CN114037966A - High-precision map feature extraction method, device, medium and electronic equipment - Google Patents

Abstract

The application discloses a high-precision map feature extraction method, device, medium and electronic equipment, and relates to the field of artificial intelligence, in particular to the field of computer vision, and specifically relates to the fields of automatic driving, high-precision maps and intelligent transportation. The specific implementation scheme is as follows: acquiring at least two frames of road images acquired in different lanes in the same road area; respectively extracting the original geometric features of the pavement markers in the at least two frames of road images; and carrying out fusion processing on at least two original geometric characteristics to obtain the target geometric characteristics of the pavement marker. By executing the technical scheme provided by the application, the geometric features of the pavement markers are extracted, and more accurate geometric features can be extracted.

Description

High-precision map feature extraction method, device, medium and electronic device

technical field

This application relates to the field of artificial intelligence, in particular to the field of computer vision, and in particular to the fields of autonomous driving, high-precision maps, and intelligent transportation.

Background technique

High-precision maps, also known as high-precision maps, are used by autonomous vehicles. High-precision maps, with accurate vehicle location information and rich road element data information, can help cars predict complex road information, such as slope, curvature, heading, etc., to better avoid potential risks. A large number of pavement signs exist on urban roads. In the process of vehicle automatic driving, high-precision map production, and intelligent traffic management, pavement signs are often extracted based on road images collected by cameras. However, affected by the acquisition perspective, the pavement markings in the road images often have a certain degree of lack and distortion, which seriously affects the accuracy of pavement marking feature extraction.

SUMMARY OF THE INVENTION

The present application discloses a high-precision map feature extraction method, device, medium and electronic device for extracting geometric features of road markings, so as to extract more accurate geometric features.

According to an aspect of the present application, a method for extracting features from a high-precision map is provided, the method comprising:

Obtain at least two frames of road images collected in different lanes for the same road area;

respectively extracting the original geometric features of the road surface markings in the at least two frames of road images;

Perform fusion processing on at least two of the original geometric features to obtain the target geometric features of the road surface marking.

According to another aspect of the present application, an electronic device is provided, the electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute any one of the embodiments of the present application High-precision map feature extraction method.

According to an aspect of the present application, a non-transitory computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to cause the computer to execute the high-precision map feature according to any one of the embodiments of the present application. Extraction Method.

By implementing the technical solution provided in the present application, the geometric features of the pavement markings are extracted, and more accurate geometric features can be extracted.

It should be understood that the content described in this section is not intended to identify key or critical features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the present application will become readily understood from the following description.

Description of drawings

The accompanying drawings are used for better understanding of the present solution, and do not constitute a limitation to the present application. in:

1 is a schematic diagram of a method for extracting features from a high-precision map according to an embodiment of the present application;

2 is a schematic diagram of yet another method for extracting features from a high-precision map according to an embodiment of the present application;

3A is a schematic diagram of yet another method for extracting features from a high-precision map according to an embodiment of the present application;

3B is a schematic flowchart of another method for extracting features of a high-precision map provided by an embodiment of the present application;

4 is a schematic diagram of yet another method for extracting features from a high-precision map according to an embodiment of the present application;

5 is a schematic diagram of yet another method for extracting features from a high-precision map according to an embodiment of the present application;

6 is a schematic diagram of an apparatus for extracting features of a high-precision map according to an embodiment of the present application;

FIG. 7 is a block diagram of an electronic device used to implement a method for extracting features of a high-precision map according to an embodiment of the present application.

Detailed ways

Exemplary embodiments of the present application are described below with reference to the accompanying drawings, which include various details of the embodiments of the present application to facilitate understanding, and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

FIG. 1 is a schematic diagram of a method for extracting features of a high-precision map according to an embodiment of the present application. This embodiment is applicable to the case of extracting the geometric features of road surface markings. The high-precision map feature extraction method disclosed in this embodiment may be executed by a high-precision map feature extraction apparatus, which may be implemented in software and/or hardware, and configured in an electronic device with computing and storage functions. Referring to FIG. 1 , a method for extracting features from a high-precision map provided by this embodiment includes:

S110: Acquire at least two frames of road images collected in different lanes for the same road area.

The road area refers to an area on the road where ground signs are drawn. Exemplarily, the road area may be a road intersection on which steering guidance, stop lines or pedestrian crossings are drawn.

Among them, the road image can be collected by a professional image acquisition vehicle, or a common vehicle can be collected by a driving recorder or a handheld device with a camera function, such as a mobile phone and a tablet computer. Here, the road image collection method and collection equipment are not limited. It is determined according to the actual situation. The road image may be a front view of the road area.

