Vehicle lane semantic object grouping method and high-precision map making method and device

Abstract

The invention discloses a lane semantic object grouping method and a high-precision map making method and device. The grouping method comprises the steps that whether the width of a single lane semantic object included in the same road section is larger than a preset width threshold value or not is judged; if yes, longitudinally segmenting the single stitching semantic object to obtain at least two stitching semantic objects; and traversing all the lane semantic objects included in the same road section in pairs, grouping the lane semantic objects according to the spatial position relationship of the lane semantic objects and / or the single-trip vehicle driving direction when the lane semantic objects are collected, and clustering by taking the lane semantic objects included in each group as grouping units. According to the method, the compatibility on the unpredictable condition of the clustering result and the strong dependence on the input condition are reduced. The technical defect that complex vehicle lines may be wrongly clustered or correct vehicle lines cannot be fitted in the prior art is overcome, the data quality of a high-precision map is integrally improved, and the error risk is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic information and the technical field of high-precision map making, and in particular relates to a lane semantic object grouping method, a high-precision map making method and device. BACKGROUND

[0002] A high-precision map can be used for automatic driving of a vehicle, and recognition and making of road markings are important components of generating the high-precision map, and the road markings are generally recognized and extracted through road image information collected by a camera mounted on a vehicle. In the process of making the high-precision map, the inventors solve the problem of large investment of resources in single closed-loop collection (single closed-loop collection), optimize the application scenario, and collect data by using a private car as a collection device under the premise of not interfering with the stable and fast driving of the vehicle. However, the use of a private car to collect data cannot strictly require a vehicle to travel along a closed-loop trajectory, and most vehicle trajectories are only one-way trajectories from a starting point to a destination, so that the multi-pass single-trajectory data is used in the later making of the high-precision map, that is, the road marking information in each pass of collected image needs to be extracted, and finally the same road marking collected in multiple passes is fused to make a high-precision map. SUMMARY

[0003] In order to solve the technical defects that the road markings collected in multiple passes generate incorrect clusters in the clustering process, and cannot output a single road marking or output an incorrect road marking during fusion (fitting), and in order to further improve the data quality, reduce the risk of errors, and enrich the technical route and increase the selection space, the present application embodiment provides a lane semantic object grouping method, a high-precision map making method and device.

[0004] In a first aspect, the present application embodiment provides a lane semantic object grouping method, which can include:

[0005] determining whether the width of a single lane semantic object included in a same road section is greater than a preset width threshold; if yes, the single lane semantic object is vertically divided to obtain at least two lane semantic objects;

[0006] all lane semantic objects included in a same road section are iterated two by two, and the lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects and / or the single-vehicle driving direction when the lane semantic objects are collected, and the lane semantic objects included in each group are taken as a grouping unit for clustering.

[0007] Optionally, the lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects, which can include:

[0008] grouping the lane semantic objects according to an included angle between two lane semantic objects, a height difference between the two lane semantic objects, and / or a direction of a vectorized point sequence of the two lane semantic objects;

[0009] if the included angle between the two lane semantic objects is greater than a preset included angle threshold, the height difference between the two lane semantic objects is greater than a preset height difference threshold, and / or the direction of the vectorized point sequence of the two lane semantic objects is inconsistent, the two lane semantic objects are divided into different groups; otherwise, the two lane semantic objects are divided into the same group.

[0010] Optionally, before grouping the lane semantic objects according to the spatial position relationship of the lane semantic objects, the method can further include: performing vector conversion on all lane semantic objects included in the same road section, to convert the lane semantic objects into vector lines.

[0011] Optionally, grouping the lane semantic objects according to the single-trip vehicle driving direction can include:

[0012] performing attribute assignment on a collection direction of the lane semantic objects collected on both sides of the single-trip vehicle driving direction, based on the single-trip vehicle driving direction;

[0013] grouping the lane semantic objects based on the attribute of the collection direction of the lane semantic objects.

[0014] Optionally, the method can further include:

[0015] acquiring image data collected in multiple trips on the same road section, and calibrating a position relationship of lane semantic objects included in single-trip image data, respectively;

[0016] performing pose optimization on the calibrated lane semantic objects, to determine whether a width of a single lane semantic object included in the same road section and subjected to the pose optimization is greater than a preset width threshold.

[0017] In a second aspect, an embodiment of the present application provides a high-precision map making method, which can include:

[0018] clustering lane semantic objects included in each group as a grouping unit, to cluster into clusters;

[0019] fitting the clusters clustered into the clusters, to obtain a lane corresponding to the clusters;

[0020] merging lanes fitted from all lane semantic objects included in the same road section;

[0021] wherein the lane semantic objects included in each group are determined according to the lane semantic object grouping method of the first aspect.

