High-precision map generation method and device, electronic equipment and medium

Abstract

The present disclosure provides a high-precision map generation method and device, electronic equipment, computer readable storage medium and computer program product, relates to the field of computers, and particularly relates to the fields of high-precision maps, autonomous driving and intelligent transportation. The implementation scheme is as follows: obtaining point cloud, image and pose data of a target scene; performing semantic segmentation on the point cloud based on the pose data; performing feature extraction on the segmented point cloud to obtain first ground element features; performing semantic segmentation on the image, and performing feature extraction on the segmented image based on the segmented point cloud and the pose data to obtain second ground element features; performing feature extraction on the image based on the segmented point cloud to obtain facade element features; and performing feature fusion on the features corresponding to the same elements to generate a high-precision map based on the fused element features.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of computers, in particular to the field of high-definition map, automatic driving and intelligent transportation, and more particularly to a high-definition map generation method and device, electronic equipment, computer readable storage medium and computer program product. BACKGROUND

[0002] Artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors of human beings (such as learning, reasoning, thinking, planning, etc.), which includes both hardware technology and software technology. Artificial intelligence hardware technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, etc. Artificial intelligence software technology mainly includes computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, etc.

[0003] In the whole automatic driving process, the high-definition map is one of the most important components, especially for L4 / L5 level automatic driving, the high-definition map depicts the static prior information in the whole scene. Although the high-definition map is of great significance in automatic driving, the production of the high-definition map is still a difficult problem to be solved in the industry. SUMMARY

[0004] The present disclosure provides a high-definition map generation method and device, electronic equipment, computer readable storage medium and computer program product.

[0005] According to one aspect of this disclosure, a method is provided for acquiring point clouds and images of a target scene, as well as pose data corresponding to a point cloud acquisition device and an image acquisition device; semantic segmentation is performed on the point cloud based on the pose data of the point cloud acquisition device to obtain a semantically segmented point cloud; feature extraction is performed on the semantically segmented point cloud to obtain a first ground feature feature corresponding to ground features, wherein the ground features include at least one of ground line features and ground point features, the ground line features corresponding to entity targets that are linearly expressed in a two-dimensional horizontal plane, and the ground point features corresponding to entity targets that are planarly expressed in a two-dimensional horizontal plane; semantic segmentation is performed on the image to obtain semantic features. The image is segmented; features are extracted from the semantically segmented image based on the point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device to obtain second ground feature features corresponding to ground features, wherein the second ground feature features include at least one of the ground line feature features and the ground point feature features; features are extracted from the image based on the segmented point cloud to obtain facade feature features corresponding to facade features, wherein the facade features correspond to entity targets expressed in three-dimensional space; and the features corresponding to the same features obtained are fused to generate a high-precision map based on the fused feature features.

[0006] According to another aspect of this disclosure, a high-precision map generation apparatus is provided, comprising: an acquisition unit configured to acquire point clouds and images of a target scene, and pose data corresponding to a point cloud acquisition device and an image acquisition device; a first semantic segmentation unit configured to perform semantic segmentation on the point cloud based on the pose data of the point cloud acquisition device to obtain a semantically segmented point cloud; a first feature extraction unit configured to extract features from the semantically segmented point cloud to obtain first ground feature features corresponding to ground features, wherein the ground features include at least one of ground line features and ground point features, the ground line features corresponding to entity targets expressed linearly in a two-dimensional horizontal plane, and the ground point features corresponding to entity targets expressed planarly in a two-dimensional horizontal plane; and a second semantic segmentation unit configured to perform semantic segmentation on the image. The system comprises: a semantic segmentation unit to obtain a semantically segmented image; a second feature extraction unit configured to extract features from the semantically segmented image based on the point cloud obtained from the semantic segmentation, the relative pose data between the point cloud acquisition device and the image acquisition device, to obtain second ground feature features corresponding to ground features, wherein the second ground feature features include at least one of ground line feature features and ground point feature features; a third feature extraction unit configured to extract features from the image based on the segmented point cloud to obtain facade feature features corresponding to facade features, wherein the facade features correspond to entity targets expressed in three-dimensional space; and a feature fusion unit configured to fuse the features corresponding to the same obtained features to generate a high-precision map based on the fused feature features.

[0007] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor to enable the at least one processor to perform the methods described in this disclosure.

[0008] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods described in this disclosure.

[0009] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods described in this disclosure.

[0010] According to one or more embodiments of this disclosure, by combining point cloud and image data, the identification of multiple types of entity targets is realized, thereby forming a data closed loop and saving the high cost of feature labeling in map production.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings exemplify embodiments and form part of the specification, serving together with the textual description to explain exemplary implementations of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0013] Figure 1 A schematic diagram of an exemplary system in which the various methods described herein may be implemented according to embodiments of the present disclosure is shown;

[0014] Figure 2 A flowchart of a high-precision map generation method according to an embodiment of the present disclosure is shown;

[0015] Figure 3 A flowchart illustrating the acquisition of corresponding map features based on point clouds according to an embodiment of the present disclosure is shown;

[0016] Figure 4 A flowchart illustrating image-based acquisition of ground feature characteristics according to an embodiment of the present disclosure is shown;

[0017] Figure 5 A flowchart illustrating image-based acquisition of facade feature characteristics according to an embodiment of the present disclosure is shown;

[0018] Figures 6-8 A flowchart is shown for geometric feature re-extraction from a stereo detection frame of facade elements according to a corresponding embodiment of the present disclosure;

[0019] Figure 9 A structural block diagram of a high-precision map generation apparatus according to an embodiment of the present disclosure is shown; and

[0020] Figure 10 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to limit the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0023] The terminology used in the description of the various examples described in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. Furthermore, the term "and / or" as used in this disclosure covers any one of the listed items and all possible combinations thereof.

