of transversely sequencing a plurality of vector marking lines in the segments, determining a reference line of the marking lines by using the priori knowledge of semantic labels of the marking lines, using the reference marking lines as dividing lines of opposite lanes, constructing lane surfaces for two adjacent marking lines according to the sequenced marking lines on the left side and the right side of the reference line, sequencing the lane surfaces lane by taking the reference line as a spacing between the left side and the right side, sequencing the left side sequentially from 1, 2.
The matching and connection of the lane planes between the segments refers to aligning joints between the segments according to the relation between the reference marked lines of adjacent segment references and the nearest neighbor; for the condition that a plurality of lanes correspond to one lane or one lane corresponds to a plurality of lanes, constructing a virtual marking line segment according to the principle that the steering angle is not smaller than the angle threshold value, and completing the longitudinal communication of the lane surface;
the construction and connection of the virtual marked lines for the intersection region are that the connection relation of the intersection is constructed according to an undirected graph, the single connected subsection and the connected relation are mapped through the position relation between grid units on the sides of the undirected graph and the longitudinal subsection section, the matching of lanes in the two subsections is carried out according to the driving rule, the left marked lines and the right marked lines of the lanes are connected according to a Bezier curve, and the virtual lane surface of the intersection is generated.
The invention also provides a terminal of the unmanned vehicle high-precision map road network generation system, which comprises a processor, a memory, a communication interface and a computer program; the computer program is stored in the memory and configured to be executed by the processor, the computer program comprising instructions for performing an unmanned vehicle height map road network generation method.
The invention also provides a computer program of the unmanned vehicle high-precision map network generation method, wherein the computer program comprises instructions for executing the unmanned vehicle high-precision map network generation method.
The present invention also provides a computer-readable storage medium storing the above computer program.
The invention has the beneficial effects that:
1. receiving a semantic segmentation image output by an image processing module by receiving track information and a laser point cloud map output by an unmanned vehicle positioning module;
2. the thinking of region segmentation, segmentation and combination is used for fusing the laser point cloud and the semantic image to rapidly extract the marking line and the pavement semantic point cloud;
3. rapidly extracting vector marking lines by using longitudinal differential segmentation, key point extraction, skeleton line tracking and skeleton line optimization;
4. constructing a road-level topological relation by using track gridding, generating an undirected graph based on a grid and judging a connected region, and identifying a crossing region and a single connected region based on the road-level topological relation; longitudinal segmentation of the marking is carried out in the single communication road section, transverse topological relation and lane construction are carried out on the marking in the segmentation, and matching and communication are carried out between the segments according to the connectivity of the marking; and constructing a virtual lane in the intersection area according to a lane steering traffic rule, and connecting the virtual lane with the single communication road section.
5. The method provides three complete tool chains of semantic map generation, element vectorization and topological map construction, and provides a lane construction and connection technology of an intersection road section and a single connected road section under the guidance of an undirected graph connected domain, so that a high-precision vector topological map can be quickly generated, the precision and the efficiency of high-precision road network map generation are remarkably improved, and the whole high-precision road network map generation system has good accuracy, timeliness and stability.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts. The system comprises a data receiving module, a semantic map module, a vectorization module and a topological map module which are connected in sequence;
FIG. 1 is a schematic view of a semantic map module operation flow;
figure 2 is a schematic diagram of a vectoring module operation flow;
FIG. 3 is a schematic view of a topology map module operation process;
FIG. 4 is a schematic diagram of a road network generating system of an unmanned vehicle high-precision map;
FIG. 5 is a schematic diagram of a terminal of an unmanned vehicle high-precision map road network generation system;
FIG. 6 is a schematic illustration of lane plane matching and connection between segments;
FIG. 7 is a schematic diagram of virtual lane surfaces of an intersection generated by connection according to Bezier curves;
FIG. 8 is a semantic point cloud map;
FIG. 9 is a vectorized reticule map;
fig. 10 is a topological map.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
In a first aspect, an embodiment of the present application provides an unmanned vehicle high-precision map road network generating system and apparatus, and the method includes the following steps:
1. and the data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module.