[0022] Optionally, before merging the lane fitted by all the lane semantic objects included in the same road section, the following can also be included:

[0023] determining whether to merge the two lanes based on the vectorized point sequence direction, the collection orientation attribute and the height of the two lanes, and / or whether there is physical isolation between the two lanes;

[0024] if the vectorized point sequence direction, the collection orientation attribute and the height of the two lanes are the same, then merging the two lanes;

[0025] if the vectorized point sequence direction of the two lanes is different, and the collection orientation attribute and the height are the same, and there is no physical isolation, then merging the two lanes;

[0026] if the vectorized point sequence direction and the height of the two lanes are the same, and the collection orientation attribute is different, and there is no physical isolation, then merging the two lanes.

[0027] In a third aspect, an embodiment of the present application provides a lane semantic object grouping device, which can include:

[0028] a first determination module for determining whether the width of a single lane semantic object included in the same road section is greater than a preset width threshold;

[0029] a cutting module for cutting the single lane semantic object longitudinally to obtain at least two lane semantic objects if the first determination module determines that the width is greater than the preset width threshold;

[0030] a grouping module for traversing all lane semantic objects included in the same road section two by two, grouping the lane semantic objects according to the spatial position relationship of the lane semantic objects and / or the single-pass vehicle driving direction when the lane semantic objects are collected, and clustering the lane semantic objects in each group as a grouping unit.

[0031] In a fourth aspect, an embodiment of the present application provides a high-precision map production device, which can include:

[0032] a clustering module for clustering lane semantic objects in each group as a grouping unit to form clusters;

[0033] a fitting module for fitting the clusters to obtain lanes corresponding to the clusters;

[0034] a merging module for merging lanes fitted by all lane semantic objects included in the same road section;

[0035] wherein the lane semantic objects included in each group are determined according to the lane semantic object grouping method of the first aspect.

[0036] In a fifth aspect, an embodiment of the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the lane semantic object grouping method according to the first aspect, or implements the high-definition map making method according to the second aspect.

[0037] In a sixth aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the lane semantic object grouping method according to the first aspect, or implements the high-definition map making method according to the second aspect when executing the program.

[0038] In a seventh aspect, an embodiment of the present application provides a high-definition map making system, which can comprise: a data collection terminal and a lane semantic object grouping device according to the third aspect; wherein the data collection terminal is configured to collect lane semantic objects; and the lane semantic object grouping device is configured to group the lane semantic objects, and to cluster lane semantic objects included in each group as a grouping unit.

[0039] The beneficial effects of the above technical solutions provided by the embodiments of the present application at least include:

[0040] The lane semantic object grouping method, the high-definition map making method and the device provided by the embodiments of the present application can be used as a preprocessing scheme for clustering into clusters, and can effectively improve the accuracy of the clustering result and improve the fitting precision of clustering. On the one hand, the compatibility of unpredictable clustering results is reduced; on the other hand, the strong dependence on input conditions is reduced. Under the condition of ensuring extreme clustering results, the lane can pass through the clustering algorithm, and output the lane data that conforms to the actual situation. By solving the technical defects that the complex lane may be clustered incorrectly or cannot be fitted to the correct lane in the prior art, the data quality of the high-definition map is improved as a whole, and the risk of errors is reduced.

[0041] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the structure particularly pointed out in the written description and the accompanying drawings.

[0042] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings are intended to provide a further understanding of the present application, and constitute a part of the specification, and together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0044] Figure 1 Flow chart of grouping method of lane semantic object provided in embodiments of the present application;

[0045] Figure 2 Flow chart of detailed grouping method of lane semantic object provided in embodiments of the present application;

[0046] Figure 3 Schematic diagram of lane semantic object to be segmented provided in embodiments of the present application;

[0047] Figure 4 Effect diagram of lane semantic object after vectorization provided in embodiments of the present application;

[0048] Figure 5 Schematic diagram of grouping based on included angle between lane semantic objects provided in embodiments of the present application;

[0049] Figure 6 Schematic diagram of grouping based on elevation difference between lane semantic objects provided in embodiments of the present application;

[0050] Figure 7 Schematic diagram of grouping based on direction of vectorization point sequence between lane semantic objects provided in embodiments of the present application;

[0051] Figure 8 Schematic diagram of grouping lane semantic objects based on driving direction of single-pass vehicle provided in embodiments of the present application;

[0052] Figure 9 Schematic diagram of grouping lane semantic objects based on direction of vectorization point sequence between lane semantic objects and driving direction of single-pass vehicle provided in embodiments of the present application;

[0053] Figure 10 Figure 4 Effect diagram after grouping;

[0054] Figure 11 Structural schematic diagram of grouping device of lane semantic object provided in embodiments of the present application;

[0055] Figure 12 Structural schematic diagram of high-definition map making system provided in embodiments of the present application;

[0056] Figure 13 Flow chart of making method of high-definition map provided in embodiments of the present application;

[0057] Figure 14 Structural schematic diagram of making device of high-definition map provided in embodiments of the present application. DETAILED DESCRIPTION

[0058] ​Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms without being limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be thoroughly and completely understood, and so that the scope of the present disclosure can be conveyed completely to those skilled in the art.