Optionally, at least two frames of road images may be collected in different lanes by a collection device, and for each lane, at least one frame of road images is collected. For example, it is possible to control the vehicles installed with the acquisition equipment to drive in each lane respectively, and control the acquisition equipment to image the road images according to a preset frequency (such as 10 frames of images per second) during the process of the vehicle driving in each lane. Acquisition, that is, to acquire multiple frames of images for each lane.

The image collection positions corresponding to different road images in this embodiment are different. It can be known that, since the collection position will affect the camera angle of view, if the same road area is photographed at different collection positions, the road area under different camera angles of view can be obtained, and the road area can be collected from multiple angles.

S120. Extract the original geometric features of the road surface markers in the at least two frames of road images respectively.

Among them, the road marking refers to the graphics drawn by the relevant departments on the road to guide pedestrians or vehicles. Exemplarily, the pavement markings may be diamond-shaped markings, lane markings, pedestrian crossings or stop lines, and the like. The original geometric features of pavement markings are the information used to describe the polygons that constitute the pavement markings. Exemplarily, when the road surface is marked as a crosswalk, the original geometric features of the crosswalk may be feature data such as corner points or edges of polygons that constitute the crosswalk.

In an optional embodiment, the road markings are road markings across lanes. Wherein, the cross-lane road marking refers to a road marking that simultaneously crosses at least one lane. Exemplarily, the road surface markings across the lanes are pedestrian crossings or stop lines or the like. Due to the camera angle of view, when the camera is used to capture images of road markings across lanes, the road markings are incomplete, and the areas where the road markings across lanes are far from the camera are prone to geometric deformation. The geometric features of pavement markings of lanes provide an accurate and efficient method.

Respectively extract the original geometric features of the pavement signs in the road images collected under each lane. Specifically, one possible implementation may be to perform semantic segmentation on each road image separately, and first use the semantic segmentation network to perform full element segmentation on the road image. , and segment the signs, vegetation, roads, and pedestrians and vehicles in the road image. Then, the ground elements are extracted, and further semantic segmentation is performed for the ground elements, and elements such as lane lines, ground arrows, pedestrian crossings, stop lines, and ground speed limits are segmented. Next, the cross-lane pavement markings are selected from the segmented ground elements, and the cross-lane pavement markings are processed by the connected domain analysis method, clustering algorithm and outer contour analysis method to determine the original geometric characteristics of the pavement markings. Optionally, in this embodiment, at least one frame of road image collected for each lane may be used to determine the original geometric features of a group of road signs; it may also be for each frame of road image, the original geometry of a group of road signs may be determined. Features, this book is limited.

Preferably, since the road signs are located on the ground, in this embodiment, when extracting the original geometric features of the road signs from the road images, the collected at least two frames of road images may be converted into a top view, and then the top view is extracted. The original geometric features of pavement markings to improve the accuracy of feature extraction.

Another possible implementation may be to input at least two frames of road images into a pre-trained feature recognition model, and extract the original geometric features of road signs in each frame of road images through the feature recognition model.

S130. Perform fusion processing on at least two of the original geometric features to obtain a target geometric feature of the road surface marking.

Among them, the original geometric features refer to the unprocessed geometric features extracted from a single road image. Due to the problem of camera angle of view, when the camera is used to capture images of road markings across lanes, the areas where the road markings across lanes are far from the camera are prone to geometric deformation. Therefore, the original geometric features may not truly reflect the geometric properties of pavement markings. The target geometric feature is a feature fusion result obtained by performing fusion processing on at least two original geometric features. Since the original geometric features are extracted from a single road image, the extracted geometric features are incomplete. The target geometric feature fuses the original geometric features captured by at least two different lanes. Compared with a single original geometric feature, the target geometric feature is more accurate and complete.

Perform fusion processing on at least two original geometric features to obtain the target geometric feature of the road surface marking. Specifically, at least two original geometric features may be superimposed first, and then combined with the standard geometric features of road markings, the superimposed geometric features may be corrected to obtain the target geometric features; or at least two original geometric features may be combined Input into the pre-built feature fusion model, the original geometric features are fused by the feature fusion model, and the target geometric features are output; it is also possible to perform feature fusion based on the regional intersection relationship between the original geometric features to obtain the target geometric features.

The technical solution of the embodiment of the present application is to obtain at least two frames of road images collected from the same road area in different lanes; to extract the original geometric features of the road surface signs in the road images collected under each lane; Fusion processing is performed to obtain the target geometric features of the pavement markings. The present application fuses at least two original geometric features, effectively integrates geometric features obtained from different camera perspectives, and obtains more accurate target geometric features including more road marking geometric feature information, and implements the technical solution provided by the present application. It can effectively solve the problem of inaccurate geometric features of pavement markings extracted from road images due to incomplete and distorted geometric features of road markings in road images.