The track information comprises time information, three-dimensional coordinate information, posture information and position translation information of each frame of data at the image acquisition moment.
The laser point cloud map is complete global point cloud data spliced by adopting all frames of laser radar data.
The semantic segmentation image is to segment an image target according to original color image data acquired by image acquisition equipment and output a mask image with a class label.
2. And the semantic map generation module is used for preprocessing the input data and performing road surface segmentation and marking segmentation of laser point cloud, as shown in figure 1.
2-1, the preprocessing of the input data refers to the segmentation and ordered organization of the laser point cloud map, the semantic image and the track data.
The input data refers to track data, laser point cloud data and semantic image data; the ordered organization means that the image sequences and the corresponding tracks are ordered according to the time sequence of the tracks, a unique ID is distributed to each segment, each image sequence and the corresponding track are also ordered according to the time sequence in each segment, and ID names are distributed; the segment length takes into account the actual road situation, for example, the segmentation is performed according to a threshold value of 20 meters.
2-2 the specific implementation mode of the road surface segmentation of the segmentation point cloud is that the segmentation point cloud is subjected to noise reduction processing of bilateral filtering, uniform sampling interpolation is carried out on the segmentation point cloud, ground point cloud near the segmentation track of the unmanned vehicle is used as an initial road surface seed surface patch, surface patch diffusion is carried out according to the normal vector consistency, the strength consistency, the elevation consistency and the density consistency of the surface patch, and the road surface point cloud is extracted. And sequentially processing the point clouds of all the segments to complete the road surface segmentation of the global map.
2-3 the specific implementation mode of the marking segmentation is that the segmented pavement point cloud is projected to a semantic image plane, multi-frame voting is carried out by utilizing the relation between the projected point cloud and the marking semantic block, probability statistics is carried out according to the hit times of the voting, the marking of a semantic label is carried out on the point cloud meeting the voting probability requirement, then the point cloud blocks in the segmentation are clustered by Euclidean distance, the point cloud intensity mean value is counted for each cluster, the cluster with larger intensity deviation is filtered according to the reflection intensity prior value of the marking point cloud and the counted intensity mean value, and thus the point cloud cluster with wrong segmentation is screened out. And sequentially processing the point cloud blocks of each segment to finish the extraction of the marking line point cloud of the global map.
3. The reticle vectorization generation module performs longitudinal differential sectioning on the sectioned reticle point cloud, clusters the cloud blocks of the inner points of each section, extracts key points and a main direction of each cluster, connects skeleton lines by using the key points, and fits and optimizes the skeleton lines to generate high-precision smooth vector reticles as shown in fig. 2.
3-1, the longitudinal differential segmentation of the reticle point cloud means that longitudinal cutting is performed at small intervals, for example, at a distance of 1 meter, according to the track traveling direction, and a certain overlap degree is maintained between adjacent segmentation units, for example, an overlap degree of 30% of the segmentation length is set.
3-2, the extraction of the key points refers to clustering the Euclidean distance of the marked line point clouds in the segments, and calculating the gravity center point of the point cloud blocks of the same cluster as the key points of the cluster.
The main direction extraction means that for the cluster of each cloud block, a direction vector with the largest variance is calculated by using a PCA (principal component analysis) mode to serve as the main direction of the cluster.