[0059] The road surface marking data is generally road image information collected by a camera mounted on a vehicle, and is made by collecting map vector elements through identification of road surface markings. When the applicant makes a high-definition map based on multiple / multi-pass collected road surface markings, the main process is as follows: road surface marking position relationship calibration → calibrated road surface marking pose optimization → road surface marking clustering → cluster fitting into a single road surface marking. Due to the deviation of vehicle positioning signals, the speed of driving, the change of driving direction and other factors, the position accuracy error of the collected map lane semantic object is too large, and the shape is deformed. The applicant has previously proposed a series of patents to make a high-definition map. In the actual process of making a high-definition map, a group of semantic objects is clustered and grouped, and a group of semantic objects finally generates a road surface marking.

[0060] Taking lane semantic objects as an example, multi-vehicle multi-lane data is clustered and grouped based on position relationship, and then a group of semantic objects generates a lane data. However, in the actual production process, the possibility that a group of lane semantic objects may generate multiple lanes is not considered. For example: 1) The middle up and down lane lines are clustered into a group due to the close distance, and finally generate a lane in the fusion(fitting) process. However, in reality, it is two lanes, which leads to errors in the generated high-definition map data; 2) The lane at the diverging merge position is clustered into a group due to the distance relationship, and finally generates an error lane in the fusion(fitting) process. However, in reality, it is two lanes that branch left and right, which leads to the subsequent inability to use the high-definition map data; 3) The lane data at the elevated interchange position is clustered into a group due to the distance relationship, and finally generates an error lane in the fusion(fitting) process. However, in reality, it is two lanes that branch left and right, which leads to the subsequent inability to use the high-definition map data. In view of the above problems, the present application is proposed in order to provide a lane semantic object grouping method, a high-definition map making method and device which overcome the above problems or at least partially solve the above problems.

[0061] Embodiment one

[0062] In the embodiment one of the present application, a lane semantic object grouping method is provided. Referring to FIG. 1, the method can include the following steps: Figure 1

[0063] ​Step S11, judge whether the width of the single lane semantic object included in the same road section is greater than the preset width threshold; if yes, execute step S12.

[0064] Step S12, longitudinally cut the single lane semantic object to obtain at least two lane semantic objects.

[0065] Step S13, traverse all lane semantic objects included in the same road section two by two, group the lane semantic objects according to the spatial position relationship of the lane semantic objects and / or the single-pass vehicle driving direction when the lane semantic objects are collected, and cluster the lane semantic objects in each group as a grouping unit.

[0066] The above grouping method provided in the embodiments of the present application can be used as a preprocessing scheme for clustering into clusters, which can effectively improve the accuracy of the clustering result and improve the fitting precision of clustering. On the one hand, the compatibility of unpredictable clustering results is reduced; on the other hand, the strong dependence on the input condition is reduced. Under the condition of guaranteeing the extreme clustering result, the lane can pass through the clustering algorithm to output the lane data conforming to the actual situation. By solving the technical defects that the complex lane may be clustered incorrectly or cannot be fitted to the correct lane in the prior art, the data quality of the high-precision map is improved as a whole, and the error risk is reduced.

[0067] In one detailed embodiment, the embodiments of the present application provide a detailed grouping method of lane semantic objects, as shown in Figure 2 The method can include the following steps:

[0068] Step S21, acquire image data collected by multiple passes of the same road section, and respectively calibrate the position relationship of the lane semantic objects included in the single-pass image data.

[0069] The calibration of the position relationship of the lane semantic objects in this step is specifically as follows:

[0070] First step, based on the azimuth angle of the trajectory point of the collection vehicle, determine the longitudinal extension direction of a single calibration object (lane semantic object) in a single frame of image and the position relationship between adjacent calibration objects in the longitudinal extension direction, to construct the longitudinal extension relationship of the lane semantic object in the single frame of image. In the embodiments of the present application, the azimuth angle of the trajectory point of the collection vehicle is taken as the reference, and the azimuth angle of the trajectory point represents the driving direction of the collection vehicle, in other words. The inventor establishes the position relationship of the lane semantic object in the collected image data based on the driving direction of the collection vehicle when collecting.

[0071] Second step, based on the projection of the calibration object in the single frame image in the lateral direction, the lateral position relationship of the calibration object in the single frame image is determined to construct the lateral extension relationship of the lane semantic object in the single frame image. That is, after the longitudinal extension relationship of the lane semantic object in the single frame image is determined, a lateral vector perpendicular to the forward direction of the road surface (track line) is established, and the lateral position relationship of the calibration object is determined according to the projection position of the calibration object in the lateral direction. The lateral position relationship represents the left and right road surface conditions of the calibration object and other lane semantic objects in the same frame image, which can also lay a foundation for subsequent high-precision map making and accurate navigation.

[0072] Third step, based on the constructed longitudinal extension relationship and lateral extension relationship of the lane semantic object in the single frame image, the position correlation relationship of the lane semantic object in the single frame image is determined.

