Map processing method, map processing device, and computer program product

Abstract

The embodiment of the application discloses a kind of map processing method, map processing device and computer program product, it is related to map data technical field.Therein method includes: based on target trajectory data, generate feature image, target trajectory data includes multiple trajectories, feature image at least characterizes the position information, density information and direction information of multiple trajectories, can further characterize speed information.Then, feature image is handled, and lane line vector data is obtained, and lane line vector data includes the position information, direction information and type information of each lane line.Finally, based on lane line vector data, lane level map is obtained.The lane level map of the application is obtained by trajectory data, to realize the construction or update of low-cost lane level map, provide basis for wider coverage and more timely update of lane level map, to further bring better navigation experience for user using lane level map, improve driving safety simultaneously.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of map data, and in particular to a map processing method, a map processing device and a computer program product. BACKGROUND

[0002] Lane-level map is a key component of navigation system and automatic driving system. At present, lane-level map is usually constructed by relying on high-precision laser radar, camera, combined navigation and other devices, and the data acquisition, transmission and storage cost is high, so it is difficult to construct lane-level map in a large area. In addition, the road will change due to road repair, road change and other reasons, and the high-cost construction method is difficult to ensure the timely update of the map, so that the road information in the lane-level map is not accurate enough, thereby affecting the navigation experience of users and driving safety. SUMMARY

[0003] Therefore, the present application provides a map processing method, a map processing device and a computer program product to obtain lane-level map in a low-cost manner.

[0004] The present application provides the following solutions:

[0005] According to a first aspect, a map processing method is provided, the method comprising: generating a feature image based on target trajectory data, the target trajectory data comprising a plurality of trajectories, the feature image at least representing position information, density information and direction information of the plurality of trajectories; processing the feature image to obtain lane line vector data, the lane line vector data comprising position information, direction information and type information of each lane line; and obtaining a lane-level map based on the lane line vector data.

[0006] According to a second aspect, a map processing device is provided, the device comprising:

[0007] An image generation unit configured to generate a feature image based on target trajectory data, the target trajectory data comprising a plurality of trajectories, the feature image at least representing position information, density information and direction information of the plurality of trajectories;

[0008] A vector generation unit configured to process the feature image to obtain lane line vector data, the lane line vector data comprising position information, direction information and type information of each lane line;

[0009] A map processing unit configured to obtain a lane-level map based on the lane line vector data.

[0010] According to a third aspect, a computer program product is provided, the computer program being executed by a processor to implement the steps of the method of any one of the first aspect.

[0011] According to the specific embodiments provided in the present application, the present application discloses the following technical effects:

[0012] The present application solves the problem that the construction of the lane-level map usually depends on high-precision devices. The present application can obtain the lane-level map through the trajectory data, so as to obtain the lane-level map in a low-cost manner, provide a basis for wider coverage and more timely update of the lane-level map, and thus bring a better navigation experience for users using the lane-level map, while improving the driving safety.

[0013] In addition, due to the low cost of the scheme provided in the present application, on the one hand, it makes it possible to construct a large range of lane-level maps, and on the other hand, it is also conducive to the timely update of the map. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0015] Figure 1 The flowchart of the map processing method provided for the embodiments of the present application.

[0016] Figure 2a The schematic diagram of a first image provided for the embodiments of the present application.

[0017] Figure 2b The schematic diagram of a second image provided for the embodiments of the present application.

[0018] Figure 3a The schematic diagram of another first image provided for the embodiments of the present application.

[0019] Figure 3b The schematic diagram of another second image provided for the embodiments of the present application.

[0020] Figure 4a The schematic diagram of the division processing unit provided for the embodiments of the present application.

[0021] Figure 4b The schematic diagram of the division grid provided for the embodiments of the present application.

[0022] Figure 5 The process schematic diagram of obtaining lane line vector data in the embodiments of the present application.

[0023] Figure 6a The schematic diagram of a lane line vector data in the embodiments of the present application.

[0024] Figure 6b FIG. 6 is a schematic diagram of another lane line vector data in an embodiment of the present application.

[0025] Figure 7 FIG. 7 is a schematic diagram of a flow of processing a lane-level map in an embodiment of the present application.

[0026] Figure 8 FIG. 8 is a schematic block diagram of a map processing apparatus provided in an embodiment of the present application.

[0027] Figure 9 FIG. 9 is a schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art belong to the scope of protection of the present application.

[0029] The terms used in the embodiments of the present application are merely for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms “a”, “an” and “the” used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0030] It should be understood that the term “and / or” used herein is merely to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character “ / ” herein generally represents an “or” relationship between the front and rear associated objects.

[0031] Depending on the context, the word “if” as used herein can be interpreted as meaning “when” or “upon” or “in response to a determination” or “in response to a detection”. Similarly, depending on the context, the phrase “if determined” or “if detecting (a stated condition or event)” can be interpreted as meaning “when determined” or “in response to a determination” or “when detecting (a stated condition or event)” or “in response to a detection (a stated condition or event)”.

[0032] In order to reduce the implementation cost on the basis of ensuring accuracy, the present application provides a new idea. Figure 1 FIG. 10 is a flowchart of a map processing method provided in an embodiment of the present application. The method can be executed by a map processing apparatus, which can be arranged on a server end or a computer terminal with strong computing power. As shown in FIG. 10, the method can include the following steps: Figure 1 FIG. 10 is a flowchart of a map processing method provided in an embodiment of the present application. The method can be executed by a map processing apparatus, which can be arranged on a server end or a computer terminal with strong computing power. As shown in FIG. 10, the method can include the following steps:

[0033] In step 101, a feature image is generated based on target trajectory data, the target trajectory data including a plurality of trajectories, and the feature image at least representing position information, density information and direction information of the plurality of trajectories.

[0034] In step 102, the feature image is processed to obtain lane line vector data, the lane line vector data including position information, direction information and type information of each lane line.

[0035] In step 103, a lane-level map is obtained based on the lane line vector data.