3-3, the connection of the skeleton lines refers to the connection of key points according to the time sequence, the spatial correlation and the main direction consistency of the cluster. <xnotran> , (v 1, v2,..., vn) (lane _ list _1,lane_list_2,..., lane _ list _ n), ; </xnotran> Entering the next segmentation, checking whether the cluster m of the current segmentation is overlapped with the cluster n of the previous segmentation, if so, pressing the key point vm of the cluster into a list lane _ list _ n of the corresponding vector marking, and completing the connection of the key point vm of the cluster m of the current segmentation; if the distance between the cluster m and the cluster n is not overlapped, but the distance and the main direction of the cluster m and the cluster n are smaller than the set threshold value, the key point vm is pressed into a list lane _ list _ n of the corresponding vector marking line to complete connection; and if the cluster m in the current segmentation does not have the overlapping degree with all clusters in the previous segmentation, and the distance threshold or the main direction angle threshold is over, determining that the key point of the cluster is a new vector marking vertex, newly building a vector marking vertex list _ n +1, and pressing in the key point vm. And the analogy is repeated to complete the connection of all key points of the cutting segment.
3-4, the fitting and optimization of the skeleton line refers to fitting the skeleton line according to a quadratic B spline curve function, and simplifying a fitted curve by using a Douglas Puck method.
4. The method comprises the steps of utilizing a topological map generation module, as shown in figure 3, calculating a map building area, building a global map grid, activating the grid by using tracks, carrying out clustering processing on the condition of multiple tracks of the same road section, building an undirected graph connected domain according to the information of track grid units, judging an intersection area and a non-intersection area according to the characteristics of single connection and multiple connection of undirected graph nodes, carrying out longitudinal segmentation on the single connected domain of the non-intersection and building a topological relation of transverse lanes in the segmentation, carrying out matching and connection between the same-name lanes between the longitudinal segmentation and virtual lanes of lane change points, carrying out construction and connection of virtual lanes on the intersection area according to lane steering information, and finally generating a global-level topological road network high-precision map.
4-1, the step of performing mesh division on the mapping region refers to creating a regular grid by using a bounding box of the mapping region as a mesh range and using an actual road surface width as the size of the mesh.
4-2, activating the grid cells by using the driving tracks, namely traversing track information to mark cells of which track points fall into the grid, and if a plurality of tracks exist in the same road section, clustering or redundancy removal processing needs to be performed on the tracks at first to ensure that the same road section keeps a complete track.
4-3, constructing an undirected graph by using the trajectory mesh, which means defining an undirected graph G = (V, E), calculating a trajectory cluster vertex Vi in the mesh for an activated mesh unit, constructing an edge connection Ei for the vertices of the mesh units in the adjacent domain, and generating an adjacency matrix of the undirected graph.
4-4, judging the single connected road section and the intersection road section by the undirected graph means that whether more than two sides are connected exists in the same vertex Vi or not is determined according to the relation between the vertex and the sides of the constructed undirected graph, if the more than two sides are connected, the vertex is considered as the intersection vertex, and if the more than two sides are connected, the vertex is considered as the vertex of the single connected road section area.
4-5, constructing the connection relationship of the intersection according to the undirected graph refers to determining the topological connection relationship between edge connections Ei1 and Ei2, ei1 and Ei3, ei2 and Ei 3.
4-6, the connection relation of the single connected road section is constructed according to the undirected graph, which means that the relation between the connection of the edges of the single connected domain road section (Ej 1, ej2, ej 3.., ejn) is determined according to the vertex Vj of the undirected graph and the two adjacent edges Ej1 and Ej2, and the vertex Vj +1 and the two adjacent edges Ej2 and Ej 3.
4-7, the specific implementation manner is that a maximum longitudinal division threshold value DL is set firstly, the change point of the vector marking is detected from the starting position of the single connected road section along the road advancing direction, if the change point is detected, longitudinal cutting is started immediately for segmentation to generate a longitudinal segment, and if the change point is not detected, cutting is started at the maximum longitudinal division threshold value DL to generate a segment. Where i represents the number of a single connected domain and m represents the number of segments within a single connected domain.
The change points of the 4-8 vector marked lines refer to head and tail points of the vector marked lines, the intersection points of a plurality of vector marked lines, change points of a broken line and a solid line of the same marked line and the like.