[0073] After the above first step of determining the longitudinal extension relationship of the lane semantic object and the second step of determining the lateral extension relationship of the lane semantic object, the position correlation relationship of the lane semantic object can be established in a coordinate system to accurately obtain the type, distance, direction and other position correlation relationships of the adjacent lane semantic objects in front, back, left and right of a lane semantic object. By establishing the position correlation relationship of the lane semantic object in the single frame image, the position of each lane semantic object in the single frame image and the relative position (distance, direction) relationship of the surrounding adjacent lane semantic objects can be clearly represented. Accurate relative position relationship information is provided for aligning multiple frames of data collected at the same place to lay a foundation for establishing a high-definition map. At the same time, the position correlation relationship of the lane semantic object can accurately guide the vehicle during navigation, and provides accurate protection for unmanned driving technology.

[0074] Step S22, pose optimization is performed on the calibrated lane semantic object to determine whether the width of the single lane semantic object after pose optimization included in the same road section is greater than a preset width threshold.

[0075] In this step, the pose optimization of the calibrated lane semantic object is performed as follows:

[0076] First step, obtaining single image data to be optimized. The obtained single image data to be optimized can be lane semantic objects and corresponding track point information processed by BA (bundle adjustment) algorithm after being collected by the collection vehicle. Or it can be image data obtained by a series of preprocessing to obtain the topological position relationship between lane semantic objects in single image data.

[0077] Second step, judge whether the actual road section corresponding to the single-trip image data to be optimized exists other single-trip image data; if there is one other single-trip image data, execute the third step; if there are at least two other single-trip image data, execute the fourth step.

[0078] The pose information of the trajectory and the lane semantic object obtained by repeatedly shooting the same road section by different vehicles and train numbers in the embodiment of the application is removed according to the maximum a posteriori estimation method. Unlike single trajectory cyclic detection, although it is also a process of "first assuming and then verifying and screening the consistency of the assumption", the key frame assumption is a matching assumption between lane semantic objects in different trajectories (different trips). The matching assumption refers to that the error of two lane semantic objects in two different image data is within the threshold range, and the observation value of the same actual lane semantic object is identified, and in each neighborhood data, the lane semantic object information with the matching relationship is mounted. It should be noted that if the actual road section does not exist one other single-trip image data, the single-trip image data to be optimized can only be corrected based on the inertial navigation, and the pose optimization between multiple trips cannot be realized, and the image data can only be saved as the first image data of the actual road section, and the pose optimization is performed after the subsequent single-trip image data is collected.

[0079] Third step, respectively extract the landmark lane semantic object in the single-trip image data to be optimized and other single-trip image data, and construct a matching error objective function based on the landmark lane semantic object. In this step, the single-trip image data to be optimized is called M, if there is one other single-trip image data, the single-trip image data is called N, and the matching is performed based on the landmark lane semantic object in M and N to construct a matching error objective function. The above landmark lane semantic object in the embodiment of the application is a lane semantic object with obvious features in image data, such as arrow, sidewalk, stop line and the like. The outer contour of such lane semantic object in the image is clear and obvious, which can provide a good reference for matching alignment and pose optimization.

[0080] Step 4: Obtain the cumulative error of at least two other single-trip image data sets and the fused landmark vehicle line semantic objects from at least two other single-trip image data sets. Construct a matching error objective function based on the fused landmark vehicle line semantic objects, the landmark vehicle line semantic objects extracted from the single-trip image data to be optimized, and the cumulative error. In this step, the single-trip image data to be optimized is referred to as M. If at least two other single-trip image data sets exist, they are referred to as N, H, I, etc. This step is an iterative step, that is, iterates based on the above step 3, performing optimization and fusion during the iteration process. Match the fused landmark vehicle line semantic objects from at least two other single-trip image data sets (N, H, I, etc.) with the landmark vehicle line semantic objects in M, and construct the matching error objective function based on the cumulative error from the previous iteration. In practical implementation, all the above data can be divided into single-trip data modules and multi-neighborhood data modules. The single-trip data module contains the single-trip image data to be optimized, which includes trajectory points and vehicle line semantic object information; while the multi-neighborhood data module only contains the vehicle line semantic object information after multiple trips are fused, and contains the structure of multiple vehicle line semantic objects to characterize its cumulative error.

[0081] Step 5: Optimize the pose of vehicle line semantic objects in the single-trip image data based on the matching error objective function. In this embodiment of the invention, during the pose optimization process, trajectories and vehicle line semantic objects obtained from repeated photographs of the same road segment by different vehicles and trips are used. A matching error objective function is constructed based on the matching relationship of iconic vehicle line semantic objects in different single-trip image data. An incremental update method is then used to optimize the pose of the vehicle line semantic objects in the single-trip image data to remove accumulated errors.

[0082] Step S23: Determine whether the width of the semantic object of a single lane included in the same road segment is greater than the preset width threshold; if so, proceed to step S24.

[0083] Reference Figure 3 As shown in the image, the red lane semantic object corresponds to two specific lanes on the actual road. However, due to data collection issues, this red lane semantic object was treated as a single lane semantic object, resulting in incorrect output during subsequent clustering and fitting steps. Therefore, this step first uses an overall width recognition algorithm for these overlapping lane semantic objects. If the horizontal width is too wide, for example, exceeding 20-40 cm, it is determined that vertical segmentation is necessary. Vertical segmentation is performed at the point of maximum width, converting the segmented object into two lanes. (Refer to...) Figure 3If the included angle between the two lines is greater than 5°, it is identified as a herringbone shape, that is, two lane lines are output by the semantic object of the adhered lane. The left and right sides of the adhered lane semantic object are cut to generate two lane semantic objects, and the purpose is to generate one corresponding lane for one lane semantic object.