[0036] As can be seen from the above process, the present application solves the problem that the construction of a lane-level map usually relies on high-precision equipment, resulting in high production cost. The present application can obtain a lane-level map through trajectory data, thereby obtaining a lane-level map in a low-cost manner, providing a basis for wider coverage and more timely updating of the lane-level map, and further providing a better navigation experience for users using the lane-level map, while improving driving safety.

[0037] In addition, due to the low cost of the scheme provided by the present application, on the one hand, it makes it possible to process a large range of lane-level maps, and on the other hand, it is also conducive to the timely updating of the map.

[0038] The steps in the above process and the effects that can be further produced will be described in detail below in conjunction with embodiments. It should be noted that the "first", "second" and the like involved in the present disclosure do not have the limitation of size, order and quantity, and are only used to distinguish in name, for example, "first image" and "second image" are used to distinguish two images.

[0039] First, the step 101, i.e., "based on target trajectory data, generating a feature image, the target trajectory data including a plurality of trajectories, and the feature image at least representing position information, density information and direction information of the plurality of trajectories", will be described in detail in conjunction with an embodiment.

[0040] The present application generates a feature image based on target trajectory data, wherein the trajectory data refers to data information obtained by sampling the movement process of one or more moving objects in a space-time environment. In the present application, the trajectory data includes a plurality of trajectories, each trajectory is composed of a set of discrete three-dimensional trajectory points at equal time intervals, and each trajectory point contains the longitude and latitude coordinates and altitude information of the trajectory point. For example, each trajectory point is usually represented by a triple (x, y, h), where x and y represent the longitude and latitude of the trajectory point, respectively, and h represents the altitude information. The feature image generated by the present application at least represents the position information, density information and direction information of the plurality of trajectories contained in the trajectory data.

[0041] As one of the implementable ways, in order to help the driving user to navigate or automatically drive to generate a lane-level map, the mobile object can be a vehicle, and the collected trajectory data can be trajectory data of the vehicle on the road. In the present application, the trajectory data of the vehicle can be obtained by collecting the position data of the vehicle, or by collecting the position data of the user terminal such as a mobile phone, a tablet computer and the like used by the driving user, or by collecting the navigation data in the navigation application used by the driving user through the user terminal as the trajectory data of the vehicle, or by collecting the navigation data in the navigation of the vehicle self-equipped as the trajectory data of the vehicle, and the like.

[0042] Before generating the feature image based on the target trajectory data, the present application can first screen the target trajectory data. This is because the trajectory data is usually collected for all the roads in the region of the map to be processed, resulting in a large amount of initial trajectory data. In order to make the generated feature image accurately represent the position information, density information and direction information of the plurality of trajectories contained in the target trajectory data, the present application can screen the initial trajectory data to eliminate the trajectories in the initial trajectory data that have abnormal conditions such as jumping and drifting.

[0043] For example, if a trajectory is composed of 10 trajectory points, the time intervals of the 10 trajectory points are the same and known, the speed of the vehicle at each trajectory point on the trajectory can be calculated according to the latitude and longitude coordinates and the altitude information contained in each trajectory point. For example, the distance between two trajectory points can be calculated by using the latitude and longitude coordinates and the altitude information of the two trajectory points, and then the speed of the vehicle at each trajectory point on the trajectory can be calculated by using the distance between the two trajectory points and the time interval between the two trajectory points. If the calculated speed is greater than a preset speed threshold (for example, the speed of the vehicle passing through each trajectory point is greater than 200 kilometers per hour), it is proved that the trajectory is abnormal, and the trajectory is eliminated from the initial trajectory data. For another example, generally, each trajectory in the trajectory data should be a straight line or a smooth curve, and the angle between adjacent line segments is greater than a preset angle threshold, which proves that the trajectory jumps and is eliminated from the initial trajectory data. For another example, if a trajectory point suddenly drifts a certain distance away from the previous trajectory point, it is proved that the trajectory drifts and is eliminated from the initial trajectory data.

[0044] By screening a large amount of initial trajectory data, a certain number of trajectories are screened for each road. Then, the present application obtains the target trajectory data by using the screened trajectory data, and generates the feature image according to the target trajectory data.

[0045] It should be noted that the feature image generated based on the trajectory data can include only one image, referred to as a first image in the embodiments of the present application, or can include two images, referred to as a first image and a second image, wherein the first image is used to represent the position information, density information and direction information of the plurality of trajectories included in the trajectory data, and the second image is used to represent the speed information of the plurality of trajectories included in the trajectory data. By introducing the speed information of the plurality of trajectories included in the second image, the perception accuracy of the stop line and the like can be improved, so that the generated lane line vector data is more accurate and comprehensive, and a more accurate and comprehensive lane-level map is obtained.

[0046] In the embodiments of the present application, if it is not necessary to determine the lane line of the category such as the stop line (for example, for the expressway, it is not necessary to determine the stop line), the feature image can include only the first image. If it is necessary to determine the stop line, the feature image can include the first image and the second image.

[0047] In generating the feature image based on the target trajectory data, since the target trajectory data represents three-dimensional information (the trajectory points include altitude), the target trajectory data can be first mapped to a two-dimensional plane under a bird's eye view (BEV, Bird's Eye View), the trajectory data of each grid on the two-dimensional plane is counted using the mapped trajectory data, and the feature image is rendered based on the counting result and the position information of each grid. The counting result of the grid at least includes the number and direction information of the trajectories passing through the grid, and the size of the grid can be 0.15m*0.15m. The bird's eye view can reduce visual occlusion and more clearly show the spatial relationship between different objects. The position information, density information and direction information of the trajectories in the trajectory data can also be more clearly represented through the trajectory data under the bird's eye view. However, in addition to the bird's eye view, other views can also be used. Moreover, the feature image of the target trajectory data is refined in the present application, which further enhances the capturing ability of local features, so that more local detailed information can be considered when predicting the lane line vector data, thereby laying a foundation for obtaining a lane-level map with high accuracy later.

[0048] As one of the implementable manners, the first image corresponding to the target trajectory data can be rendered according to the counting result and the position information of each grid on the two-dimensional plane, wherein the counting result of the grid at least includes the number and direction information of the trajectories passing through the grid.