4-9, constructing a transverse topological relation for vector marking lines in a segment, namely transversely sequencing a plurality of vector marking lines in the segment, determining a reference line of the marking lines by using the priori knowledge of semantic labels of the marking lines, using the reference marking lines as dividing lines of opposite lanes, constructing lane surfaces for two adjacent marking lines according to the sequenced marking lines on the left side and the right side of the reference line, sequencing lane surface lanes by taking the reference line as a separation on the left side and the right side, sequencing the left side sequentially from 1, 2.
4-10, the matching and connection of lane planes between the segments refers to aligning joints between the segments according to the relation between the reference mark lines of the adjacent segments and the nearest neighbors. And for the condition that a plurality of lanes correspond to one lane or one lane corresponds to a plurality of lanes, constructing a virtual marking line segment according to the principle that the steering angle is not less than an angle threshold (for example, 120 degrees), and completing the longitudinal penetration of the lane surface. As shown in FIG. 6, the lane surface with the ID of-2 of the segment section1 corresponds to the lane surfaces with the IDs of-2 and-3 of the segment section2, the section1 (-2) is directly connected with the section2 (-2), the section1 (-2) and the section2 (-3) need to construct the following virtual reticle shown in FIG. 6, and the connection of the section1 (-2) and the section2 (-3) is completed.
4-11, the construction and connection of the virtual marking for the intersection region means that according to the side connection relationship described in 4-5, the position relationship between the grid unit of the non-directional graph side and the longitudinal segment section maps the single connected segment and the connected relationship, the matching between the lanes in the two segments is performed according to the driving rules (for example, the left turn is connected with the right lane of the left turn segment according to the right-side left-turn lane of the current segment reference line, the right turn is connected with the right lane of the right turn segment according to the right-side lane of the current segment reference line), the left and right marking of the lane are connected according to the bezier curve, and the virtual lane surface of the intersection is generated. The Bezier curve is not limited to a specific order in the invention, and can be flexibly selected according to actual conditions. As shown in fig. 7, the disclosure and schematic illustration are made according to the construction principle of the second order bezier curve.
P i ＝(1-t) 2 P a +2t(1-t)P e +t 2 P c i＝1,2,...,m
P j ＝(1-t) 2 P b +2t(1-t)P f +t 2 P d j＝1,2,...,n
Wherein t is ∈ [0,1 ]]E and f are the intersection points of the extension lines of the left marked line (a-c) and the right marked line (b-d) matched with the sectional lane, P i And P j The number of vertexes of the virtual marking, which are corresponding points i and j on the left marking line and the right marking line of the virtual lane, can be calculated by dividing the value range of t into m equal parts and n equal parts respectively.
In a second aspect, an embodiment of the present application provides an unmanned vehicle advanced map road network generating system, as shown in fig. 4, including:
the data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module;
the semantic map module is used for performing road surface point cloud segmentation and marking line point cloud segmentation according to input data to generate a point cloud map with semantic labels, as shown in fig. 8;
the vectorization module is used for extracting vector skeleton lines from the point cloud data segmented from the semantic map to generate a reticle string containing semantic category attributes, and the obtained vectorization reticle map is shown in fig. 9;
and the topological map module is used for constructing lanes of the vector marking lines, completing virtual lanes, constructing a spatial inclusion relationship between the marking lines and the lanes and an adjacent relationship between the lanes to form a global and complete lane network map, as shown in fig. 10.
In a third aspect, an embodiment of the present application provides a terminal, as shown in fig. 5, including: a processor, a memory, a communication interface, and a computer program;
wherein the computer program is stored in the memory and configured to be executed by the processor, the computer program comprising instructions for performing the method of the first aspect.
In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where a computer program is stored, and the computer program causes a server to execute the method according to the first aspect.