[0084] Step S24, the single lane semantic object is longitudinally cut to obtain at least two lane semantic objects.

[0085] Step S25, all lane semantic objects included in the same road segment are vector converted to convert the lane semantic object into a vector line.

[0086] Referring to Figure 3 As described above, the lane semantic object is a surface object, and there are many inconveniences in subsequent data processing. In order to facilitate grouping in subsequent steps, the inventors convert the surface object into a vector line, which improves the efficiency of effective judgment in the subsequent grouping process. Referring to Figure 4 As shown, the vector line obtained after the lane semantic object is vector converted, the first aspect of the vector line has a vector direction, which can effectively judge the direction; the second aspect provides an effective basis for local lateral distance judgment in the subsequent clustering process; the third aspect reduces the number of lane semantic object vector points in the subsequent production of high-precision maps and fitting into a lane, and improves the mapping efficiency.

[0087] Step S26, all lane semantic objects included in the same road segment are traversed two by two, and the lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects and / or the single-vehicle driving direction when the lane semantic objects are collected, and the lane semantic objects included in each group are taken as a grouping unit for clustering.

[0088] This step is to group the lane semantic objects (based on the converted vector line), and in the grouping process, the lane semantic objects are divided into different groups based on the spatial position relationship of the lane semantic objects and / or the single-vehicle driving direction when the lane semantic objects are collected. In the embodiment of the application, the lane semantic objects are divided according to the specific spatial position relationship and single-vehicle driving direction, which will be described in detail as follows:

[0089] <Method one> Grouping lane semantic objects according to spatial position relationship

[0090] In the grouping manner, the lane semantic objects are grouped according to the included angle between two lane semantic objects, the elevation difference between two lane semantic objects and / or the direction of the vectorized point sequence of two lane semantic objects; if the included angle between two lane semantic objects is greater than a preset included angle threshold, the elevation difference between two lane semantic objects is greater than a preset elevation difference threshold and / or the direction of the vectorized point sequence of two lane semantic objects is inconsistent, the two lane semantic objects are divided into different groups; otherwise, the two lane semantic objects are divided into the same group.

[0091] 1.1 Grouping based on the included angle of lane semantic objects

[0092] The head and tail of the lane semantic objects after the vector conversion in the above step S25 are connected, and then two-by-two traversal is performed to determine the included angle between the straight lines formed by the head and tail connection of two-by-two lane semantic objects; if the included angle is greater than a preset included angle threshold, the lane semantic objects are divided into two groups; if the included angle between the straight lines formed by the head and tail connection of two-by-two lane semantic objects is less than a preset included angle threshold, the lane semantic objects are divided into one group. Referring to Figure 5 , Figure 5 The straight lines of the head and tail connection of the two lines in red exist an included angle, and the divergent confluence cannot be divided into one group. However, when the existing method is processed, the position is likely to be divided into one group, and the existing divergent confluence road condition cannot be clearly determined in the subsequent high-precision map clustering process; and the technical problem is successfully solved by the grouping manner in the embodiment of the application.

[0093] 1.2 Grouping based on the elevation difference of lane semantic objects

[0094] Two-by-two traversal is performed on the lane semantic objects after the vector conversion in the above step S25 to determine the elevation difference between two-by-two lane semantic objects; if the elevation difference value is greater than a preset elevation difference threshold, the lane semantic objects are divided into two groups and are fitted respectively; if the elevation difference between two lane semantic objects is less than a preset elevation difference threshold, the lane semantic objects are divided into one group. Referring to Figure 6 , Figure 6 The two lines in red may be an elevated uphill and an elevated downhill, there is a height difference, which belongs to different spatial planes, and is not the same lane on the actual road, and should be divided into different groups. However, when the traditional method is processed, the position is likely to be divided into one group, and the existing elevated road condition cannot be clearly determined in the clustering.

[0095] 1.3 Grouping based on the direction of the vectorized point sequence of lane semantic objects

[0096] In the actual operation process, two vehicle lanes may be divided into a group due to the close distance, and then become a cluster in clustering. At this time, only one vehicle lane is output in the clustering output, but there should be two vehicle lanes at the actual position, and the output of the vehicle lane does not match the actual situation. In view of the above situation, the embodiment of the present application divides the vehicle lane semantic objects with consistent vectorization point sequence directions into a group by identifying the vectorization point sequence directions of the vehicle lane semantic objects two by two, and respectively performs fitting based on the grouping result. Referring to Figure 7 As shown in the figure, the two vehicle lanes framed in the middle of the road are actually two vehicle lanes in the middle of the road, and the left and right lanes are the above-mentioned two vehicle lane semantic objects collected by the uplink and downlink lanes, which belong to two lanes respectively, and have direction attributes. The vehicle compartment with direction attributes on the generated high-precision map is convenient for the use of data in the process of unmanned driving. Therefore, Figure 7 If the vectorization point sequence directions of the two vehicle lane semantic objects in the middle are inconsistent, the two vehicle lane semantic objects are divided into different groups.