[0049] As another implementable manner, the first image and the second image corresponding to the target trajectory data can be respectively rendered according to the counting result and the position information of each grid on the two-dimensional plane, wherein the counting result of the grid includes the number, direction information and speed information of the trajectories passing through the grid.

[0050] For each grid, the number of trajectories passing through is added by 1, and finally the number of trajectories passing through is counted. The position of the grid embodies the position information, and the number of trajectories passing through the grid embodies the density information. When determining the direction information of the trajectory passing through the grid, the direction value of each trajectory point passing through the grid can be calculated first, and then the direction information of the trajectory passing through the grid is obtained by averaging the direction value of each trajectory point. The direction value of the trajectory point a can be embodied as the direction value of the previous trajectory point of the trajectory point a pointing to the trajectory point a.

[0051] It should be noted that, on the one hand, the direction information helps to distinguish the direction of the lane; on the other hand, the direction information can reflect the degree of curvature of the road, helping to predict the lane line in the sparse trajectory scene; therefore, the direction information is introduced into the feature image in the embodiment of the present application, which can effectively ensure the prediction accuracy of the lane line vector data.

[0052] When determining the speed information of the trajectory passing through the grid, the speed value of each trajectory point passing through the grid can be calculated first, and then the speed information of the trajectory passing through the grid is obtained by averaging the speed value of each trajectory point.

[0053] For the rendering process of the first image, the present application can map each grid to each pixel region of the first image according to the position information of each grid, and map the number of trajectories passing through each grid to the first channel of the multi-channel color space adopted by the first image, and further map the direction information of the trajectory passing through each grid to the second channel of the multi-channel color space, to obtain the first image. In this way, different information contained in the target trajectory data can be mapped to different channels in the multi-channel color space to obtain the first image, which is convenient for subsequent analysis and research on the first image corresponding to the target trajectory data.

[0054] The multi-channel color space can be an RGB (red, green, blue) color space, or an HSV (hue, saturation, lightness) color space, and the present application does not make specific limitation, as long as the method in the present application can be implemented. The present application provides a more preferred embodiment, the multi-channel color space can be HSV, and when the first image adopting HSV is obtained, the first image adopting HSV is converted into the first image adopting RGB, and the conversion process is lossless.

[0055] The HSV is a commonly used color representation, especially suitable for computer image processing, design and visual arts, etc. The HSV color space is named after its three components: Hue, Saturation, and Value. Hue represents the type of color, usually in the form of an angle (ranging from 0 to 360 degrees), for example, red (which can correspond to 0 degrees), green (which can correspond to 120 degrees), blue (which can correspond to 240 degrees), etc. Saturation represents the purity or intensity of the color, ranging from 0 to 1 (or 0% to 100%), with a saturation of 0 indicating no color (gray), and a saturation of 1 indicating a fully saturated color, i.e. a pure color. Increasing the saturation makes the color appear more vibrant, while decreasing the saturation makes the color appear more gray or soft. Value represents the brightness or intensity of the color, ranging from 0 to 1 (or 0% to 100%), with a value of 0 indicating black, and a value of 1 indicating the maximum brightness of the color, with intermediate values possibly resulting in different brightness effects.

[0056] Since the HSV color space is more consistent with human perception of color compared to the RGB color space, the present application first uses HSV to represent the first image. And in the rendering of the first image, it is easier and more intuitive to use HSV for color adjustment (such as color filtering or color replacement) than to operate directly in the RGB color space. When the lane line map needs to be obtained from the first image later, the first image in HSV needs to be converted to the first image in RGB for easier visual analysis.

[0057] For the rendering process of the second image, the present application can map each grid to a pixel area of the second image according to the position information of each grid, and map the speed information of the trajectory passed by each grid to the single-channel color space adopted by the second image, to obtain the second image. This way, the speed information contained in the target trajectory data can be mapped to the single-channel color space to obtain the second image, which is convenient for subsequent analysis and research of the second image corresponding to the target trajectory data.

[0058] Taking the multi-channel color space as the HSV color space and the single-channel color space as the grayscale space as an example. The HSV color space is composed of hue (H), saturation (S), and value (V), and the grayscale space is a single-channel image with only one gray value channel. For the target trajectory data on the highway, the direction information of the trajectory passed by each grid on the two-dimensional plane corresponding to the target trajectory data is mapped to the H channel of the HSV color space, and the number of the trajectory passed by each grid on the two-dimensional plane corresponding to the target trajectory data is mapped to the V channel of the HSV color space, and the S channel is set to a preset value, to obtain the first image corresponding to the target trajectory data in the HSV color space, as shown inFigure 2a As shown in FIG. 6, since the sensitivity of human eyes to saturation is relatively low compared to chroma and luminance, only the H channel and the V channel are used when mapping the direction information and the density information (i.e., the number of trajectories passing through each grid), and the S channel is set to a preset value, i.e., the direction and density features are more significantly reflected by the H channel and the V channel.

[0059] The speed information of the trajectories passing through each grid on the two-dimensional plane corresponding to the target trajectory data is filtered to retain only the speed information below a preset speed threshold (e.g., 20 km / h), and is mapped into the gray value channel of the gray space to obtain a second image in the gray space corresponding to the target trajectory data, as shown in FIG. 7. Figure 2b As shown in FIG. 7, since the speed of vehicles on the highway is rarely below 20 km / h, it can be seen from FIG. 7 that almost all the pixels are white. Figure 2b

[0060] For a target trajectory data containing a crossroads, the direction information of the trajectories passing through each grid on the two-dimensional plane corresponding to the target trajectory data is mapped onto the H channel of the HSV color space, and the number of trajectories passing through each grid on the two-dimensional plane corresponding to the target trajectory data is mapped onto the V channel of the HSV color space, and the S channel is set to a preset value to obtain a first image in the HSV color space corresponding to the target trajectory data, as shown in FIG. 8. Figure 3a The speed information of the trajectories passing through each grid on the two-dimensional plane corresponding to the target trajectory data is filtered to retain only the speed information below a preset speed threshold (e.g., 20 km / h), and is mapped into the gray value channel of the gray space to obtain a second image in the gray space corresponding to the target trajectory data, as shown in FIG. 9. Figure 3b As shown in FIG. 9, since there are some low-speed situations of vehicles at the crossroads, it can be seen from FIG. 9 that there are some pixels with lower gray values. Figure 3b