In a fifth aspect, the present application provides a computer program product, which includes computer instructions for executing the method of the first aspect by a processor.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus embodiment, the above is only a preferred embodiment of the present invention, and since it is basically similar to the method embodiment, it is described simply, and the relevant points can be referred to the partial description of the method embodiment. The above description is only for the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and the protection scope of the present invention should be covered by the principle of the present invention without departing from the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (9)
1. The utility model provides an unmanned vehicle height map road network generating system which characterized in that: the system comprises a data receiving module, a semantic map module, a vectorization module and a topological map module which are connected in sequence;
the data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module;
the semantic map module is used for performing road surface point cloud segmentation and marking point cloud segmentation according to input data to generate a point cloud map with semantic labels;
the vectorization module is used for extracting vector skeleton lines from the point cloud data segmented from the semantic map to generate a reticle line string containing semantic category attributes;
and the topological map module is used for constructing lanes for the vector marking lines, completing virtual lanes, constructing a spatial inclusion relationship between the marking lines and the lanes and an adjacent relationship between the lanes and forming a global and complete lane network map.
2. An unmanned vehicle high-precision map road network generation method is characterized by comprising the following steps:
1) The data receiving module is used for receiving the track information and the laser point cloud map sent by the unmanned vehicle positioning module and receiving the semantic segmentation image sent by the image processing module;
2) Preprocessing the input data by using a semantic map generation module, and performing pavement segmentation and marking segmentation of laser point cloud;
3) Utilizing a marking vectorization generation module to perform longitudinal differential sectioning on the sectioned marking point cloud, clustering the point cloud blocks in each sectioning, extracting key points and a main direction of each clustering cluster, connecting skeleton lines by utilizing the key points, and fitting and optimizing the skeleton lines to generate high-precision smooth vector markings;
4) The method comprises the steps of calculating a mapping area and constructing a global map grid by using a topological map generation module, activating the grid by using tracks, carrying out clustering processing on the condition of multiple tracks of the same road section, constructing an undirected graph connected domain according to information of track grid units, judging an intersection area and a non-intersection area according to characteristics of single connection and multi-connection of undirected graph nodes, constructing a topological relation between longitudinal sections and transverse lanes in the sections of the single connected domain of the non-intersection, matching and connecting homonymous lanes between the longitudinal sections and constructing a virtual lane of a lane change point, constructing and connecting virtual lanes on the intersection area according to lane steering information, and finally generating a global lane-level topological road network high-precision map.
3. The unmanned vehicle high-precision map road network generation method according to claim 2, characterized in that: in the step 1), the track information comprises time information, three-dimensional coordinate information, posture information and position translation information of each frame of data at the image acquisition moment;
the laser point cloud map is complete global point cloud data spliced by adopting all frames of laser radar data;
the semantic segmentation image is to segment an image target according to original color image data acquired by image acquisition equipment and output a mask image with a class label.
4. The unmanned vehicle high-precision map road network generation method according to claim 2, characterized in that: in the step 2), the input data is preprocessed by segmenting and orderly organizing the laser point cloud map, the semantic image and the track data;
the input data refers to track data, laser point cloud data and semantic image data; the ordered organization means that the image sequences and the corresponding tracks are ordered according to the time sequence of the tracks, a unique ID is distributed to each segment, each image sequence and the corresponding track are also ordered according to the time sequence in each segment, and ID names are distributed; the segment length takes into account the actual road situation.
5. The unmanned vehicle height map road network generation method according to claim 2 or 4, characterized in that: in the step 2), the road surface segmentation of the segmented point cloud is implemented by performing bilateral filtering denoising processing on the segmented point cloud, performing uniform sampling interpolation on the segmented point cloud, performing surface patch diffusion according to the normal vector consistency, the intensity consistency, the elevation consistency and the density consistency of the surface patches by using the ground point cloud near the unmanned vehicle segmentation track as an initial road surface seed surface patch, and extracting the road surface point cloud; sequentially processing the point clouds of all the segments to complete the road surface segmentation of the global map;
the specific implementation mode of the marking segmentation is that segmented pavement point clouds are projected to a semantic image plane, multi-frame voting is carried out by utilizing the relation between the projected point clouds and marking semantic blocks, probability statistics is carried out according to the number of times of voting, marking of semantic labels is carried out on the point clouds with voting probability meeting the requirements, then Euclidean distance clustering is carried out on the point cloud blocks in the segments, point cloud intensity mean value statistics is carried out on each cluster, the cluster clusters with large intensity deviation are filtered according to the reflection intensity prior value of the marking point clouds and the statistical intensity mean value, and therefore point cloud clusters which are wrongly classified are screened out; and sequentially processing the point cloud blocks of each segment to finish the extraction of the marking line point cloud of the global map.