[0097] <Method two> Grouping vehicle lane semantic objects according to the single-vehicle driving direction when collecting vehicle lane semantic objects

[0098] In this grouping method, the collection positions of the vehicle lane semantic objects collected on both sides of the single-vehicle driving direction are attribute valued based on the single-vehicle driving direction; then, the vehicle lane semantic objects are grouped based on the collection position attributes of the vehicle lane semantic objects.

[0099] In the actual operation process, two vehicle lanes may be divided into a group due to the close distance, and then become a cluster in clustering. At this time, only one vehicle lane is output in the clustering output, but there should be two vehicle lanes at the actual position. For example, there are double yellow solid lines in the middle of the road, and the output of the vehicle lane does not match the actual situation.

[0100] In the above method two provided in the embodiment of the present application, the vehicle lane collected on the right side of the driving direction of each vehicle is divided into a group, and the vehicle lane collected on the left side of the driving direction is divided into a group. For example, referring to Figure 8As shown, for vehicle A, its travel direction is from bottom to top. The lanes collected to its left are grouped together. Based on the grouping results, fitting can be performed to obtain a pre-clustering result, which is a single lane A'. For vehicle B, its travel direction is exactly the opposite, from top to bottom. The lanes collected to its left are also grouped together. Based on the grouping results, fitting can also be performed to obtain a pre-clustering result, which is also a single lane B'. Now, based on the same vehicle, two pre-clustering results are obtained. Comparing the pre-clustered lanes A' and B', if the distance between them is less than a certain threshold and their vector directions are opposite, it can be determined that the pre-clustering result of this cluster should be two lanes. If clustering is performed directly based on the lanes collected at this location, only a single lane clustering result is obtained. The grouping method corrects the defects in the existing technology, allowing the improved clustering result to match the actual road lane situation.

[0101] <Method 3> Grouping based on the common factors of vectorized point order direction between vehicle lane semantic objects and the travel direction of a single vehicle.

[0102] In actual operation, there may be situations where two lanes are grouped together due to their proximity, thus forming a cluster during clustering. In this case, the clustering output will only show one lane, but there should actually be two lanes at that location. The output lanes do not match the actual road conditions. For example... Figure 9 There are two lanes on adjacent roads traveling in the same direction. (Refer to...) Figure 9 As shown in the embodiment of the present invention, if the direction of travel of a single vehicle is the same, it is determined whether the direction of the vectorized point sequence is the same. If they are all the same, they are divided into the same group. If the direction of travel of a single vehicle is different, even if the direction of the vectorized point sequence is the same, they are divided into two groups.

[0103] Reference Figure 10 As shown, Figure 4 The resulting image shows the grouped vehicle lane semantic objects, distinguished by color into two groups, which are then clustered and fitted separately. The detailed vehicle lane semantic object grouping method provided in this embodiment addresses potential clustering errors by proposing an improvement scheme. Clusters that may generate clustering errors are further grouped, and the grouping results are then fitted and merged as preprocessing before clustering to improve the accuracy of the clustering results and enhance the precision of downstream applications. This solves the technical shortcomings of existing technologies where complex vehicle lanes may be clustered incorrectly or unable to fit the correct lanes, thus improving the overall data quality of high-precision maps and reducing the risk of errors.

[0104] Based on the same inventive concept, this invention also provides a grouping device for vehicle lane semantic objects, referring to... Figure 11As shown, the apparatus can comprise: a first judging module 111, a cutting module 112 and a grouping module 113, which work as follows:

[0105] The first judging module 111 is configured to judge whether the width of a single lane semantic object included in the same road section is greater than a preset width threshold value;

[0106] If the first judging module 111 judges yes, the cutting module 112 performs longitudinal cutting on the single lane semantic object to obtain at least two lane semantic objects;

[0107] The grouping module 113 is configured to traverse all lane semantic objects included in the same road section two by two, and group the lane semantic objects according to the spatial positional relationship of the lane semantic objects and / or the single-vehicle driving direction when the lane semantic objects are collected, and cluster the lane semantic objects in each group as a grouping unit.

[0108] In an optional embodiment, the grouping module 113 is specifically configured to group the lane semantic objects according to the included angle, the elevation difference and / or the vectorization point sequence direction between two lane semantic objects;

[0109] If the included angle between the two lane semantic objects is greater than a preset included angle threshold value, the elevation difference between the two lane semantic objects is greater than a preset elevation difference threshold value, and / or the vectorization point sequence direction between the two lane semantic objects is inconsistent, the two lane semantic objects are divided into different groups; otherwise, they are divided into the same group.

[0110] In another optional embodiment, referring to Figure 11 As shown, the apparatus can further comprise a vector conversion module 114, which is configured to perform vector conversion on all lane semantic objects included in the same road section to convert the lane semantic objects into vector lines.

[0111] In an optional embodiment, the grouping module 113 is specifically further configured to attribute the collection orientation of the lane semantic objects collected on both sides of the single-vehicle driving direction according to the single-vehicle driving direction;

[0112] Group the lane semantic objects according to the collection orientation attribute of the lane semantic objects.

[0113] In another optional embodiment, referring to Figure 11 As shown, the apparatus can further comprise an acquisition module 115, a position calibration module 116 and a pose optimization module 117, which work as follows:

[0114] The acquisition module 115 is configured to acquire image data collected in multiple passes of the same road section;

[0115] The position calibration module 116 is configured to calibrate the position relationship of the lane semantic object included in the single image data respectively;

[0116] The pose optimization module 117 is configured to perform pose optimization on the calibrated lane semantic object, so that the first determination module 111 determines whether the width of the single lane semantic object included in the same road section after the pose optimization is greater than a preset width threshold.

[0117] Based on the same inventive concept, the present embodiment further provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the lane semantic object grouping method.

[0118] Based on the same inventive concept, the present embodiment further provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the lane semantic object grouping method.

[0119] A high-precision map making system, referring to Figure 12 The system can comprise a data acquisition terminal 1 and the lane semantic object grouping device; wherein the data acquisition terminal 1 is configured to acquire lane semantic objects; and the grouping device is configured to group the lane semantic objects, and cluster the lane semantic objects included in each group as a grouping unit.

[0120] The principles of the problems solved by the above-mentioned device, medium, related equipment and system in the present embodiment are similar to those of the foregoing method, so the implementation thereof can be referred to the implementation of the foregoing method, and the repeated parts will not be described herein.

[0121] Embodiment two

[0122] The present embodiment two provides a high-precision map making method, referring to Figure 13 The method can comprise the following steps:

[0123] Step S131, clustering the lane semantic objects included in each group as a grouping unit to form clusters. The lane semantic objects included in each group are determined according to the lane semantic object grouping method described in embodiment one.

[0124] Step S132, fitting the clusters formed by clustering to obtain the lane corresponding to the cluster.

[0125] In another optional embodiment, before merging the lanes fitted by all lane semantic objects included in the same road section, the following steps can be further included:

[0126] Step S133, based on the vectorization point sequence direction of the two lane lines, the collection azimuth attribute and the height, and / or whether there is physical isolation between the lane lines, it is judged whether the two lane lines are merged; if the vectorization point sequence direction of the two lane lines, the collection azimuth attribute and the height are all the same, step S134 is executed; if the vectorization point sequence direction of the two lane lines is different, and the collection azimuth attribute and the height are the same, and there is no physical isolation, step S134 is executed; if the vectorization point sequence direction and the height of the two lane lines are the same, and the collection azimuth attribute is different, and there is no physical isolation, step S134 is executed.

[0127] Step S134, all lane lines semantic objects included in the same road section are fitted into lane lines and merged.

[0128] The above lane line merging in the embodiment of the application can be performed in the manner of breaking at the intersection of the two lane lines and connecting the head and tail, and in the manner of projection and eliminating the redundant part when the two lane lines do not intersect, and the above merging process is not specifically limited in the embodiment of the application. It should be noted that the connection and the preset distance extended from the connection can be smoothed after merging, so as to achieve the effect of lane line smoothing.

[0129] Based on the same inventive concept, the embodiment of the application further provides a high-precision map making device, as shown in Figure 14 The device can include a clustering module 141, a fitting module 142, a second judgment module 143 and a merging module 144, and the working principle is as follows:

[0130] The clustering module 141 is used for clustering in each grouping including lane line semantic objects as a grouping unit, to cluster into clusters;

[0131] The fitting module 142 is used for fitting the clusters clustered into the lane lines corresponding to the clusters;

[0132] The second judgment module 143 is used for judging whether the two lane lines are merged based on the vectorization point sequence direction of the two lane lines, the collection azimuth attribute and the height, and / or whether there is physical isolation between the lane lines;

[0133] In the second judgment module 143, the vectorization point sequence direction of the two lane lines, the collection azimuth attribute and the height are all the same, in the second judgment module 143, the vectorization point sequence direction of the two lane lines is different, and the collection azimuth attribute and the height are the same, and there is no physical isolation, and in the second judgment module 143, the vectorization point sequence direction and the height of the two lane lines are the same, and the collection azimuth attribute is different, and there is no physical isolation, the merging module 144 is used for merging the lane lines fitted from all lane line semantic objects included in the same road section.

[0134] Based on the same inventive concept, the embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the method for manufacturing the high-precision map.

[0135] Based on the same inventive concept, the embodiment of the present application further provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the program to realize the method for manufacturing the high-precision map.

[0136] The above device, medium, related device and the principle for solving problems in the embodiment of the present application are similar to the foregoing method, and therefore the implementation can be referred to the implementation of the foregoing method, and the redundant parts will not be described herein.

[0137] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can be in the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware. Moreover, the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

[0138] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device implemented in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flow(s) or block(s).