[0024] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0025] Figure 1 A schematic diagram of an exemplary system 100 in which the various methods and apparatus described herein can be implemented according to embodiments of this disclosure is shown. Reference Figure 1 The system 100 includes one or more client devices 101, 102, 103, 104, 105 and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105 and 106 can be configured to execute one or more applications.

[0026] In embodiments of this disclosure, server 120 may run one or more services or software applications that enable the execution of methods for generating high-precision maps.

[0027] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtual and virtual environments. In some embodiments, these services may be provided as web-based services or cloud services, such as to users of client devices 101, 102, 103, 104, 105, and / or 106 under a Software as a Service (SaaS) model.

[0028] exist Figure 1In the configuration shown, server 120 may include one or more components that implement the functions performed by server 120. These components may include software components, hardware components, or combinations thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 can sequentially interact with server 120 using one or more client applications to utilize the services provided by these components. It should be understood that various different system configurations are possible and may differ from system 100. Therefore, Figure 1 This is an example of a system used to implement the various methods described herein, and is not intended to be limiting.

[0029] Users can use client devices 101, 102, 103, 104, 105, and / or 106 to acquire or receive corresponding data. The client devices can provide an interface that allows users to interact with the client devices. The client devices can also output information to the user through this interface. Although... Figure 1 Only six client devices are described, but those skilled in the art will understand that this disclosure can support any number of client devices.

[0030] Client devices 101, 102, 103, 104, 105, and / or 106 may include various types of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptops), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors, or other sensing devices. These computer devices can run various types and versions of software applications and operating systems, such as Microsoft Windows, Apple iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as Google Chrome OS); or include various mobile operating systems, such as Microsoft Windows Mobile OS, iOS, Windows Phone, and Android. Portable handheld devices may include cellular phones, smartphones, tablets, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Gaming systems may include various handheld gaming devices, internet-enabled gaming devices, etc. Client devices are capable of executing various applications, such as various internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols.

[0031] Network 110 can be any type of network well known to those skilled in the art, and can use any of a variety of available protocols (including but not limited to TCP / IP, SNA, IPX, etc.) to support data communication. By way of example only, one or more networks 110 can be a local area network (LAN), an Ethernet-based network, a token ring network, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0032] Server 120 may include one or more general-purpose computers, special-purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running a virtual operating system, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices for servers). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0033] The computing unit in server 120 can run one or more operating systems, including any of the aforementioned operating systems and any commercially available server operating system. Server 120 can also run any of a variety of additional server applications and / or middleware applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0034] In some implementations, server 120 may include one or more applications to analyze and merge data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105, and 106. Server 120 may also include one or more applications to display data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and 106.

[0035] In some implementations, server 120 can be a server for a distributed system or a server integrated with blockchain. Server 120 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in the cloud computing service system, designed to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0036] System 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as acquired point clouds, images, etc. Databases 130 may reside in various locations. For example, a database used by server 120 may be local to server 120, or it may be located remotely to server 120 and may communicate with server 120 via a network-based or dedicated connection. Databases 130 may be of different types. In some embodiments, the database used by server 120 may be, for example, a relational database. One or more of these databases may store, update, and retrieve data from and from the databases in response to commands.

[0037] In some embodiments, one or more of the databases 130 may also be used by an application to store application data. The databases used by the application may be of different types, such as key-value stores, object stores, or regular stores supported by a file system.

[0038] Figure 1 The system 100 can be configured and operated in various ways to enable the application of the various methods and apparatus described in this disclosure.

[0039] High-precision maps depict static prior information about the entire scene, including various information such as lane geometry / attributes, traffic lights, pedestrian crossings, stop lines, traffic poles, and traffic signs. All algorithm templates on autonomous vehicles, including localization, perception, and decision-making modules, rely on high-precision maps. It can be said that high-precision maps are the foundation of the entire L4 / L5 level autonomous driving system.

[0040] Currently, the production of high-precision maps remains a critical challenge in the industry, primarily due to two main difficulties: first, the creation of highly accurate 3D maps; and second, an efficient process for annotating elements within these maps. The second challenge directly impacts the production efficiency of high-precision maps, further affecting their production costs and consequently hindering the commercialization and mass production of autonomous driving. Therefore, for all autonomous driving companies aiming for mass production, the production efficiency of high-precision maps is of paramount importance.

[0041] In actual production, there are two main ways to improve annotation: first, by improving the operation of annotation tools; and second, by using automated recognition to provide prior annotation results, followed by manual correction to improve annotation efficiency. Theoretically, improving efficiency through annotation tools will have certain limitations, since manual work is unavoidable no matter how the tools are improved. However, automated recognition can achieve truly manual work-free operation. For simple road sections, the prior annotation results provided by automated recognition can achieve results without human intervention, which greatly improves operational efficiency.