6. The unmanned vehicle height map road network generation method according to claim 2, characterized in that: in the step 3), the longitudinal differential sectioning is carried out on the marked line point cloud, namely according to the track advancing direction and according to smaller intervals;
extracting the key points, namely clustering the Euclidean distance of the marked line point clouds in the segments, and calculating the gravity center point of the point cloud blocks of the same cluster as the key points of the cluster;
the extraction of the main direction refers to that for the cluster of each cloud block, a direction vector with the largest variance is calculated by using a PCA (principal component analysis) mode to serve as the main direction of the cluster;
the connection of the skeleton lines refers to the connection of key points according to the time sequence, the spatial correlation and the main direction consistency of the cluster;
the fitting and optimization of the skeleton line refers to fitting the skeleton line according to a quadratic B spline curve function, and simplifying the fitted curve by using a Douglas pock method.
7. The unmanned vehicle high-precision map road network generation method according to claim 6, characterized in that: the specific implementation mode of the connection of the skeleton line is that different vector marking container lists (lane _ list _1, lane _list_2.. And lane _ list _ n) are newly built in the first segmentation according to the key points (v 1, v 2.. And vn) extracted in the previous step, and the corresponding key points are pressed in; entering the next segmentation, checking whether the cluster m of the current segmentation is overlapped with the cluster n of the previous segmentation, if so, pressing the key point vm of the cluster into a list lane _ list _ n of the corresponding vector marking, and completing the connection of the key point vm of the cluster m of the current segmentation; if the distance between the cluster m and the cluster n is not overlapped, but the distance and the main direction of the cluster m and the cluster n are smaller than the set threshold value, the key point vm is pressed into a list lane _ list _ n of the corresponding vector marking line to complete connection; if the cluster m in the current segmentation does not have the overlapping degree with all clusters in the previous segmentation, and the distance threshold or the main direction angle threshold is over, the key point of the cluster is considered as the vertex of a new vector marking, a vector marking vertex list Lane _ list _ n +1 is newly built, and the key point vm is pressed; and analogizing in turn to complete the connection of all key points of the cutting segments.
8. The unmanned vehicle height map road network generation method according to claim 2, characterized in that: in the step 4), the grid division is performed on the mapping region, namely a regular grid is created by taking a bounding box of the mapping region as a grid range and taking the actual pavement width as the size of the grid;
the activation of the grid unit by using the driving track means that traversing track information marks a unit of which a track point falls into a grid, if a plurality of tracks exist in the same road section, clustering or redundancy removal processing needs to be carried out on the tracks at first, and the same road section is ensured to keep a complete track;
the method for constructing the undirected graph by using the trajectory grid is characterized in that the undirected graph G = (V, E) is defined, the vertex Vi of a trajectory cluster in a grid is calculated for an activated grid unit, the vertex of a grid unit in an adjacent field is constructed with an edge connection Ei, and an adjacency matrix of the undirected graph is generated;
judging a single connected road section and an intersection road section by the undirected graph, namely determining whether more than two sides are connected with one vertex Vi according to the relation between the vertex and the sides of the constructed undirected graph, if more than two sides are connected, considering the vertex as the intersection vertex, otherwise, considering the vertex as the vertex of the single connected road section area;
the connection relation of the intersection is constructed according to the undirected graph, namely the topological connection relation between the edges connecting Ei1 and Ei2, ei1 and Ei3, ei2 and Ei3 is determined according to the topological relation between the vertex Vi of the intersection and a plurality of edges (Ei 1, ei2, ei 3.);
the connection relation of the single connected road section is established according to the undirected graph, namely the relation between the connection of the edges of the single connected domain road section (Ej 1, ej2, ej3,.. Multidot.Ejn) is determined according to the vertex Vj of the undirected graph, the front adjacent edge Ej1 and Ej2, the vertex Vj +1, the front adjacent edge Ej2 and Ej 3;
the method comprises the following steps of firstly setting a maximum longitudinal segmentation threshold DL, detecting a change point of a vector marking along the road advancing direction from the initial position of the single connected road section, immediately starting longitudinal cutting for segmentation if the change point is detected to generate a longitudinal segment, and starting cutting at the maximum longitudinal segmentation threshold DL to generate the segment if the change point is not detected; where i represents the number of a single connected domain and m represents the number of a segment within a single connected domain.