[0139] These computer program instructions can also be stored in a computer readable memory to guide the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including an instruction apparatus, which implements the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flow(s) or block(s).

[0140] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable devices provide processes for implementing the functions specified in the flowchart Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps of the functions specified in the flowchart

[0141] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method of grouping lane semantic objects, characterized in that, The method comprises: determining whether the width of a single lane semantic object included in the same road section is greater than a preset width threshold; if yes, the single lane semantic object is vertically divided to obtain at least two lane semantic objects; all lane semantic objects included in the same road section are traversed two by two, and the lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects and / or the single-vehicle driving direction when the lane semantic objects are collected, and the lane semantic objects included in each group are taken as a grouping unit for clustering.

2. The method of claim 1, wherein, The lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects, which comprises: the lane semantic objects are grouped according to the included angle, elevation difference and / or vectorized point sequence direction between the two lane semantic objects; if the included angle between the two lane semantic objects is greater than a preset included angle threshold, the elevation difference between the two lane semantic objects is greater than a preset elevation difference threshold, and / or the vectorized point sequence directions of the two lane semantic objects are inconsistent, the two lane semantic objects are divided into different groups; otherwise, the two lane semantic objects are divided into the same group.

3. The method according to claim 1 or 2, characterized in that, Before the lane semantic objects are grouped according to the spatial position relationship of the lane semantic objects, the lane semantic objects included in the same road section are further subjected to vector conversion to convert the lane semantic objects into vector lines.

4. The method of claim 1, wherein, The lane semantic objects are grouped according to the single-vehicle driving direction when the lane semantic objects are collected, which comprises: the collection orientations of the lane semantic objects collected on both sides of the single-vehicle driving direction are subjected to attribute assignment based on the single-vehicle driving direction; the lane semantic objects are grouped based on the collection orientation attribute of the lane semantic objects.

5. The method of claim 1, wherein, Further comprising: acquiring image data collected in multiple passes of the same road section, and calibrating the position relationship of the lane semantic objects included in the single-pass image data respectively; performing pose optimization on the calibrated lane semantic objects to determine whether the width of a single lane semantic object included in the same road section after pose optimization is greater than a preset width threshold.

6. A method of producing a high-definition map, characterized by, The method comprises: clustering the lane semantic objects included in each group as a grouping unit to form clusters; fitting the clusters to obtain the lane corresponding to the clusters; merging the lanes fitted from all lane semantic objects included in the same road section; wherein the lane semantic objects included in each group are determined according to the lane semantic object grouping method of any one of claims 1-5.

7. The method of claim 6, wherein, Before the lanes fitted from all lane semantic objects included in the same road section are merged, further comprising: determining whether two lanes are merged based on the vectorized point sequence direction, collection orientation attribute and height of the two lanes, and / or whether there is physical isolation between the two lanes; if the vectorized point sequence direction, collection orientation attribute and height of the two lanes are all the same, the two lanes are merged; if the vectorized point sequence direction of the two lanes is different, and the collection orientation attribute and the height are the same, and there is no physical isolation, the two lanes are merged; If the directions of the vectorized point sequences of the two vehicle lanes are the same and the heights are the same, and the collection orientation attributes are different, and there is no physical isolation, the two vehicle lanes are merged.

8. A lane semantic object grouping apparatus, comprising: The method comprises the following steps: The first judging module is configured to judge whether the width of a single vehicle lane semantic object included in the same road section is greater than a preset width threshold value; The segmentation module is configured to perform longitudinal segmentation on the single vehicle lane semantic object to obtain at least two vehicle lane semantic objects if the first judging module judges that the width of the single vehicle lane semantic object is greater than the preset width threshold value; The grouping module is configured to traverse all vehicle lane semantic objects included in the same road section two by two, and group the vehicle lane semantic objects according to the spatial position relationship of the vehicle lane semantic objects and / or the single-vehicle-traveling direction when the vehicle lane semantic objects are collected, and cluster the vehicle lane semantic objects in each group as a grouping unit. 9.A device for producing a high-definition map, comprising: The method comprises the following steps: The clustering module is configured to cluster the vehicle lane semantic objects in each group as a grouping unit to form clusters; The fitting module is configured to fit the clusters to obtain vehicle lanes corresponding to the clusters; The merging module is configured to merge the vehicle lanes fitted from all vehicle lane semantic objects included in the same road section; The vehicle lane semantic objects included in each group are determined according to the grouping method of the vehicle lane semantic object in any one of claims 1-5.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the grouping method of the vehicle lane semantic object in any one of claims 1-5, or implement the method for making a high-definition map in claim 6 or 7.

11. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the grouping method of the vehicle lane semantic object in any one of claims 1-5, or implement the method for making a high-definition map in claim 6 or 7.

12. A high-definition map making system, comprising: The method comprises the following steps: The data acquisition terminal and the grouping device of the vehicle lane semantic object in claim 8 are included; the data acquisition terminal is configured to collect vehicle lane semantic objects; and the grouping device is configured to group the vehicle lane semantic objects and cluster the vehicle lane semantic objects in each group as a grouping unit.

Images