[0061] It should be noted that before rendering the image according to the statistical results and the position information of each grid on the two-dimensional plane, in order to make the obtained feature image better represent the statistical results, the statistical results can be converted to the same scale first, so as to better compare and analyze. Specifically, the statistical results of each grid can be normalized (i.e., the number of trajectories passing through each grid, the direction information and the speed information are respectively scaled to the range of 0 to 1), and then the normalized statistical results are mapped to the interval of 0 to 255 to obtain the mapped statistical results, and the feature image is rendered according to the mapped statistical results. In this way, when the feature extraction network needs to extract the features of the feature image subsequently, the obtained feature representation is more accurate, and the stability and ability of the feature extraction network can be improved.

[0062] ​​It can be seen that the present application maps the number and direction information of the trajectories passed through each grid into a color space, and maps the speed information into another color space, to obtain a first image and a second image respectively, thereby mapping the expression of the target trajectory data in the geographical space into the color space, which is a more understandable expression mode for the model, so that more accurate lane lines can be extracted from the target trajectory data by using the model in the subsequent process.

[0063] As one of the implementable modes, a complete road can be taken as a whole to generate a feature image, and then generate lane line vector data of the road. For example, a first image corresponds to a road, that is, a first image is generated by using the trajectory data on the road. For another example, a first image and a second image correspond to a road.

[0064] However, the size and data amount of the feature image generated by this implementable mode are large, which is not conducive to subsequent generation of lane line vector data. Therefore, in order to reduce the processing granularity of the feature image and obtain a more accurate lane-level map, an implementable mode is provided, that is, the road corresponding to the trajectory data to be processed can be divided in advance. Specifically, the trajectory data to be processed is obtained, the road corresponding to the trajectory data to be processed is divided to obtain a plurality of processing units, the trajectory data corresponding to each processing unit is taken as target trajectory data respectively, a feature image of the trajectory data corresponding to each processing unit is generated, and then lane line vector data of each processing unit is generated.

[0065] The embodiment of the present application can obtain N feature images corresponding to N processing units according to the statistical results of the trajectory data corresponding to the N processing units on the road, wherein the N processing units are obtained by dividing the road according to a preset first size, and N is a positive integer. For example, the first size can be 120m*180m. The first size is selected as 120m*180m in the present application for the convenience of calculation, that is, one processing unit can cover the width of a general road (for example, a two-way highway including multiple lanes), and a series of processing units can cover the entire road. The target trajectory data in the present application corresponds to the trajectory data in one processing unit. Other first sizes can also be used, for example, two processing units arranged side by side can cover the width of the road.

[0066] As shown in FIG. 1, Figure 4a The purple line shows the road, the black line shows the trajectory on the road, and the blue rectangle shows the plurality of processing units. The plurality of processing units essentially divide the road, and the present application takes the trajectory data in each processing unit as target trajectory data. There can be partial overlap between adjacent processing units.

[0067] The application cuts the road according to the preset first size to obtain a feature image corresponding to each processing unit. The local feature image obtained in this way can clearly represent the track characteristics in each processing unit, thereby improving the prediction accuracy of the subsequent lane line vector data and further improving the accuracy of the lane-level map. Moreover, the multiple processing units can be processed in parallel, thereby improving the calculation efficiency.

[0068] Further, as a preferred embodiment, the trajectory data can be counted for each grid in the N processing units, wherein the grid is obtained by cutting the processing unit according to a preset second size, and the second size is smaller than the first size, for example, the second size can be 0.15m*0.15m. Figure 4b To intercept Figure 4a A grid diagram obtained by one of the processing units is shown in FIG. 2, wherein the yellow squares represent the grids in the processing unit. It should be noted that the number of grids in the figure is only illustrative. Figure 4b

[0069] Then, as one of the implementable ways, the first image corresponding to each of the N processing units can be respectively rendered according to the statistical results and position information of each grid in the N processing units, wherein the statistical results of the grid at least include the number and direction information of the track passing through the grid.

[0070] As another implementable way, the first image and the second image corresponding to each of the N processing units can be respectively rendered according to the statistical results and position information of each grid in the N processing units, wherein the statistical results of the grid include the number, direction information and speed information of the track passing through the grid.

[0071] The application further cuts each processing unit according to the second size to obtain the grids, and obtains a feature image corresponding to each processing unit according to the statistical results and position information of each grid in each processing unit. In this way, the application further refines the feature image of the trajectory data in the road, further enhances the local feature capturing capability, and can consider more local detailed information when predicting the lane line vector data, thereby laying a foundation for obtaining a lane-level map with high accuracy.

[0072] For the rendering process of the first image, the application maps each grid in each processing unit to each pixel region of the first image according to the position information of each grid in the processing unit, maps the number of tracks passing through each grid in the processing unit to a first channel of a multi-channel color space adopted by the first image, and maps the direction information of the tracks passing through each grid in the processing unit to a second channel of the multi-channel color space adopted by the first image, to obtain the first image corresponding to the processing unit in the multi-channel color space.​

[0073] For the rendering process of the second image, the present application maps each grid in the processing unit to each pixel region of the second image according to the position information of each grid in the processing unit, and maps the speed information of the trajectory passed by each grid in the processing unit to the single-channel color space adopted by the second image, to obtain the second image corresponding to the N processing units in the single-channel color space.

[0074] The above step 102, i.e., "processing the feature image to generate lane line vector data, the lane line vector data including position information, direction information and type information of each lane line", will be described in detail below in conjunction with an embodiment.