FIG. 2 is a schematic diagram of another method for extracting features of a high-precision map according to an embodiment of the present application; the present embodiment is an optional solution proposed on the basis of the above-mentioned embodiment. Specifically, it is a refinement of the operation "fusing at least two of the original geometric features to obtain the target geometric feature of the road surface marking".

Referring to FIG. 2 , the method for extracting features of a high-precision map provided by this embodiment includes:

S210: Acquire at least two frames of road images collected in different lanes for the same road area.

S220. Extract the original geometric features of the road surface markers in the at least two frames of road images respectively.

Specifically, at least a set of original geometric features of road surface markers are extracted from each frame of road image.

S230. Determine an intersection area and a non-intersection area of at least two of the original geometric features.

The intersection area of the original geometric features refers to the image area where the original geometric features describing the same part of the pavement mark in different road images are located. In contrast, the image area where other original geometric features in the road image are located is the non-intersection area of the original geometric features.

Exemplarily, when the road surface is marked as a pedestrian crossing, the extracted original geometric features are polygonal outlines that constitute the pedestrian crossing. Since a pedestrian crossing is a road marking that crosses lanes, some pedestrian crossings may be missing in the road image due to the problem of the collection location, and there may be differences in the missing part of the pedestrian crossing in different road images. The overlapping part of the crosswalk in the road image, that is, the part of the crosswalk included in each road image, is the intersection area of the original geometric features.

S240. Extract at least two candidate local features of the original geometric features in the intersection region, and determine a first local feature of the intersection region according to the confidence of the candidate local features in the intersection region.

Among them, candidate local features refer to some original geometric features located in the intersection area. Confidence refers to the accuracy of the original geometric features located in the intersection area. Generally speaking, the area with small deformation of pavement markings in the road image will have higher confidence than other areas. The confidence level can be obtained through network model analysis, can be determined by comparing and analyzing candidate local features based on standard geometric features, and can also be determined based on the distance between the road marking and the collection location. The method for determining the confidence is not limited here, and is specifically determined according to the actual situation.

The first local feature refers to a candidate local feature used to generate the target geometric feature.

The candidate local features refer to the original geometric features located in the intersection area in different road images. Due to the difference in the deformation degree of the candidate local features, the corresponding confidence levels of the candidate local features are different.

The candidate local features of at least two original geometric features in the intersection area are extracted, and the first local feature of the intersection area is determined according to the confidence of the candidate local features in the intersection area. Specifically, candidate local features of at least two original geometric features in the intersection area are extracted, confidence levels corresponding to each candidate local feature are calculated, and a candidate local feature corresponding to the maximum confidence level is selected as the first local feature. In this way, under the circumstance that the geometric features of the pavement signs are redundant, the candidate local features can be screened according to the confidence, and the relatively accurate geometric features can be selected, thus ensuring the accuracy of the geometric features.

In an optional embodiment, the confidence level of the candidate local feature in the intersection region is determined according to the distance value between the position of the intersection region and the collection position of the candidate local feature.

Wherein, the collection position of the candidate local feature may be the collection position of the road image. It can be known that in the process of using the camera to capture images, due to the problem of the camera's angle of view, the deformation of the area farther from the shooting position is more serious, especially for the ground signs that cross lanes.

A road image is an image collected from the same road area at different collection locations. The intersection area is in different road images, and there are also differences in the collection locations of the candidate local features. According to the distance value between the two, it is determined that the candidate local features are in the intersection area. confidence. Specifically, in the world coordinate system, the Euclidean distance between the center position of the intersection area and the collection position of the road image can be calculated to determine the confidence level of the candidate local feature in the sub-area. Generally speaking, the distance value is negatively correlated with the confidence, that is, the larger the distance value, the smaller the confidence.

In the present application, the confidence level of the candidate local features in the intersection region is determined according to the distance value between the position of the intersection region and the collection position of the candidate local features, thereby improving the extraction accuracy of the geometric features corresponding to the pavement markings.

S250. Extract second local features of at least two of the original geometric features in the non-intersection region.

The original geometric features located in the non-intersection area describe the geometric characteristics of different parts of the pavement marking respectively. In order to ensure the integrity of the pavement marking information, the original geometric features located in the non-intersection area are respectively extracted as the second local feature. The second local feature refers to the original geometric feature located in the non-intersection area, and the second local feature is used to generate the target geometric feature together with the first local feature.

S260. Determine the target geometric feature of the road marking according to the first local feature and the second local feature.

The first local feature is generated in the intersection region of the original geometric features, and the second local feature is generated in the non-intersection region of the original geometric features. According to the first local feature and the second local feature, the target geometric feature of the road surface marking is determined, and specifically, the target geometric feature is obtained by splicing the first local feature and the second local feature.

The technical solution of the embodiment of the present application distinguishes the intersection area and non-intersection area of the original geometric features, where there is a redundant intersection area in the original geometric features, screen candidate local features according to the confidence, and select the relatively accurate first Local features ensure the accuracy of target geometric features. The present application extracts the second local features of at least two original geometric features in the non-intersection area, and determines the target geometric feature of the pavement marking according to the first local feature and the second local feature, which ensures the integrity of the pavement marking information.

FIG. 3A is a schematic diagram of another method for extracting features of a high-precision map according to an embodiment of the present application; this embodiment is an optional solution proposed on the basis of the above-mentioned embodiment. Specifically, it is a refinement of the operation "determine the first local feature of the intersection region according to the confidence of the candidate local features in the intersection region".

Referring to FIG. 3A , the method for extracting features of a high-precision map provided by this embodiment includes:

S310: Acquire at least two frames of road images collected in different lanes for the same road area.

S320. Extract the original geometric features of the road surface markers in the at least two frames of road images respectively.

S330. Determine an intersection area and a non-intersection area of at least two of the original geometric features.

S340: Extract at least two candidate local features of the original geometric features in the intersection area, and divide the intersection area into at least two sub-areas.

The intersection area is divided into at least two sub-areas. Specifically, the vehicle driving direction is taken as the reference direction, and the intersection area corresponding to the road surface markings perpendicular to the reference direction is divided.

Optionally, the intersection area is divided into equal parts by a straight line parallel to the reference direction to obtain at least two sub-areas.

S350. Determine the first local feature of the intersection region according to the confidence level of the candidate local feature in each subregion.

Different sub-regions correspond to different confidence levels, and the confidence levels corresponding to each sub-region are calculated separately to determine the confidence levels of candidate local features located in the sub-regions. According to the confidence of the candidate local features, the candidate local features located in the intersection area are screened, and the first local feature is screened out from the intersection area.

Specifically, the confidence levels of the candidate local features in each sub-region are sorted in a certain order, and the candidate local feature corresponding to the maximum confidence level is selected as the first local feature.

In an optional embodiment, the confidence level of the candidate local feature in the subregion is determined according to the distance value between the subregion position and the collection position of the candidate local feature. Wherein, the collection position of the candidate local feature may be the collection position of the road image. The distance value between the position of the sub-region and the collection position of the candidate local feature. Specifically, in the world coordinate system, the Euclidean distance between the center position of the sub-region and the collection position of the road image can be calculated to determine that the candidate local feature is in the sub-region. confidence. Generally speaking, the distance value is negatively correlated with the confidence, that is, the larger the distance value, the smaller the confidence. In the present application, the confidence level of the candidate local feature in the sub-region is determined according to the distance value between the sub-region position and the collection position of the candidate local feature, thereby improving the extraction accuracy of the geometric feature corresponding to the road surface marking.

S360. Extract second local features of at least two of the original geometric features in the non-intersection region.

S370. Determine the target geometric feature of the road surface marking according to the first local feature and the second local feature.

The technical solutions of the embodiments of the present application divide the intersection area into at least two sub-areas, and determine the first local feature of the intersection area according to the confidence of the candidate local features in each sub-area. The fine-grained calculation of the corresponding confidence of the candidate local features is realized, the candidate local features are screened according to the confidence, and the accuracy of the first local feature is further improved, thereby ensuring the extraction accuracy of the geometric features corresponding to the pavement markings.

For ease of understanding, FIG. 3B is a schematic flowchart of another method for extracting features of a high-precision map provided by an embodiment of the present application. As shown in FIG. 3B , first, extract the original geometric feature A of the pavement mark in the road image (such as a top view of the road image) a, and the original geometric feature B of the pavement mark in the road image (such as the top view of the road image) b. Then, the intersection area C and the non-intersection area D ₁ and D ₂ of the original geometric feature A and the original geometric feature B are determined. Next, the intersection area C is divided into at least two sub-areas, C _i is a sub-area in the intersection area C, A _i and B _i are the candidate parts of the original geometric feature A and the original geometric feature B in the sub-region C _i , respectively feature. According to the confidence of the candidate local features A _i and B _i in the sub-region C _i , B _i is determined as the first local feature of the intersection region C. Extract the second local feature located in the non-intersection area D ₁ in the original geometric feature A and extract the second local feature located in the non-intersection area D ₂ in the original geometric feature B. According to the first local feature B _i and the second local features D ₁ and D ₂ , the target geometric feature E of the road surface marking is determined.

FIG. 4 is a schematic diagram of another method for extracting features of a high-precision map according to an embodiment of the present application; this embodiment is an optional solution proposed on the basis of the above-mentioned embodiment. Specifically, in the case that the road area contains at least two road surface markers, the operation of "merging at least two of the original geometric features to obtain the target geometric features of the road surface markers" is refined.

Referring to FIG. 4 , the method for extracting features of a high-precision map provided by this embodiment includes:

S410: Acquire at least two frames of road images collected in different lanes for the same road area.

S420. Extract the original geometric features of the road surface markers in the at least two frames of road images respectively.

S430. If the road area contains at least two road surface markers, determine at least two original geometric features associated with the same road surface marker according to the intersection ratio relationship between the original geometric features.

The road area includes at least two road signs, which means that there are at least two road signs of the same type in the same road area at the same time. Exemplarily, in the case that the road area is an intersection formed by the intersection of two roads, the road area usually includes four pedestrian crossings. In the case that the road area includes at least two road surface markers, it is necessary to determine the original geometric features belonging to the same road surface marker from the original geometric features corresponding to the road surface markers.

Among them, the intersection ratio relationship is used to describe the degree of coincidence between the original geometric features, and the intersection ratio relationship between the original geometric features can be determined by using the Intersection over Union (IOU, Intersection over Union) strategy in the target detection task, which will not be further discussed here. Expand.

At least two original geometric features associated with the same road surface mark are determined according to the intersection ratio relationship between the original geometric features. Specifically, according to the intersection ratio relationship between the original geometric features, the road surface markings corresponding to each original geometrical feature are determined, and the original geometrical features and the corresponding road surface markings are associated. For example, at least two original geometric features whose intersection ratio is greater than a preset threshold are used as the original geometric features corresponding to the same road surface mark.

S440. Perform fusion processing on at least two original geometric features associated with the same road surface markings to obtain target geometrical features of the at least two road surface markings.

The original geometric features corresponding to the same pavement mark are fused to obtain the target geometric features corresponding to the pavement mark. Optionally, in the case that the road area contains at least two pavement marks, fusion processing is performed on the original geometric features associated with each pavement mark, respectively, to obtain target geometric features corresponding to each pavement mark.

In the technical solution of the embodiment of the present application, when the road area contains at least two pavement marks, at least two original geometric features associated with the same pavement mark are determined according to the intersection ratio between the original geometric features; for the same pavement mark The associated at least two original geometric features are fused to obtain at least two target geometric features of road surface markings. In the present application, the fusion processing is performed by determining the original geometric features associated with the pavement markings, thereby ensuring the extraction accuracy of the geometrical features corresponding to the pavement markings.

FIG. 5 is a schematic diagram of another method for extracting features of a high-precision map according to an embodiment of the present application; this embodiment is an optional solution proposed on the basis of the above-mentioned embodiment. Specifically, it is a refinement of the operation "respectively extract the original geometric features of road surface markers in the at least two frames of road images".

Referring to FIG. 5 , the method for extracting features of a high-precision map provided by this embodiment includes:

S510: Acquire at least two frames of road images collected in different lanes for the same road area.

S520. Determine the road top view of the at least two frames of road images.

Since road images are generally front views, and road signs are graphics drawn on the road surface, it is more accurate to extract the geometric features of road signs from the top view of the road than the geometric features extracted from formal images.

Specifically, when determining the road top view, the front view of the road can be converted into a top view of the road based on the principles of perspective transformation and affine transformation, or the point cloud data of the road area can be combined with the front view of the road area to generate the top view of the road. In this embodiment, the point cloud data is preferably combined to generate the overhead view of the road image.

In an optional embodiment, determining the road top view of the at least two frames of road images includes: determining, according to at least two frames of road images and at least two frames of point cloud data collected on the road area in each lane, the The top view of the road corresponding to the road area under each lane.

Among them, the point cloud data and road images of the road area are collected from the same road area in different lanes by the lidar and camera configured on the image acquisition vehicle, respectively.

Specifically, for the point cloud data and road image data collected under each lane, you can first perform semantic segmentation on each frame of road image collected, determine the ground object area in the image, and perform ground point analysis on the point cloud data. Cloud detection; then, according to the calibration relationship between the lidar and the camera, project the detected multi-frame ground point cloud into each frame of road image, and judge whether each projection point is the target projection corresponding to the semantically segmented ground object area point; if yes, select the optimal frame from the multi-frame ground object area corresponding to each target projection point, and perform fusion to generate a road top view. That is, for the data collected in each lane, a frame of road top view is determined. In the embodiment of the present application, the road top view is generated by combining the point cloud data of the road area and the road image, and the position of the ground is determined based on the point cloud data, which can improve the accuracy of feature extraction.

It should be noted that, when determining the road top view of at least two frames of road images in this embodiment, one frame of road image top view may be determined for each frame of image; Frame road image top view. In order to ensure the accuracy of the top view, the second embodiment is preferred.

S530. Extract the pavement marking area from the at least two frames of road top views respectively, and perform feature extraction on the pavement marking area to obtain the original geometric features of the pavement marking in the at least two frames of road images.

Usually, in addition to the road area, the collected road images often include the surrounding environment of the road area, such as vehicles, pedestrians, vegetation and other elements. In the embodiment of the present application, the road surface marking area is used as the area of interest, and the area of interest on the road surface needs to be extracted from the road image. Feature extraction is performed on the pavement marking area, and the features of the polygons that constitute the pavement marking are extracted as the original geometric features of the pavement marking in the road image.

S540. Perform fusion processing on at least two of the original geometric features to obtain a target geometric feature of the road surface marking.

According to the technical solutions of the embodiments of the present application, by determining the road top views of at least two frames of road images, respectively extracting road marking areas from the at least two frames of road top views, and performing feature extraction on the pavement marking areas to obtain road markings in at least two frames of road images The original geometric features of the road surface marking are obtained by fusing at least two original geometric features to obtain the target geometric features of the pavement marking. The road sign is a figure drawn on the road surface, and the embodiment of the present application extracts the geometric features of the road sign from the top view of the road, which effectively improves the extraction accuracy of the geometric feature corresponding to the road sign.

6 is a schematic diagram of an apparatus for extracting features of a high-precision map according to an embodiment of the present application; referring to FIG. 6 , an embodiment of the present application discloses an apparatus 600 for extracting features of a high-precision map. The apparatus 600 may include: a road image acquisition module 610, The original geometric feature extraction module 620 and the original geometric feature fusion module 630.

a road image acquisition module 610, configured to acquire at least two frames of road images collected in different lanes for the same road area;

an original geometric feature extraction module 620, configured to respectively extract the original geometric features of the road surface markers in the at least two frames of road images;

The original geometric feature fusion module 630 is configured to perform fusion processing on at least two of the original geometric features to obtain the target geometric feature of the road surface marking.

The technical solution of the embodiment of the present application is to obtain at least two frames of road images collected from the same road area in different lanes; to extract the original geometric features of the road surface signs in the road images collected under each lane; Fusion processing is performed to obtain the target geometric features of the pavement markings. The present application fuses at least two original geometric features, effectively integrates geometric features obtained from different camera perspectives, and obtains more accurate target geometric features including more road marking geometric feature information, and implements the technical solution provided by the present application. It can effectively solve the problem of inaccurate geometric features of pavement markings extracted from road images due to incomplete and distorted geometric features of road markings in road images.

Optionally, the original geometric feature fusion module 630 includes: an intersection area determination sub-module for determining an intersection area and a non-intersection area of at least two of the original geometric features; a first local feature determination sub-module for Extracting at least two candidate local features of the original geometric features in the intersection region, and determining the first local feature of the intersection region according to the confidence of the candidate local features in the intersection region; the second local feature; an extraction submodule for extracting at least two second local features of the original geometric features in the non-intersection area; a target geometric feature determination submodule for extraction according to the first local feature and the second local feature , and determine the target geometric feature of the road marking.

Optionally, the first local feature determination sub-module includes: a sub-region dividing unit, used to divide the intersection area into at least two sub-regions; a first local feature determination unit, used for according to the candidate local features At the confidence level of each sub-region, the first local feature of the intersection region is determined.

Optionally, the device further includes: a confidence level determination module, which is specifically configured to determine whether the candidate local feature is in the Confidence for the intersection region or subregion.

Optionally, if the road area contains at least two road surface markers, the original geometric feature fusion module 630 includes: a sub-module associated with the original geometric feature, for determining the same road surface according to the intersection ratio relationship between the original geometric features. at least two associated original geometric features are identified; a target geometric feature determination sub-module is used to fuse at least two original geometric features associated with the same road surface marker to obtain the target geometric features of the at least two road surface markers.

Optionally, the original geometric feature extraction module 620 includes: a road top view determination sub-module for determining the road top view of the at least two frames of road images; a road surface identification area feature extraction sub-module for respectively extracting from the at least two frames. The road surface marking area is extracted from the top view of the road, and feature extraction is performed on the road surface marking area to obtain the original geometric features of the road surface marking in the at least two frames of road images.

Optionally, the road top view determination sub-module is specifically configured to determine, according to at least two frames of road images and at least two frames of point cloud data collected for the road area in each lane, the corresponding road area of the road area under each lane. Road top view.

Optionally, the road marking is a road marking across lanes.

The high-precision map feature extraction device provided by the embodiment of the present application can execute the high-precision map feature extraction method provided by any embodiment of the present application, and has corresponding functional modules and beneficial effects for executing the high-precision map feature extraction method.

In the technical solution of this application, the relevant data of any application involved (such as the authorization code of the application, the application identifier and the authorization parameters of the application, etc.), the relevant data of the open platform (such as historical access records) and the third-party organization (such as The acquisition, storage and application of relevant data of target institutions and other institutions, etc., are in compliance with relevant laws and regulations, and do not violate public order and good customs.

In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of the user's personal information involved are in compliance with relevant laws and regulations, and do not violate public order and good customs.

"It should be noted that the head model in this embodiment is not a head model for a specific user, and cannot reflect the personal information of a specific user";

"It should be noted that the two-dimensional face image in this embodiment comes from a public dataset" and so on.

According to the embodiments of the present application, the present application further provides an electronic device, a readable storage medium, and a computer program product.

FIG. 7 shows a schematic block diagram of an example electronic device 700 that may be used to implement embodiments of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the application described and/or claimed herein.

As shown in FIG. 7 , the device 700 includes a computing unit 701 that can be executed according to a computer program stored in a read only memory (ROM) 702 or loaded into a random access memory (RAM) 703 from a storage unit 708 Various appropriate actions and handling. In the RAM 703, various programs and data necessary for the operation of the device 700 can also be stored. The computing unit 701 , the ROM 702 , and the RAM 703 are connected to each other through a bus 704 . An input/output (I/O) interface 705 is also connected to bus 704 .

Various components in the device 700 are connected to the I/O interface 705, including: an input unit 706, such as a keyboard, mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, an optical disk, etc. ; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 709 allows the device 700 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

Computing unit 701 may be various general-purpose and/or special-purpose processing components with processing and computing capabilities. Some examples of computing units 701 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 executes the various methods and processes described above, such as a high-precision map feature extraction method. For example, in some embodiments, the HD map feature extraction method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 708 . In some embodiments, part or all of the computer program may be loaded and/or installed on device 700 via ROM 702 and/or communication unit 709 . When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the high-precision map feature extraction method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the HD map feature extraction method by any other suitable means (eg, by means of firmware).

Various implementations of the systems and techniques described herein above may be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips system (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.

Program code for implementing the methods of the present application may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, performs the functions/functions specified in the flowcharts and/or block diagrams. Action is implemented. The program code may execute entirely on the machine, partly on the machine, partly on the machine and partly on a remote machine as a stand-alone software package or entirely on the remote machine or server.

In the context of this application, a machine-readable medium may be a tangible medium that may contain or store the program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), blockchain networks, and the Internet.

A computer system can include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or a cloud host. It is a host product in the cloud computing service system to solve the traditional physical host and VPS services, which are difficult to manage and weak in business scalability. defect. The server can also be a server of a distributed system, or a server combined with a blockchain.

Artificial intelligence is the study of making computers to simulate certain thinking processes and intelligent behaviors of people (such as learning, reasoning, thinking, planning, etc.), both hardware-level technology and software-level technology. AI hardware technologies generally include technologies such as sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing; AI software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, and machine learning/depth Learning technology, big data processing technology, knowledge graph technology and other major directions.

Cloud computing refers to accessing elastically scalable shared physical or virtual resource pools through the network. Resources can include servers, operating systems, networks, software, applications and storage devices, etc., and can be self-service on demand and on demand. A technical system for deploying and managing resources in a way. Through cloud computing technology, it can provide efficient and powerful data processing capabilities for artificial intelligence, blockchain and other technical applications and model training.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present application can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in the present application can be achieved, no limitation is imposed herein.

The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims (19)

1. A high-precision map feature extraction method, comprising: Obtain at least two frames of road images collected in different lanes for the same road area; respectively extracting the original geometric features of the road surface markings in the at least two frames of road images; Perform fusion processing on at least two of the original geometric features to obtain the target geometric features of the road surface marking. 2. The method according to claim 1, wherein the performing fusion processing on at least two of the original geometric features to obtain the target geometric features of the road surface marking, comprising: determining an intersection area and a non-intersection area of at least two of the original geometric features; extracting at least two candidate local features of the original geometric features in the intersection region, and determining the first local feature of the intersection region according to the confidence of the candidate local features in the intersection region; extracting at least two second local features of the original geometric features in the non-intersection area; Based on the first local feature and the second local feature, a target geometric feature of the road surface marking is determined. 3. The method according to claim 2, wherein the determining the first local feature of the intersection region according to the confidence of the candidate local features in the intersection region comprises: dividing the intersection area into at least two sub-areas; The first local feature of the intersection region is determined according to the confidence level of the candidate local feature in each subregion. 4. The method of claim 2 or 3, further comprising: According to the distance value between the position of the intersection region or the sub-region and the collection position of the candidate local feature, the confidence level of the candidate local feature in the intersection region or sub-region is determined. 5 . The method according to claim 1 , wherein, if the road area contains at least two road surface markers, performing fusion processing on at least two of the original geometric features to obtain the target geometric features of the road surface markers, comprising: 6 . : Determine at least two original geometric features associated with the same pavement mark according to the intersection ratio relationship between the original geometric features; Perform fusion processing on at least two original geometric features associated with the same road surface markings to obtain target geometrical features of the at least two road surface markings. 6. The method according to claim 1, wherein extracting the original geometric features of road surface markings in the at least two frames of road images respectively comprises: determining a road top view of the at least two frames of road images; The pavement marking area is extracted from the at least two frames of road top views, respectively, and feature extraction is performed on the pavement marking area to obtain the original geometric features of the pavement marking in the at least two frames of road images. 7. The method of claim 6, wherein determining the road top view of the at least two frames of road images comprises: According to at least two frames of road images and at least two frames of point cloud data collected for the road area in each lane, a road top view corresponding to the road area under each lane is determined. 8. The method of any of claims 1-7, wherein the pavement marking is a cross-lane pavement marking. 9. A high-precision map feature extraction device, comprising: a road image acquisition module, used for acquiring at least two frames of road images collected in different lanes for the same road area; an original geometric feature extraction module, configured to extract the original geometric features of the road surface markings in the at least two frames of road images respectively; The original geometric feature fusion module is configured to perform fusion processing on at least two of the original geometric features to obtain the target geometric feature of the road surface marking. 10. The apparatus according to claim 9, wherein the original geometric feature fusion module comprises: an intersection area determination submodule, configured to determine an intersection area and a non-intersection area of at least two of the original geometric features; The first local feature determination submodule is used to extract at least two candidate local features of the original geometric features in the intersection region, and determine the intersection region according to the confidence of the candidate local features in the intersection region The first local feature of ; a second local feature extraction submodule, configured to extract the second local features of at least two of the original geometric features in the non-intersection area; A target geometric feature determination sub-module, configured to determine the target geometric feature of the road surface mark according to the first local feature and the second local feature. 11. The apparatus according to claim 10, wherein the first local feature determination sub-module comprises: a sub-region dividing unit, configured to divide the intersection region into at least two sub-regions; The first local feature determining unit is configured to determine the first local feature of the intersection region according to the confidence of the candidate local feature in each subregion. 12. The apparatus according to claim 10 or 11, further comprising: a confidence level determination module, which is specifically configured to, according to the distance value between the position of the intersection area or the position of the sub-area and the collection position of the candidate local feature, A confidence level of the candidate local feature in the intersection region or sub-region is determined. 13. The device according to claim 9, wherein, if the road area includes at least two road surface markers, the original geometric feature fusion module comprises: The associated original geometric feature determination sub-module is used to determine at least two original geometric features associated with the same road surface mark according to the intersection ratio relationship between the original geometric features; The target geometric feature determination sub-module is configured to perform fusion processing on at least two original geometric features associated with the same road surface mark to obtain the target geometric features of the at least two road surface marks. 14. The apparatus according to claim 9, wherein the original geometric feature extraction module comprises: a road top view determination submodule, configured to determine the road top view of the at least two frames of road images; The feature extraction sub-module of pavement marking area is used to extract the pavement marking area from the at least two frames of road top view respectively, and perform feature extraction on the pavement marking area to obtain the original geometry of the pavement marking in the at least two frames of road images feature. 15. The apparatus according to claim 14, wherein the road top view determination sub-module is specifically configured to determine the The top view of the road corresponding to the road area under each lane. 16. The apparatus of any of claims 9-15, wherein the pavement marking is a cross-lane pavement marking. 17. An electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the execution of any of claims 1-8 Methods. 18. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any of claims 1-8. 19. A computer program product comprising a computer program which, when executed by a processor, implements the method of any of claims 1-8.

Images