[0042] Given the crucial role of automated annotation in the entire high-precision map production process, all autonomous driving companies and map providers are involved in automated annotation. However, currently available methods are mostly designed for identifying one type or category of targets, failing to create a complete production line capable of producing all target elements. Almost all available methods operate on a single sensor, but a single sensor cannot provide sufficient redundancy or guarantee full coverage of target geometry and attributes. Furthermore, multiple sensors cannot interact effectively to ensure target recall. Current methods only focus on initial target identification; for example, traffic light detection primarily focuses on the recognition of traffic lights in images. However, in practical high-precision map applications, the 3D location and attributes of traffic lights are more critical, making post-recognition processing equally important. Finally, the overall design does not consider data closure and incurs high annotation costs.

[0043] Therefore, an embodiment of the present disclosure provides a method for generating high-precision maps. Figure 2 A flowchart of a high-precision map generation method according to an embodiment of the present disclosure is shown, such as... Figure 2As shown, method 200 includes: acquiring point cloud and image of target scene, and pose data corresponding to point cloud acquisition device and image acquisition device (step 210); performing semantic segmentation on point cloud based on pose data of point cloud acquisition device to obtain semantically segmented point cloud (step 220); performing feature extraction on semantically segmented point cloud to obtain first ground feature features corresponding to ground features, wherein ground features include at least one of ground line features and ground point features, the ground line features corresponding to entity targets that are linearly expressed in a two-dimensional horizontal plane, and the ground point features corresponding to entity targets that are planarly expressed in a two-dimensional horizontal plane (step 230); performing semantic segmentation on image to obtain semantically segmented image. (Step 240); Based on the semantically segmented point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device, feature extraction is performed on the semantically segmented image to obtain second ground feature features corresponding to ground features, wherein the second ground feature features include at least one of the ground line feature features and the ground point feature features (Step 250); Based on the segmented point cloud, feature extraction is performed on the image to obtain facade feature features corresponding to facade features, wherein the facade features correspond to entity targets expressed in three-dimensional space (Step 260); and the features corresponding to the same obtained features are fused to generate a high-precision map based on the fused feature features (Step 270).

[0044] According to embodiments of this disclosure, by combining point cloud and image data, the identification of multiple types of entity targets is achieved, thereby forming a data closed loop and saving the high cost of feature labeling in map production.

[0045] In this disclosure, the target scene can be a road and its surrounding environment that allows vehicles to travel. Alternatively, it can be the interior of a building that allows robots to move, a flight path that allows drones to fly at low altitudes and its surrounding environment, etc., without limitation. The target scene includes at least one target entity corresponding to a map element. The target entity is a feature entity that actually exists in the target scene, and the map element corresponding to the feature entity is a feature presented in the map matched to the target scene.

[0046] Specifically, map elements and their corresponding entity elements will vary depending on the application scenario. For example, in a scenario assisting vehicle navigation, map elements include lane lines, ground markings, curbs, fences, and traffic signs. Correspondingly, entity elements can include real-world entities such as lane lines, ground markings, curbs, fences, and traffic signs. Alternatively, in a scenario assisting low-altitude drone flight, map elements include streetlights, vegetation, buildings, and traffic signs. Correspondingly, entity elements can include real-world entities such as directional lights, vegetation, buildings, and traffic signs.

[0047] In some examples, the raw point cloud and images acquired during the data acquisition process are obtained, along with the pose data of the image acquisition device and the point cloud acquisition device. For example, the image acquisition device can include a variety of devices, such as a visual camera, an infrared camera, a camera for ultraviolet or X-ray imaging, etc. The point cloud acquisition device can also include a variety of devices, such as LiDAR, millimeter-wave radar, ultrasonic sensors, etc. Different devices can provide different detection accuracies and ranges.

[0048] According to some embodiments, multiple sensors for data acquisition can be located on at least one acquisition vehicle or roadside equipment. For example, while the acquisition vehicle is in motion, multiple sensors can be mounted in front of, behind, or other locations on the vehicle to collect data on the surrounding road conditions.

[0049] The map elements in this embodiment include ground line elements, ground point elements, and elevation elements. Ground line elements can be understood as targets that can be linearly represented on a two-dimensional horizontal plane. For example, line elements include road elements that can be represented by straight lines and curves, such as lane lines and road boundaries. Ground point elements can be understood as targets that can be planar represented on a two-dimensional horizontal plane, such as stop lines, sidewalks, no-parking zones, and safety islands. Elevation elements can be understood as targets represented in three-dimensional space, such as traffic lights, poles, and signs.

[0050] According to some embodiments, semantic segmentation of the point cloud based on corresponding pose data to obtain a semantically segmented point cloud may include: inputting the point cloud and pose data of the point cloud acquisition device into a corresponding network model to obtain the segmented point cloud. In some examples, semantic segmentation can be performed on the point cloud to obtain the ground feature category corresponding to each point; based on the ground feature category of each point, segmented point cloud data corresponding to different ground feature categories can be extracted respectively.

[0051] Specifically, point cloud data can be input into a pre-trained semantic segmentation model to obtain the ground feature category corresponding to each point. Each point can be labeled with a category tag, which indicates the ground feature category of the point. For example, lane lines are considered a type of ground feature. In some examples, the semantic segmentation model can be a network model such as SparseConv, DeepLabV3+, or RandLANet. Multiple sample point cloud data can be acquired. For each sample point cloud data, each point in the sample point cloud data is labeled to obtain a ground feature category label for each point. This ground feature category label can be understood as a category ground truth. Then, a sample point cloud data and its corresponding category ground truth are used as a sample pair. The neural network model is trained based on multiple sample pairs to obtain the semantic segmentation model.