[0075] The present application can process the feature image to generate lane line vector data, wherein the lane line vector data includes position information, direction information and type information of a plurality of lane lines. The type information herein is used to represent the type of each lane line in the lane line vector data. The lane line is mainly used to distinguish the area of each lane and can be divided into a left boundary line (usually a yellow solid line in an actual road scene, used to separate opposite driving motor vehicle lanes), an outer boundary line (usually a white solid line in an actual road scene), a lane separation line (usually a white dashed line or a white solid line in an actual road scene, used to separate motor vehicle lanes driving in the same direction), and can further include a stop line and the like. Each lane line is composed of geometric elements such as points and lines. The lane line is essentially composed of continuous position points, and its direction can be represented by the order of the position points. For example, assuming that a lane line includes position points a, b, c, …, z, the coordinates of each position point can represent the position information of the lane line. The order of the position points on the lane line, such as the order from a to z or the order from z to a, represents the direction information of the lane line. In the embodiments of the present application, the vector data format can be used to represent the shape and position of the geometric elements.

[0076] After obtaining the feature image, the image processing method such as image segmentation, boundary recognition, etc. can be used to extract the lane center line, and then the lane line can be further extracted based on the lane center line. According to the relative position relationship between the lane line and the lane center line, the type of the lane boundary line can be identified. However, this method involves multiple steps and processing results, which leads to the accumulation of errors, and the accuracy of the obtained lane line vector data is not high. Therefore, the present application provides a more optimal method, which uses a model based on a Transformer network to realize the prediction from the feature image to the lane line vector data.

[0077] Specifically, if the feature image only includes the first image, the application can use a feature extraction network (i.e., the feature extraction model mentioned in the above process) to extract features from the first image to obtain a feature representation of the first image, which integrates the position information, density information, and direction information of the multiple trajectories included in the trajectory data. The feature extraction network can be a convolutional neural network (CNN), which is characterized by convolution operations and pooling operations for feature extraction and dimension reduction. It can also be a residual network (ResNet), which is characterized by residual learning, i.e., learning a residual function instead of directly learning an output, so that the residual network can directly pass the features of the previous layers to the subsequent layers, thereby avoiding the problems of gradient vanishing and gradient explosion. Further, the application uses a decoder based on Transformer (i.e., the generation model mentioned in the above process) to decode the feature representation of the first image to obtain the lane line vector data.

[0078] If the feature image includes the first image and the second image, as shown in Figure 5 Figure 5 The process of obtaining lane line vector data provided by the embodiments of the application is shown. The application can use a feature extraction network to extract features from the first image and the second image respectively to obtain a feature representation of the first image and a feature representation of the second image, and then fuse the feature representation of the first image and the feature representation of the second image to obtain a fused feature representation. At this time, the fused feature representation integrates the position information, density information, direction information, and speed information of the multiple trajectories included in the trajectory data. Then, a decoder based on Transformer is used to decode the fused feature representation to obtain the lane line vector data.

[0079] Wherein, when fusing the feature representation of the first image and the feature representation of the second image, a multi-head attention mechanism based on Transformer can be used. Through the multi-head attention mechanism, the association and weight between the feature representations of the two images can be learned, and effective fusion between them can be achieved.

[0080] ​When decoding with a Transformer-based decoder, the fused feature representation is processed using a Transformer layer, such as self-attention, and the latent vector output by the Transformer layer is mapped to the position and direction of the lane line. For example, the position and direction of the lane line are predicted using regression. Regarding lane type information, this can be understood as mapping the latent vector output by the Transformer layer to the lane type space, obtaining the confidence level for each lane type, and outputting the type with the highest confidence level as the lane type.

[0081] In addition, in addition to using the Transformer-based decoder to achieve the above-mentioned lane line type prediction, other methods can also be used, such as a classifier based on Softmax (an activation layer).

[0082] In execution Figure 1 Before the process shown, the end-to-end model can be pre-trained. For example, the first image and the second image of some roads with known lane line vector data can be obtained in advance using the method shown in step 101. The first image, the second image and the corresponding lane line vector data of the road are used as training samples, and training data is obtained by constructing multiple training samples, and the above-mentioned end-to-end model is trained using the training data. That is, the first image and the second image in the training sample are used as model inputs to obtain the lane line vector data predicted by the model. In each round of iteration, the difference between the predicted lane line vector data and the lane line vector data in the training sample is used to obtain the loss function, and the value of the loss function is used to update the model parameters using methods such as gradient descent until the training end condition is reached. The training end condition can be, for example, that the loss function is less than or equal to a preset loss function threshold, or that the number of iterations reaches a preset round number threshold.

[0083] This application utilizes an end-to-end model based on a Transformer network to predict lane vector data directly from feature images, avoiding the potential for error accumulation associated with multiple steps and processing results. Furthermore, the Transformer network's powerful ability to capture both global and local features improves lane vector prediction accuracy. Furthermore, this end-to-end model eliminates the need for complex post-processing, resulting in higher processing efficiency.

[0084] The above step 103, i.e., "obtaining a lane-level map based on the lane line vector data," is described in detail below with reference to an embodiment.

[0085] The application can obtain a lane-level map based on lane line vector data. If the lane line vector data is in a two-dimensional plane, the obtained lane-level map can be a two-dimensional map. The application can further combine height values to obtain a three-dimensional lane-level map.

[0086] For example, if the feature image in the embodiment of the application is generated in a bird's eye view in step 101, and the lane-level vector data in the bird's eye view is generated in step 102, the height information of each lane line can be determined in step 102 by using the height information of the neighboring trajectory point set of each lane line in the trajectory data, wherein the neighboring trajectory point set of each lane line includes trajectory points within a preset distance range from the lane line. Then, three-dimensional lane line vector data can be obtained according to the height information of each lane line and the lane line vector data in the bird's eye view. Further, the lane-level map in the three-dimensional space can be obtained in step 103 according to the three-dimensional lane line vector data.

[0087] For determining the height information of each lane line, the application provides two methods.

[0088] One method is to use the Ransac algorithm to fit the height value of each lane line.

[0089] Specifically, for each lane line, the neighboring trajectory point set is found in the trajectory data, and a mathematical model is assumed to represent the height of the lane line in the three-dimensional space. Then, a group of trajectory points is randomly selected from the neighboring trajectory point set as inliers (i.e., the group of inliers is assumed to conform to the mathematical model), the mathematical model is fitted using the group of inliers to obtain the parameters of the mathematical model, and then the fitting model is determined. Then, the distances of all trajectory points in the neighboring trajectory point set to the fitting model are calculated, and according to a preset threshold, it is determined which trajectory points are inliers (conform to the fitting model) and which are outliers (do not conform to the fitting model).

[0090] The above process is repeated multiple times, and after each iteration, the fitting model is evaluated according to the number of inliers corresponding to the fitting model, and the fitting model is selected as the final model according to the evaluation result. Finally, for each lane line, the corresponding height value is calculated according to the final model.

[0091] Another method is to map each lane line included in the lane line vector data to a grid in a two-dimensional plane, and determine the height information of each lane line according to the height values of the trajectory data corresponding to each grid.

[0092] Specifically, each trajectory included in the trajectory data is projected into a grid of a two-dimensional plane (for example, the size of the grid can be 2m*2m), and the height value of the trajectory data contained in each grid is determined by the trajectory data. Assuming that there are 10 trajectory points in a grid, the height value corresponding to the grid is determined according to the average value of the altitude information of the 10 trajectory points. Then, each lane line included in the lane line vector data is also mapped into the grid of the two-dimensional plane, and the height value of the lane line is determined according to the height value corresponding to the mapped grid. That is, the trajectory points mapped into the same grid as the lane line are taken as the neighbor trajectory point set of the lane line, and the height average value of the neighbor trajectory point set is taken as the height value of the lane line.

[0093] It can be seen that the present application first determines the lane line vector data under the bird's eye view, and then determines the height information of each lane line included in the lane line vector data under the bird's eye view according to the trajectory data, and further obtains the three-dimensional lane line vector data. Such a way can obtain three-dimensional lane line vector data in three-dimensional space, and further obtain a three-dimensional lane-level map, and since the height of the lane line is determined by the actual height of the trajectory data, the road situation in the actual scene is more similar, which can bring better navigation experience to the user.

[0094] In step 101, if the road corresponding to the trajectory data to be processed is divided into N processing units, and N is equal to 1, the lane-level map can be directly processed according to the lane line vector data corresponding to the processing unit.

[0095] If N is an integer greater than 1, and there is no overlapping area between adjacent processing units, the lane line vector data corresponding to the N processing units can be directly spliced to obtain global lane line vector data, and then the lane-level map is processed according to the global lane line vector data.

[0096] As a preferred embodiment, if N is an integer greater than 1, and there is an overlapping area between adjacent processing units, the lane line vector data corresponding to the N processing units can be fused to obtain the lane-level map of the road.

[0097] For the fusion of lane line vector data between different processing units, the present application provides a method. When the road is divided according to a first size, a certain overlapping area is left between adjacent processing units, which is to facilitate the fusion of lane line vector data between adjacent processing units to obtain a more complete and more accurate lane-level map. Specifically, for adjacent processing units with overlapping areas, Hungarian matching is used to match the lane line vector data of the two processing units, and the matched lane lines can be fused, or a more complex weighted fusion strategy can be used.

[0098] The lane line vector data corresponding to the N processing units is fused, alignment between adjacent processing units is realized, more complete and accurate lane line vector data is obtained, and a basis is provided for construction and updating of a lane-level map.

[0099] Figure 2a a schematic diagram of a first image corresponding to trajectory data on a road, Figure 2b a schematic diagram of a second image corresponding to trajectory data on the same road, Figure 2b In the method, only speed information lower than a preset speed threshold is retained based on the speed threshold. Figure 2a and Figure 2b The corresponding lane line vector data is as shown in Figure 6a For example, Figure 3a a schematic diagram of a first image corresponding to trajectory data on a road in a crossroad region, Figure 3b a schematic diagram of a second image corresponding to trajectory data on the road in the region, Figure 3a and Figure 3b The corresponding lane line vector data is as shown in Figure 6b In Figure 6a and Figure 6b Different colors are used to distinguish different types of lane lines, wherein an outer boundary line can be displayed in red, a left boundary line can be displayed in blue, a stop line can be displayed in green, and a guide lane line can be displayed in black. In fact, the lane line vector data is a set including position information, direction information, and type information of each lane line.

[0100] As can be seen from the above embodiments, the target trajectory data is composed of trajectory points, and each trajectory point includes latitude, longitude, and altitude information of the trajectory point. The target trajectory data is usually in a global coordinate system, for example, the global coordinate system is a coordinate system with the center of the earth as the origin, the geographic north pole as the positive direction of the Z axis, the geographic equatorial plane as the XOY plane, and the X axis pointing to the geographic prime meridian. However, the expression of the global coordinate system is not convenient for data processing when generating feature images of each processing unit. Therefore, the target trajectory data in the N processing units can be converted into N local coordinate systems corresponding to the N processing units. For each processing unit, the local coordinate system corresponding to the processing unit takes the geometric center of the processing unit as the origin. Then, the feature image and the lane line vector diagram are generated based on the expression of the target trajectory data in the local coordinate system.

[0101] In the conversion of the target trajectory data in the N processing units to the local coordinate system corresponding to the N processing units respectively, a SD (Standard Definition Map) map can be used as a reference. For example, the longitude and latitude coordinates of each trajectory point and the longitude and latitude coordinates of the geometric center of each processing unit are determined by using the SD map, the geometric center of each processing unit is taken as the origin of the local coordinate system respectively, and then the target trajectory data in the N processing units are converted to the local coordinate system corresponding to the N processing units respectively.

[0102] Further, after the lane line vector data corresponding to the N processing units is generated, for each processing unit, the lane line vector data corresponding to the processing unit is converted from the local coordinate system corresponding to the processing unit to the global coordinate system, and based on the lane line vector data corresponding to the N processing units in the global coordinate system, a lane-level vector map is obtained.