The change points of the vector marking lines refer to head and tail points of the vector marking lines, the intersection points of a plurality of vector marking lines, and change points of a broken line and a solid line of the same marking line;
the method comprises the steps of constructing a transverse topological relation for vector marking lines in a subsection, transversely sequencing a plurality of vector marking lines in the subsection, determining a reference line of the marking lines by using the priori knowledge of semantic labels of the marking lines, using the reference marking lines as dividing lines of opposite lanes, constructing lane surfaces for two adjacent marking lines according to the sequenced marking lines on the left side and the right side of the reference line, sequencing lane surfaces Lane by taking the reference line as a left side and a right side at intervals, sequencing the left side sequentially from 1,2,. To. Starting sequencing, and sequencing the right side sequentially from-1, -2,. To. Starting sequencing;
the matching and connection of the lane planes between the subsections refers to the alignment of joints between the subsections according to the relation between the reference mark lines of the adjacent subsection and the nearest neighbor; for the condition that a plurality of lanes correspond to one lane or one lane corresponds to a plurality of lanes, constructing a virtual marking line segment according to the principle that the steering angle is not smaller than the angle threshold value, and completing the longitudinal communication of the lane surface;
the construction and connection of the virtual marked lines for the intersection region are that the connection relationship of the intersection is constructed according to an undirected graph, the single connected subsection and the connected relationship are mapped through the position relationship between the grid units on the sides of the undirected graph and the longitudinal subsection section, the lanes in the two subsections are matched according to the driving rule, the left marked lines and the right marked lines of the lanes are connected according to a Bezier curve, and the virtual lane surface of the intersection is generated.
9. A terminal of an unmanned vehicle high-precision map road network generation system is characterized in that: comprising a processor, memory, a communications interface, and a computer program; the computer program stored in the memory and configured to be executed by the processor, the computer program comprising instructions for performing the method of claim 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210906460.5A CN115435798A (en) | 2022-07-29 | 2022-07-29 | Unmanned vehicle high-precision map road network generation system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210906460.5A CN115435798A (en) | 2022-07-29 | 2022-07-29 | Unmanned vehicle high-precision map road network generation system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115435798A true CN115435798A (en) | 2022-12-06 |
Family
ID=84243376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210906460.5A Pending CN115435798A (en) | 2022-07-29 | 2022-07-29 | Unmanned vehicle high-precision map road network generation system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115435798A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116106853A (en) * | 2023-04-12 | 2023-05-12 | 陕西欧卡电子智能科技有限公司 | Method for identifying dynamic and static states of water surface scene target based on millimeter wave radar |
| CN116309943A (en) * | 2023-05-24 | 2023-06-23 | 联友智连科技有限公司 | Parking lot semantic map road network construction method and device and electronic equipment |
| CN116878487A (en) * | 2023-09-07 | 2023-10-13 | 河北全道科技有限公司 | Method and device for establishing automatic driving map, vehicle and server |
| CN117128984A (en) * | 2023-02-21 | 2023-11-28 | 荣耀终端有限公司 | Navigation map generation method and device |
-
2022