[0052] In some embodiments, feature extraction can be performed directly on the point cloud after semantic segmentation to obtain the first ground feature features corresponding to the ground features. In some examples, after semantic segmentation, the segmented ground line features can be vectorized to extract vector lines. Vectorization is easily implemented and will not be elaborated here.

[0053] According to some embodiments, the point cloud includes a point cloud obtained by repeatedly capturing data from the target scene. Additionally or alternatively, such as... Figure 3 As shown, directly extracting features from the semantically segmented point cloud to obtain the first ground feature corresponding to the ground feature may include: extracting features from the semantically segmented point cloud to obtain features corresponding to preset road signs (step 310); stitching together the point cloud obtained from multiple acquisitions based on the features corresponding to the preset road signs to obtain a stitched point cloud (step 320); generating a reflectance map based on the stitched point cloud (step 330); and obtaining the first ground feature based on the reflectance map (step 340).

[0054] In some examples, the preset road markings may include lane lines, curbs, guardrails, poles, signs, tree trunks, etc., and the extracted feature information of the road markings may include information such as the location and size of the road markings. Therefore, the point cloud is stitched together based on the extracted feature information of the road markings to obtain more accurate point cloud data. The stitched point cloud data can be used to produce map data and obtain reflectance maps. Those skilled in the art can easily implement reflectance maps based on point cloud data, which will not be elaborated upon here.

[0055] Specifically, according to some embodiments, both the second ground feature and the first ground feature include the ground line feature. Obtaining the first ground feature based on the reflectance map includes: inputting the ground line feature from the second ground feature and the reflectance map into a trained first network model to obtain the ground line feature from the first ground feature.

[0056] Specifically, in some examples, the input to the first network model is a reflectance map generated based on a stitched point cloud and ground line feature features identified from the image. The output is the geometry of the identified ground line features, such as lane lines and road boundaries. Geometric correction is performed on the ground line features identified from the image within the reflectance map to obtain more accurate ground line feature features.

[0057] According to some embodiments, both the second ground feature and the first ground feature include the ground point feature. Obtaining the first ground feature based on the reflectance map includes: inputting the ground point feature from the second ground feature and the reflectance map into a trained second network model to obtain the ground point feature from the first ground feature.

[0058] Specifically, in some examples, the input to the second network model is a reflectance map generated based on the stitched point cloud and ground point feature features identified from the image, and the output is the complete geometric features of the identified ground point features. Geometric correction is performed on the ground point features identified from the image in the reflectance map to obtain more accurate ground point feature features.

[0059] According to some embodiments, the second ground feature includes the ground point feature. Inputting the ground point feature and the reflectance map from the second ground feature into a trained second network model includes: clustering the pixels corresponding to the ground point features determined based on the image to obtain cluster centers; and expanding a preset range based on the cluster centers, using the expanded range as the ground point feature to be input into the second network model.

[0060] To ensure the completeness of the extracted target features, extraction needs to be carried out over a large area (e.g., covering the entire intersection) to maximize the integrity of the extracted results. Considering computing power, prior ground point features extracted from upstream can be used for clustering, and the receptive field can be expanded based on the cluster center points. This reduces the computation of irrelevant areas and ensures the geometric accuracy of the target features through detection operations on the reflectance map.

[0061] It is understood that the first and second network models can be any suitable network models, such as convolutional neural network models, residual neural networks, etc., without any restrictions.

[0062] In some examples, pixel-based clustering may include superpixel clustering, K-means clustering, etc., and there are no restrictions here.

[0063] According to some embodiments, such as Figure 4 As shown, the process of extracting features from the semantically segmented image based on the semantically segmented point cloud and the corresponding pose data to obtain the second ground feature includes: inputting the semantically segmented point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device, and the semantically segmented image into a trained third network model to obtain a bird's-eye view (step 410); performing semantic segmentation on the bird's-eye view to obtain a semantically segmented bird's-eye view (step 420); and extracting features from the semantically segmented bird's-eye view to obtain the second ground feature (step 430).

[0064] In some examples, commonly used network models for image semantic segmentation may include, but are not limited to, ResNet, Deeplabv3p, MobileNet, etc., and are not limited here. Furthermore, it is understood that obtaining the corresponding bird's-eye view based on point clouds can be easily achieved by those skilled in the art. Based on this, obtaining a bird's-eye view from point clouds, relative pose data, and images can be easily achieved through a trained network model, and will not be elaborated further here.

[0065] According to some embodiments, such as Figure 5 As shown, the feature extraction of the image based on the segmented point cloud to obtain facade element features includes: identifying the facade elements based on the image to obtain detection boxes of the facade elements (step 510); performing 3D reconstruction on the detection boxes of the facade elements to obtain stereo detection boxes of the facade elements (step 520); projecting the stereo detection boxes of the facade elements onto the corresponding point cloud to determine the overlapping area (step 530); and in response to determining that the overlapping area is greater than a first preset threshold, using the stereo detection boxes of the facade elements obtained by 3D reconstruction as the obtained facade element features (step 540).

[0066] For facade elements such as traffic lights, traffic signs, and man-made poles, image-based feature reconstruction and point cloud geometric verification were used to ensure the consistency (rich texture, high recall) and accuracy (geometric accuracy) of the extracted target elements with the vehicle-side data.

[0067] According to some embodiments, extracting features from the image based on the segmented point cloud to obtain facade element features includes: in response to determining that the overlapping area is not greater than the first preset threshold, performing geometric feature re-extraction on the stereo detection box of the facade element based on the segmented point cloud to obtain a corrected stereo detection box of the facade element as the obtained facade element features.

[0068] When the difference between the facade elements determined based on the image and the corresponding elements in the point cloud is large, geometric features can be further extracted based on the point cloud to ensure the integrity and accuracy of the obtained facade elements.

[0069] According to some embodiments, the facade element includes rod-like structures. Therefore, as... Figure 6 As shown, the geometric feature re-extraction of the stereo detection box of the facade element based on the segmented point cloud may include: dividing at least a portion of the stereo detection box into multiple sub-stereo detection boxes along a first direction (step 610); for each of the multiple sub-stereo detection boxes (step 620): projecting the points in the point cloud corresponding to the sub-stereo detection box onto a plane perpendicular to the first direction to determine the radius of the smallest circumcircle surrounding the projection area (step 6201); in response to determining that the radius is within a preset range, determining the points corresponding to the sub-stereo detection box as seed points (step 6202); merging the seed points corresponding to the multiple sub-stereo detection boxes to form a first seed point set, wherein the Euclidean distance between any two seed points in the first seed point set is less than a preset distance threshold (step 630); and performing region growing based on the first seed point set to obtain the corrected stereo detection box of the facade element (step 640).

[0070] According to some embodiments, region growing based on the first seed point set includes: determining a first detection box surrounding at least one seed point in the first seed point set, such that during region growing, the number of points within the first detection box and the number of points outside the first detection box are greater than a second preset threshold. This ensures the geometric accuracy of the facade elements obtained after region growing.

[0071] For example, rod-shaped objects in a target scene identified from an image may be occluded by trees, flags, or other objects, causing distortion in the reconstructed stereo detection box and preventing accurate reconstruction of the target entity. However, by performing region growing based on determined seed points, the geometry of the occluded rod-shaped objects can be automatically reconstructed.

[0072] According to some embodiments, the facade elements include sign-like structures. For example... Figure 7As shown, the geometric feature re-extraction of the stereo detection box of the facade element based on the segmented point cloud includes: determining the dimensional features of the point cloud corresponding to the stereo detection box to determine the set of two-dimensional points (step 710); clustering the points in the set of two-dimensional points to obtain one or more clustered point cloud clusters (step 720); filtering the one or more point cloud clusters based on preset point cloud cluster size information to obtain filtered point cloud clusters (step 730); performing plane fitting on the points in the filtered point cloud clusters to determine the center position and normal direction of the fitted plane (step 740); and determining the corrected stereo detection box of the facade element based on the center position and normal direction (step 750).

[0073] According to some embodiments, the point cloud cluster size information includes at least one of the height and width of the point cloud cluster. By filtering the obtained point cloud clusters based on preset point cloud cluster size information, point clouds that are clearly not the target elements to be extracted, such as trees, trash cans, and other entities that may interfere with the recognition effect, can be filtered out, thereby improving the recognition effect while saving computing power.

[0074] In some examples, the acquired point cloud can be preprocessed to obtain the dimensional features of the points, such as one-dimensional, two-dimensional, and three-dimensional points. For a sign-shaped object, the two-dimensional points are clustered based on distance to obtain clustered point cloud clusters. Plane fitting is then performed on the clustered point cloud clusters, using any suitable algorithm, including but not limited to SVD (Singular Value Decomposition).

[0075] According to some embodiments, the facade elements include traffic lights. For example... Figure 8 As shown, the geometric feature re-extraction of the stereo detection boxes of the facade elements based on the segmented point cloud includes: clustering the points in the segmented point cloud based on a first distance to determine a first cluster center point and a first normal direction (step 810); clustering the points of the same category and the same first normal direction based on a second distance to determine a second cluster center point and a second normal direction, wherein the second distance is greater than the first distance, and wherein the category is determined based on semantic segmentation (step 820); determining a first detection box based on the second cluster center point and the second normal direction (step 830); and determining one or more second detection boxes located within the first detection box based on the first cluster center point and the first normal direction, as the corrected stereo detection boxes of the facade elements (step 840).

[0076] Specifically, in the first clustering, information about each traffic light can be roughly obtained. However, due to issues such as foliage obstruction, the location and range of the traffic lights obtained by clustering are not precise and contain some errors. In the second clustering, based on pre-determined category information and normal direction, points of the same category are clustered together to generate large detection boxes that roughly surround each traffic light. Within these detection boxes, smaller detection boxes corresponding to each traffic light are then determined.

[0077] In some examples, after geometric feature re-extraction, the re-extracted stereo detection boxes can be compared with the detection boxes of the corresponding elements in the point cloud. As mentioned above, the re-extracted stereo detection boxes are projected onto the corresponding point cloud to determine the overlapping area, thereby further verifying the target recognition effect.

[0078] According to some embodiments, both the first ground feature feature and the second ground feature feature include ground line features. Feature fusion of all obtained feature features includes at least one of the following: determining a predetermined number of second ground line features closest to the first ground line feature; voting on the predetermined number of second ground line features based on a preset first attribute; and assigning the first attribute of the corresponding ground line feature in the second ground line features to the first ground feature based on the voting results; determining a predetermined number of first ground line features closest to the second ground line feature; and adjusting the second attribute of the second ground line feature based on a preset second attribute corresponding to the predetermined number of first ground line features. The first ground line feature is the ground line feature corresponding to the ground line feature feature in the first ground feature feature, and the second ground line feature is the ground line feature corresponding to the ground line feature feature in the second ground feature feature.

[0079] In some examples, the first attribute can be color (e.g., white, yellow), dashed line, single solid line, double solid line, etc.; the second attribute can be geometric structure, location, etc., without limitation. By fusion of features, the advantages of each sensor are fully utilized to more accurately determine map information, thereby improving map accuracy.

[0080] In some examples, the order of the operations can be disregarded when performing ground line feature fusion based on the two operations described above.

[0081] In specific application scenarios, high-precision maps are an indispensable and crucial component for achieving assisted driving or autonomous driving. They can realistically recreate the target scene, thereby improving positioning accuracy and effectively compensating for the shortcomings of environmental perception devices. The methods provided by the various embodiments of this invention not only fully guarantee the accuracy of high-precision maps but also effectively reduce their production costs and improve their production efficiency, facilitating large-scale mass production of high-precision maps.

[0082] According to embodiments of this disclosure, such as Figure 9 As shown, a high-precision map generation device 900 is also provided, comprising: an acquisition unit 910 configured to acquire point clouds and images of a target scene, as well as pose data corresponding to the point cloud acquisition device and the image acquisition device; a first semantic segmentation unit 920 configured to perform semantic segmentation on the point cloud based on the pose data of the point cloud acquisition device to obtain semantically segmented point clouds; a first feature extraction unit 930 configured to extract features from the semantically segmented point cloud to obtain first ground feature features corresponding to ground features, wherein the ground features include at least one of ground line features and ground point features, the ground line features corresponding to entity targets that are linearly expressed in a two-dimensional horizontal plane, and the ground point features corresponding to entity targets that are planarly expressed in a two-dimensional horizontal plane; and a second semantic segmentation unit 940 configured to perform semantic segmentation on the image. The system performs semantic segmentation to obtain a semantically segmented image; a second feature extraction unit 950 is configured to extract features from the semantically segmented image based on the point cloud after semantic segmentation, the relative pose data between the point cloud acquisition device and the image acquisition device, to obtain second ground feature features corresponding to ground features, wherein the second ground feature features include at least one of ground line feature features and ground point feature features; a third feature extraction unit 960 is configured to extract features from the image based on the segmented point cloud to obtain facade feature features corresponding to facade features, wherein the facade features correspond to entity targets expressed in three-dimensional space; and a feature fusion unit 970 is configured to fuse the features corresponding to the same obtained features to generate a high-precision map based on the fused feature features.

[0083] Here, the operation of each of the above-mentioned units 910 to 970 of the high-precision map generation device 900 is similar to the operation of steps 210 to 270 described above, and will not be repeated here.

[0084] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0085] According to embodiments of this disclosure, an electronic device, a readable storage medium, and a computer program product are also provided.

[0086] refer to Figure 10 The present invention describes a structural block diagram of an electronic device 1000 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0087] like Figure 10 As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of the electronic device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0088] Multiple components in electronic device 1000 are connected to I / O interface 1005, including: input unit 1006, output unit 1007, storage unit 1008, and communication unit 1009. Input unit 1006 can be any type of device capable of inputting information to electronic device 1000. Input unit 1006 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device, and may include, but is not limited to, a mouse, keyboard, touchscreen, trackpad, trackball, joystick, microphone, and / or remote control. Output unit 1007 can be any type of device capable of presenting information, and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 1008 may include, but is not limited to, a hard disk and an optical disk. The communication unit 1009 allows the electronic device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and / or chipsets, such as Bluetooth™ devices, 802.11 devices, WiFi devices, WiMax devices, cellular communication devices and / or the like.

[0089] The computing unit 1001 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as method 200. For example, in some embodiments, method 200 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of method 200 described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to execute method 200 by any other suitable means (e.g., by means of firmware).

[0090] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0091] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0092] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0093] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0094] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0095] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0096] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0097] While embodiments or examples of this disclosure have been described with reference to the accompanying drawings, it should be understood that the methods, systems, and devices described above are merely exemplary embodiments or examples, and the scope of the invention is not limited by these embodiments or examples, but only by the granted claims and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. Furthermore, the steps may be performed in a different order than that described in this disclosure. Further, various elements in the embodiments or examples may be combined in various ways. Importantly, as the technology evolves, many elements described herein can be replaced by equivalents that appear after this disclosure.

Claims

1. A method for generating high-precision maps, comprising: Acquire point cloud and image data of the target scene, as well as pose data corresponding to the point cloud acquisition device and the image acquisition device; Based on the pose data of the point cloud acquisition device, semantic segmentation is performed on the point cloud to obtain a semantically segmented point cloud. Feature extraction is performed on the point cloud after semantic segmentation to obtain the first ground feature feature corresponding to the ground feature, wherein the ground feature includes at least one of ground line feature and ground point feature, the ground line feature corresponds to the entity target that is linearly expressed in a two-dimensional horizontal plane, and the ground point feature corresponds to the entity target that is planarly expressed in a two-dimensional horizontal plane. Perform semantic segmentation on the image to obtain a semantically segmented image; Based on the semantically segmented point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device, feature extraction is performed on the semantically segmented image to obtain the second ground feature corresponding to the ground feature, wherein the second ground feature includes at least one of the ground line feature and the ground point feature. Based on the segmented point cloud, feature extraction is performed on the image to obtain the facade element features corresponding to the facade elements, wherein the facade elements correspond to the entity targets expressed in three-dimensional space. as well as The features corresponding to the same elements obtained are fused to generate a high-precision map based on the fused feature features.

2. The method as described in claim 1, wherein, The point cloud includes point clouds obtained by multiple acquisitions of the target scene, and wherein feature extraction from the semantically segmented point cloud to obtain first ground feature features includes: Feature extraction is performed on the semantically segmented point cloud to obtain the features corresponding to the preset road markings; Based on the features corresponding to the preset road signs, the point clouds obtained from the multiple acquisitions are stitched together to obtain the stitched point cloud. A reflectance map is generated based on the stitched point cloud; and The first ground feature feature is obtained based on the reflectance value map.

3. The method as described in claim 2, wherein, Both the second ground feature and the first ground feature include the ground line feature, and wherein, Obtaining the first ground feature feature based on the reflectance map includes: inputting the ground line feature feature from the second ground feature feature and the reflectance map into a trained first network model to obtain the ground line feature feature from the first ground feature feature.

4. The method as described in claim 2 or 3, wherein, Both the second ground feature and the first ground feature include the ground point feature, and wherein, Obtaining the first ground feature feature based on the reflectance map includes: inputting the ground point feature feature in the second ground feature feature and the reflectance map into a trained second network model to obtain the ground point feature feature in the first ground feature feature.

5. The method of claim 4, wherein, The second ground feature includes the ground point feature, and wherein inputting the ground point feature and the reflectance map from the second ground feature into the trained second network model includes: Cluster the pixels corresponding to the ground point features determined based on the image to obtain cluster center points; and The cluster center points are expanded within a preset range, and the expanded range is used as the ground point feature to be input into the second network model.

6. The method of claim 1, wherein, Based on the semantically segmented point cloud and the corresponding pose data, feature extraction is performed on the semantically segmented image to obtain second ground feature features, including: The semantically segmented point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device, and the semantically segmented image are input into a trained third network model to obtain a bird's-eye view. Perform semantic segmentation on the bird's-eye view to obtain a semantically segmented bird's-eye view; and Feature extraction is performed on the semantically segmented bird's-eye view to obtain the second ground feature features.

7. The method of claim 1, wherein, Based on the segmented point cloud, feature extraction is performed on the image to obtain facade feature features, including: Based on the image, the facade elements are identified to obtain the detection bounding boxes of the facade elements; The detection frames of the facade elements are reconstructed in 3D to obtain the three-dimensional detection frames of the facade elements. The 3D detection frame of the facade elements is projected onto the corresponding point cloud to determine the overlapping area; and In response to determining that the overlapping area is greater than a first preset threshold, the stereo detection frame of the facade element obtained by 3D reconstruction is used as the obtained facade element feature.

8. The method of claim 7, wherein feature extraction of the image based on the segmented point cloud to obtain facade feature features includes: In response to determining that the overlapping area is not greater than the first preset threshold, geometric features are re-extracted from the stereo detection box of the facade element based on the segmented point cloud to obtain the corrected stereo detection box of the facade element, which is used as the obtained facade element feature.

9. The method of claim 8, wherein, The facade elements include rod-shaped structures, wherein the geometric feature re-extraction of the stereo detection box of the facade elements based on the segmented point cloud includes: At least a portion of the stereo detection frame is divided into multiple sub-stereo detection frames along a first direction; For each of the plurality of sub-stereo detection frames: The points in the point cloud corresponding to the sub-stereo detection frame are projected onto a plane perpendicular to the first direction to determine the radius of the smallest circumcircle surrounding the projection area. In response to determining that the radius is within a preset range, the point corresponding to the sub-stereoscopic detection frame is determined as a seed point; The seed points corresponding to the plurality of sub-stereo detection frames are merged to form a first seed point set, wherein the Euclidean distance between any two seed points in the first seed point set is less than a preset distance threshold; and Region growing is performed based on the first set of seed points to obtain the corrected 3D detection frame of the facade elements.

10. The method of claim 9, wherein, Region growth based on the first set of seed points includes: determining a first detection box surrounding at least one seed point in the first set of seed points, such that during the region growth process, the number of points within the first detection box and the number of points outside the first detection box are higher than a second preset threshold.

11. The method of claim 8, wherein, The facade elements include sign-shaped objects, wherein the geometric feature re-extraction of the stereo detection bounding box of the facade elements based on the segmented point cloud includes: Determine the dimensional features of the point cloud corresponding to the stereo detection box to determine the set of two-dimensional points; Cluster the points in the set of two-dimensional points to obtain one or more clustered point cloud clusters; The one or more point cloud clusters are filtered based on preset point cloud cluster size information to obtain filtered point cloud clusters; The points in the filtered point cloud cluster are subjected to plane fitting to determine the center position and normal direction of the fitted plane; and Based on the center position and normal direction, the three-dimensional detection frame of the corrected facade element is determined.

12. The method of claim 11, wherein, The point cloud cluster size information includes at least one of the height and width of the point cloud cluster.

13. The method of claim 8, wherein, The facade elements include traffic lights, wherein the geometric feature re-extraction of the stereo detection bounding box of the facade elements based on the segmented point cloud includes: The points in the segmented point cloud are clustered based on the first distance to determine the first cluster center point and the first normal direction; Clustering is performed on points of the same category and with the same first normal direction based on the second distance to determine the second cluster center point and the second normal direction, wherein the second distance is greater than the first distance, and wherein the category is determined based on semantic segmentation. A first detection box is determined based on the second cluster center point and the second normal direction; and Based on the first cluster center point and the first normal direction, one or more second detection boxes located within the first detection box are determined as the three-dimensional detection boxes of the corrected facade elements.

14. The method of claim 2, wherein, Both the first ground feature and the second ground feature include ground line feature features, and the feature fusion of all obtained feature features includes at least one of the following: A predetermined number of second ground line elements are determined to be closest to the first ground line element. The predetermined number of second ground line elements are then voted on based on a preset first attribute. Based on the voting results, the first attribute of the corresponding ground line element in the second ground line elements is assigned to the first ground element. A predetermined number of first ground line features closest to the second ground line feature are identified, and the second attribute of the second ground line feature is adjusted based on a preset second attribute corresponding to the predetermined number of first ground line features. Wherein, the first ground line feature is the ground line feature corresponding to the ground line feature feature in the first ground feature feature, and the second ground line feature is the ground line feature corresponding to the ground line feature feature in the second ground feature feature.

15. A high-precision map generation device, comprising: The acquisition unit is configured to acquire point clouds and images of the target scene, as well as pose data corresponding to the point cloud acquisition device and the image acquisition device. The first semantic segmentation unit is configured to perform semantic segmentation on the point cloud based on the pose data of the point cloud acquisition device, and obtain the semantically segmented point cloud. The first feature extraction unit is configured to extract features from the semantically segmented point cloud to obtain the first ground feature features corresponding to the ground features, wherein the ground features include at least one of ground line features and ground point features, the ground line features correspond to entity targets that are linearly expressed in a two-dimensional horizontal plane, and the ground point features correspond to entity targets that are planarly expressed in a two-dimensional horizontal plane. The second semantic segmentation unit is configured to perform semantic segmentation on the image to obtain a semantically segmented image; The second feature extraction unit is configured to extract features from the semantically segmented image based on the relative pose data between the point cloud after semantic segmentation, the point cloud acquisition device, and the image acquisition device, so as to obtain the second ground feature corresponding to the ground feature, wherein the second ground feature includes at least one of the ground line feature and the ground point feature. The third feature extraction unit is configured to extract features from the image based on the segmented point cloud to obtain facade element features corresponding to the facade elements, wherein the facade elements correspond to entity targets expressed in three-dimensional space; and The feature fusion unit is configured to fuse the features corresponding to the same elements obtained, so as to generate a high-precision map based on the fused feature features.

16. The apparatus of claim 15, wherein, The point cloud includes point clouds obtained by repeatedly capturing data from the target scene, and the device further includes: The first feature extraction subunit is configured to extract features from the semantically segmented point cloud to obtain features corresponding to preset road markings. The stitching subunit is configured to stitch together the point cloud corresponding to the point cloud obtained from multiple acquisitions based on the features corresponding to the preset road signs, so as to obtain the stitched point cloud. The map generation subunit is configured to generate a reflectance map based on the stitched point cloud; and The second feature extraction subunit is configured to obtain the first ground feature features based on the reflectance value map.

17. The apparatus of claim 15, wherein, The second feature extraction unit includes: The image processing subunit is configured to input the semantically segmented point cloud, the relative pose data between the point cloud acquisition device and the image acquisition device, and the semantically segmented image into a trained third network model to obtain a bird's-eye view. A semantic segmentation subunit is configured to perform semantic segmentation on the bird's-eye view to obtain a semantically segmented bird's-eye view; and The third feature extraction subunit is configured to extract features from the semantically segmented bird's-eye view to obtain the second ground feature features.

18. The apparatus of claim 15, wherein, The third feature extraction unit includes: The fourth feature extraction subunit is configured to identify the facade elements based on the image to obtain the detection box of the facade elements; The 3D reconstruction subunit is configured to perform 3D reconstruction on the detection frame of the facade element to obtain the three-dimensional detection frame of the facade element. The projection subunit is configured to project the stereo detection frame of the facade element onto the corresponding point cloud to determine the overlapping area; and The first determining subunit is configured to, in response to determining that the overlapping area is greater than a first preset threshold, use the stereo detection frame of the facade element obtained by 3D reconstruction as the obtained facade element feature.

19. The apparatus of claim 18, wherein, The third feature extraction unit includes: The second determining subunit is configured to, in response to determining that the overlapping area is not greater than the first preset threshold, perform geometric feature re-extraction on the stereo detection box of the facade element based on the segmented point cloud to obtain the corrected stereo detection box of the facade element as the obtained facade element feature.

20. The apparatus of claim 16, wherein, Both the first ground feature and the second ground feature include ground line feature features, and the feature fusion unit performs at least one of the following operations: A predetermined number of second ground line elements are determined to be closest to the first ground line element. The predetermined number of second ground line elements are then voted on based on a preset first attribute. Based on the voting results, the first attribute of the corresponding ground line element in the second ground line elements is assigned to the first ground element. A predetermined number of first ground line features closest to the second ground line feature are identified, and the second attribute of the second ground line feature is adjusted based on a preset second attribute corresponding to the predetermined number of first ground line features. Wherein, the first ground line feature is the ground line feature corresponding to the ground line feature feature in the first ground feature feature, and the second ground line feature is the ground line feature corresponding to the ground line feature feature in the second ground feature feature.

21. An electronic device, comprising: At least one processor; as well as A memory that is communicatively connected to the at least one processor; in The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-14.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-14.

23. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method of any one of claims 1-14.

Images