[0103] In order to more clearly illustrate the method provided in the present application, a specific embodiment is used for explanation as a preferred embodiment, as shown in Figure 7 Figure 7 is a flowchart of obtaining a lane-level map in the embodiment of the present application. As can be seen from Figure 7 , the present application first needs to collect trajectory data on the road, and pre-process the trajectory data, which includes mapping the trajectory data to a two-dimensional plane in the bird's eye view. Then the road is divided into N processing units, each processing unit including grids. For each processing unit, according to the statistical results of the grids in the processing unit, the first image and the second image corresponding to the processing unit are respectively rendered. The end-to-end model is used to generate lane-level vector data corresponding to each processing unit, which is lane line vector data in the two-dimensional plane in the bird's eye view. Then, according to the collected original trajectory data, the height information of each lane line in the lane line vector data corresponding to the N processing units is determined, and then three-dimensional lane line vector data corresponding to the N processing units is obtained. The three-dimensional lane line vector data corresponding to the N processing units is fused, and according to the fused global lane line vector data, a lane-level map is obtained.

[0104] The above method provided in the embodiment of the present application can be applied to various application scenarios, including but not limited to: used for navigation of urban roads, in urban roads, the road conditions are usually complex, such as temporary widening or narrowing of lanes, continuous intersections, bus lanes and other special situations, the lane-level map obtained based on the method of the present application can more accurately display the actual road conditions, helping users to make lane-changing decisions in advance; used for automatic driving assistance, in some advanced automatic driving systems, the lane-level map is combined with vehicle sensor data to realize autonomous navigation and obstacle avoidance functions of the vehicle. ​

[0105] The above describes particular embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims can be performed in a different order than those in the embodiments and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or necessary.

[0106] According to an embodiment of another aspect, a map processing apparatus is provided. Figure 8 A schematic block diagram of the map processing apparatus according to an embodiment is shown. As shown, the apparatus 800 comprises an image generating unit 801, a vector generating unit 802 and a map processing unit 803, and can further comprise a road segmentation unit 804. The main functions of each component unit are as follows: Figure 8

[0107] The image generating unit 801 is configured to generate a feature image based on target trajectory data, the target trajectory data comprising a plurality of trajectories, the feature image characterizing at least position information, density information and direction information of the plurality of trajectories.

[0108] The vector generating unit 802 is configured to process the feature image to obtain lane line vector data, the lane line vector data comprising position information, direction information and type information of each lane line.

[0109] The map processing unit 803 is configured to obtain a lane level map based on the lane line vector data.

[0110] As one of the implementable manners, the feature image comprises a first image and a second image; wherein the first image is used to characterize position information, density information and direction information of the plurality of trajectories; and the second image is used to characterize speed information of the plurality of trajectories.

[0111] As one of the implementable manners, when generating the feature image based on the target trajectory data, the image generating unit 801 can be configured to: map the target trajectory data to a two-dimensional plane under a bird's eye perspective; perform trajectory data statistics on each grid on the two-dimensional plane based on the mapped trajectory data; and render the feature image based on the statistics result and position information of each grid, the statistics result of each grid at least comprising the number of trajectories passing through the grid and direction information.

[0112] ​As one of the implementable manners, the feature image includes the first image, and when rendering the feature image based on the statistical result and the position information of each grid, the image generation unit 801 can be configured to: map each grid to each pixel region of the first image according to the position information of each grid, map the number of trajectories passed by each grid to a first channel of a multi-channel color space adopted by the first image, and map the direction information of the trajectories passed by each grid to a second channel of the multi-channel color space, to obtain the first image.

[0113] As one of the implementable manners, the multi-channel color space is HSV.

[0114] When obtaining the first image, the image generation unit 801 can also be configured to: obtain the first image in HSV, and convert the first image in HSV into the first image in RGB.

[0115] As one of the implementable manners, the feature image further includes the second image, and when rendering the feature image based on the statistical result and the position information of each grid, the image generation unit 801 can also be configured to: map each grid to each pixel region of the second image according to the position information of each grid, and map the speed information of the trajectories passed by each grid to a single-channel color space adopted by the second image, to obtain the second image.

[0116] As one of the implementable manners, when processing the feature image to obtain the lane line vector data, the vector generation unit 802 can be configured to: perform feature extraction on the first image and the second image respectively by using a feature extraction network to obtain a feature representation of the first image and a feature representation of the second image; fuse the feature representation of the first image and the feature representation of the second image to obtain a fused feature representation; and decode the fused feature representation by using a decoder of a Transformer network to obtain the lane line vector data.

[0117] As one of the implementable manners, when obtaining the lane-level map based on the lane line vector data, the map processing unit 803 can be configured to: determine the height information of each lane line by using the height information of the near-neighbor trajectory point set of each lane line in the target trajectory data, the near-neighbor trajectory point set of the lane line including trajectory points within a preset distance range from the position of the lane line; obtain three-dimensional lane line vector data according to the height information and the lane line vector data under the bird's-eye view; and obtain a three-dimensional lane-level map based on the three-dimensional lane line vector data.

[0118] Further, before generating the feature image based on the target trajectory data, the road segmentation unit 804 can be configured to: obtain the trajectory data to be processed; segment the road corresponding to the trajectory data to be processed into a plurality of processing units according to a first size; and take the trajectory data corresponding to each processing unit as the target trajectory data respectively.

[0119] As one of the implementable manners, the map processing unit 803 can be configured to: fuse the lane line vector data corresponding to adjacent processing units to obtain lane line vector data of the road; and obtain the lane-level map based on the lane line vector data of the road when obtaining the lane-level map based on the lane line vector data.

[0120] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. Especially, the device embodiments are described more simply because they are basically similar to the method embodiments. The relevant parts can be referred to the part of the description of the method embodiments. The device embodiments described above are only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0121] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0122] In addition, the present application also provides a computer program product, comprising a computer program which, when executed by a processor, implements the steps of the method of any one of the preceding method embodiments.

[0123] In addition, the present application also provides a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method of any one of the preceding method embodiments.

[0124] and an electronic device comprising:

[0125] one or more processors; and

[0126] a memory associated with the one or more processors, the memory for storing program instructions that, when read and executed by the one or more processors, perform the steps of the method of any of the preceding method embodiments.

[0127] wherein, Figure 9 An exemplary architecture of the electronic device is shown, which can specifically include a processor 910, a video display adapter 911, a disk drive 912, an input / output interface 913, a network interface 914, and a memory 920. The processor 910, the video display adapter 911, the disk drive 912, the input / output interface 913, the network interface 914, and the memory 920 can be communicatively connected through a communication bus 930.

[0128] The processor 910 can be implemented in a general-purpose CPU, a microprocessor, an application specific integrated circuit (ASIC), or one or more integrated circuits, etc., for executing related programs to implement the technical solutions provided in the present application.

[0129] The memory 920 can be implemented in a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 920 can store an operating system 921 for controlling the operation of the electronic device 900, a basic input / output system (BIOS) 922 for controlling the low-level operation of the electronic device 900. In addition, a web browser 923, a data storage management system 924, and a map processing apparatus 800, etc. can also be stored. The map processing apparatus 800 can be an application program for implementing the above steps in the embodiments of the present application. In general, when the technical solutions provided in the present application are implemented by software or firmware, the related program codes are stored in the memory 920 and executed by the processor 910.

[0130] The input / output interface 913 is used to connect input / output modules to realize information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input devices can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices can include a display, a speaker, a vibrator, an indicator light, etc.

[0131] The network interface 914 is configured to connect a communication module (not shown in the figure) to realize the communication interaction between the device and other devices. The communication module can realize communication through wired mode (such as USB, network cable, etc.), or can realize communication through wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0132] The bus 930 includes a path for transmitting information between various components (such as the processor 910, the video display adapter 911, the disk drive 912, the input / output interface 913, the network interface 914, and the memory 920) of the device.

[0133] It should be noted that although the above device only shows the processor 910, the video display adapter 911, the disk drive 912, the input / output interface 913, the network interface 914, the memory 920, the bus 930, etc., in the specific implementation process, the device can also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device can also only contain the components necessary to implement the scheme of the present application, and does not have to contain all the components shown in the figure.

[0134] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and the necessary general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer program product, which can be stored in a storage medium such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions for making a computer device (which can be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0135] The above describes the technical solutions provided by the present application in detail, and the principles and implementation manners of the present application are described by applying specific examples; the above description of the embodiments is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A map processing method, characterized in that: The method comprises: generating a feature image based on target trajectory data, wherein the target trajectory data includes a plurality of trajectories, position information of the plurality of trajectories is mapped to pixel areas of the feature image, and density information and direction information of the plurality of trajectories are mapped to different channels of a multi-channel color space used by the feature image; Using a Transformer network-based model to predict lane line vector data based on the feature image, the lane line vector data includes position information, direction information, and type information of each lane line; A lane-level map is obtained based on the lane line vector data.

2. The method according to claim 1, characterized in that The characteristic image includes a first image and a second image; The first image is used to represent the position information, density information, and direction information of the multiple tracks; and the second image is used to represent the speed information of the multiple tracks.

3. The method according to claim 1, characterized in that The generating of the feature image based on the target trajectory data includes: Mapping the target trajectory data to a two-dimensional plane from a bird's-eye view; Based on the mapped trajectory data, statistics are performed on each grid on the two-dimensional plane; The characteristic image is rendered based on the statistical results and position information of each grid, where the statistical results of the grid at least include the number and direction information of the tracks through which the grid passes.

4. The method according to claim 3, characterized in that The characteristic image includes a first image, and the rendering to obtain the characteristic image based on the statistical results and position information of each grid includes: According to the position information of each grid, each grid is mapped to each pixel area of ​​the first image, the number of tracks passed by each grid is mapped to the first channel of the multi-channel color space used by the first image, and the direction information of the tracks passed by each grid is mapped to the second channel of the multi-channel color space to obtain the first image.

5. The method according to claim 4, characterized in that The multi-channel color space is HSV; The obtaining of the first image includes: obtaining the first image using the HSV, and converting the first image using the HSV into a first image using RGB.

6. The method according to claim 4, characterized in that The characteristic image further includes a second image, and the rendering of the characteristic image based on the statistical results and position information of each grid further includes: The grids are mapped onto pixel regions of the second image according to their position information, and the speed information of the trajectories through which the grids pass is mapped onto a single-channel color space used by the second image to obtain the second image.

7. The method according to claim 2, characterized in that The method of using a Transformer network-based model to predict lane line vector data based on the feature image includes: Using a feature extraction network to perform feature extraction on the first image and the second image respectively, to obtain a feature representation of the first image and a feature representation of the second image; fusing the feature representation of the first image and the feature representation of the second image to obtain a fused feature representation; The fused feature representation is decoded using a decoder based on a Transformer network to obtain the lane line vector data.

8. The method according to claim 3, characterized in that The obtaining of a lane-level map based on the lane line vector data further includes: Determining the height information of each lane line using the height information of a neighboring trajectory point set of each lane line in the target trajectory data, wherein the neighboring trajectory point set of the lane line includes trajectory points within a preset distance range from the position of the lane line; Obtaining three-dimensional lane line vector data based on the height information and the lane line vector data from the bird's-eye view; Based on the three-dimensional lane line vector data, a three-dimensional lane-level map is obtained.

9. The method according to any one of claims 1 to 8, characterized in that Before generating the feature image based on the target trajectory data, the method includes: Get the trajectory data to be processed; Dividing the road corresponding to the trajectory data to be processed into a plurality of processing units according to a first size; The trajectory data corresponding to each of the processing units is used as the target trajectory data.

10. The method according to claim 9, characterized in that The obtaining of a lane-level map based on the lane line vector data includes: Fusing lane line vector data corresponding to adjacent processing units to obtain lane line vector data of the road; A lane-level map is obtained based on the lane line vector data of the road.

11. A map processing device, characterized in that: The device comprises: an image generation unit configured to generate a feature image based on target trajectory data, wherein the target trajectory data includes a plurality of trajectories, position information of the plurality of trajectories is mapped to pixel areas of the feature image, and density information and direction information of the plurality of trajectories is mapped to different channels of a multi-channel color space used by the feature image; A vector generation unit is configured to use a Transformer network-based model to predict lane line vector data based on the feature image, wherein the lane line vector data includes position information, direction information, and type information of each lane line; The map processing unit is configured to obtain a lane-level map based on the lane line vector data.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images