- 2022-07-29 CN CN202210906460.5A patent/CN115435798A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117128984A (en) * | 2023-02-21 | 2023-11-28 | 荣耀终端有限公司 | Navigation map generation method and device |
| CN116106853A (en) * | 2023-04-12 | 2023-05-12 | 陕西欧卡电子智能科技有限公司 | Method for identifying dynamic and static states of water surface scene target based on millimeter wave radar |
| CN116106853B (en) * | 2023-04-12 | 2023-09-01 | 陕西欧卡电子智能科技有限公司 | Method for identifying dynamic and static states of water surface scene target based on millimeter wave radar |
| CN116309943A (en) * | 2023-05-24 | 2023-06-23 | 联友智连科技有限公司 | Parking lot semantic map road network construction method and device and electronic equipment |
| CN116309943B (en) * | 2023-05-24 | 2023-08-08 | 联友智连科技有限公司 | Parking lot semantic map road network construction method and device and electronic equipment |
| CN116878487A (en) * | 2023-09-07 | 2023-10-13 | 河北全道科技有限公司 | Method and device for establishing automatic driving map, vehicle and server |
| CN116878487B (en) * | 2023-09-07 | 2024-01-19 | 河北全道科技有限公司 | Method and device for establishing automatic driving map, vehicle and server |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111462275B (en) | Map production method and device based on laser point cloud | |
| CN115435798A (en) | Unmanned vehicle high-precision map road network generation system and method | |
| CN109241069B (en) | Road network rapid updating method and system based on track adaptive clustering | |
| CN108955702B (en) | Lane-level map creation system based on three-dimensional laser and GPS inertial navigation system | |
| CN110220521B (en) | High-precision map generation method and device | |
| WO2020029601A1 (en) | Method and system for constructing transverse topological relationship of lanes in map, and memory | |
| CN108171131B (en) | Improved MeanShift-based method for extracting Lidar point cloud data road marking line | |
| CN111985322A (en) | Road environment element sensing method based on laser radar | |
| CN105667518A (en) | Lane detection method and device | |
| CN114812581B (en) | Cross-country environment navigation method based on multi-sensor fusion | |
| CN109256028B (en) | Method for automatically generating unmanned high-precision road network | |
| CN110021072B (en) | Holographic mapping-oriented multi-platform point cloud intelligent processing method | |
| CN103703490A (en) | Device for generating three-dimensional feature data, method for generating three-dimensional feature data, and recording medium on which program for generating three-dimensional feature data is recorded | |
| CN113723715B (en) | Method, system, equipment and storage medium for automatically matching public transport network with road network | |
| CN111652241B (en) | Building contour extraction method integrating image features and densely matched point cloud features | |
| CN114020015A (en) | Unmanned aerial vehicle path planning system and method based on barrier map bidirectional search | |
| Zhang et al. | An efficient LiDAR-based localization method for self-driving cars in dynamic environments | |
| CN113310491A (en) | Unmanned aerial vehicle navigation network automatic generation method considering specific structure | |
| CN115564865A (en) | Construction method and system of crowdsourcing high-precision map, electronic equipment and vehicle | |
| CN115861968A (en) | Dynamic obstacle removing method based on real-time point cloud data | |
| CN113903173B (en) | Vehicle track feature extraction method based on directed graph structure and LSTM | |
| CN115077550A (en) | Unmanned vehicle route planning method based on matching of navigation map and high-precision map | |
| CN114509065A (en) | Map construction method, map construction system, vehicle terminal, server side and storage medium | |
| Bastani et al. | Inferring and improving street maps with data-driven automation | |
| Dørum | Deriving double-digitized road network geometry from probe data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |