Semantic map construction method and device, equipment and medium
By acquiring standardized attributes (location, orientation, and size) of semantic elements for semantic association and optimization, the problem of errors introduced by the lack of observation contour points in semantic maps is solved, and higher-precision semantic map construction is achieved.
Patent Information
- Application Number
- CN202511266268.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-05
- Publication Date
- 2025-11-18
AI Technical Summary
In existing technologies, due to occlusion and other reasons, semantic elements such as pillars and storage locations may only have two or three outline points observed, resulting in insufficient semantic map accuracy. Existing methods introduce errors when supplementing missing outline points.
By acquiring semantic observation data of semantic elements at each observation time, determining their standardized attributes (location, orientation, and size), and performing semantic association and optimization based on these attributes, semantic element clusters are formed, and finally a semantic map is generated, avoiding errors introduced by contour points lacking observation.
This improves the accuracy of semantic maps, avoids errors caused by a lack of observed contour points, and enhances the accuracy of semantic mapping.
Smart Images
Figure CN120970618A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure relates to the technical field of semantic map, and particularly relates to a semantic map construction method and device, equipment and medium. BACKGROUND
[0002] The semantic map is constructed by using frame-by-frame semantic observation data provided by a perception module, and a global map composed of semantic elements is constructed, so that an absolute pose observation that corrects cumulative errors is provided for positioning, which is a high-reliability way to realize indoor high-precision positioning.
[0003] In related technologies, a laser perception scheme or a visual perception scheme is used to obtain contour points of semantic elements to construct a semantic map. For an indoor scene, due to occlusion and other reasons, semantic elements such as pillars and storage locations may only be observed to have two or three contour points. The missing contour points are filled in before the perception module outputs semantic observation data. However, these actually observed contour points introduce errors for constructing a semantic map, and the semantic map accuracy needs to be improved. SUMMARY
[0004] To solve the above technical problems, the present disclosure provides a semantic map construction method, device, equipment and medium.
[0005] In a first aspect, the present disclosure provides a semantic map construction method, comprising:
[0006] Obtaining semantic observation data of a semantic element at each observation time, and determining a standardized attribute of the semantic element according to the semantic observation data; the standardized attribute includes a position, an orientation and a size;
[0007] According to the position, the semantic elements at multiple observation times are semantically associated to obtain at least one semantic element cluster;
[0008] According to the position, the orientation and the size of the semantic elements in each semantic element cluster, semantic optimization is performed to generate a semantic map according to the optimized semantic element cluster.
[0009] In a second aspect, the present disclosure provides a semantic map construction device, comprising:
[0010] An obtaining module is configured to obtain semantic observation data of a semantic element at each observation time, and determine a standardized attribute of the semantic element according to the semantic observation data; the standardized attribute includes a position, an orientation and a size;
[0011] An associating module is configured to associate the semantic elements at multiple observation times according to the position to obtain at least one semantic element cluster;
[0012] The generating module is configured to perform semantic optimization according to the position, orientation and size of the semantic elements in each semantic element cluster, and generate a semantic map according to the optimized semantic element cluster.
[0013] In a third aspect, an electronic device is provided, including: a processor; a memory for storing executable instructions of the processor; and the processor is configured to read the executable instructions from the memory and execute the instructions to implement the semantic map construction method of the first aspect.
[0014] In a fourth aspect, a computer readable storage medium is provided, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the semantic map construction method of the first aspect.
[0015] The technical solution provided by the embodiments of the present disclosure has the following advantages compared with the prior art: by obtaining semantic observation data of semantic elements at each observation time, determining standardized attributes of the semantic elements according to the semantic observation data, using position, orientation and size as the standardized attributes to represent semantic information, then performing semantic association on the semantic elements at multiple observation times according to the position to obtain at least one semantic element cluster, and performing semantic optimization according to the position, orientation and size of the semantic elements in each semantic element cluster, the semantic map construction is realized, and in the process of semantic association and semantic optimization, the error introduced by the contour points lacking observation is avoided, and the semantic mapping accuracy is improved. BRIEF DESCRIPTION OF DRAWINGS
[0016] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced here. Obviously, for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.
[0018] Figure 1 A flowchart of a semantic map construction method provided by the embodiments of the present disclosure is shown.
[0019] Figure 2 A flowchart of another semantic map construction method provided by the embodiments of the present disclosure is shown.
[0020] Figure 3 A flowchart of a semantic mapping is shown.
[0021] Figure 4A scene schematic diagram of constructing a residual term based on standardized attributes of a storage location and a column is shown.
[0022] Figure 5 A structural schematic diagram of a semantic map construction device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0023] In order to more clearly understand the above-mentioned purposes, features and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.
[0024] In the following description, many specific details are set forth in order to provide a thorough understanding of the present disclosure, but the present disclosure can also be implemented in other ways different from those described herein; obviously, the embodiments in the description are only some of the embodiments of the present disclosure, not all the embodiments.
[0025] Figure 1 A flowchart of a semantic map construction method provided by an embodiment of the present disclosure is shown, and the method provided by the embodiment of the present disclosure can be executed by a semantic map construction device, which can be implemented by software and / or hardware, and can be integrated on any electronic device with computing capability.
[0026] As shown in Figure 1 The semantic map construction method provided by the embodiment of the present disclosure can include:
[0027] In step 101, semantic observation data of a semantic element at each observation time is acquired, and standardized attributes of the semantic element are determined according to the semantic observation data.
[0028] The method of the embodiment of the present disclosure is applied to construct a semantic map, for example, a semantic map construction for an indoor parking scene.
[0029] In the embodiment, data of a scene to be constructed is collected at each observation time to acquire semantic observation data of a semantic element. Optionally, a laser perception sensor and a visual perception sensor are arranged on a vehicle, and the semantic observation data of the semantic element at each observation time is acquired by moving the vehicle in the scene and collecting data, and then the semantic observation data of the semantic element is processed to obtain standardized attributes of the semantic element.
[0030] The semantic element includes but is not limited to a storage location, a column, a wall line and the like, the semantic observation data is for example an observation profile point obtained by observation, and the standardized attributes include a position, an orientation and a size.
[0031] As an example, for each observation time, semantic observation data at the observation time can be obtained in the vehicle body coordinate system. Since the conversion relationship of the vehicle body coordinate system relative to the global coordinate system changes with the movement of the vehicle, the semantic observation data is converted from the vehicle body coordinate system to the global coordinate system according to the vehicle pose at each observation time, so that it is consistent with the coordinate system of each semantic observation data at the historical observation time. The vehicle pose at the current observation time can be obtained by interpolation according to the timestamp at the current observation time. In this example, the coordinate conversion is as follows:
[0032]
[0033] wherein, is the observation contour point in the vehicle body coordinate system, k is the kth observation time, and the pose R k and t k is the vehicle pose at the kth observation time, and the vehicle pose is used for conversion between the vehicle body coordinate system and the global coordinate system, is the observation contour point in the global coordinate system.
[0034] In step 102, the semantic elements at multiple observation times are semantically associated according to the position to obtain at least one semantic element cluster.
[0035] In this embodiment, the semantic elements of the same observation target observed at different observation times are associated to the same cluster by semantically associating the semantic elements at multiple observation times, to obtain at least one semantic element cluster. The semantic element cluster includes semantic elements belonging to the same observation target at different observation times.
[0036] In one embodiment of the present disclosure, the semantic elements at multiple observation times are semantically associated according to the position to obtain at least one semantic element cluster, including: for any target semantic element at the current observation time, calculating the distance between the position of the target semantic element and the fitting position of the existing semantic element cluster; determining a first semantic element cluster with a distance less than or equal to a preset distance threshold from the existing semantic element cluster, and adding the target semantic element to the first semantic element cluster; if there is no first semantic element cluster in the existing semantic element cluster, a second semantic element cluster is newly built according to the target semantic element. The existing semantic element cluster can be a semantic element cluster newly built at the historical observation time. If there is no existing semantic element cluster at the current observation time, a semantic element cluster is directly newly built according to the target semantic element. The fitting position of the existing semantic element cluster can be determined according to the positions of the semantic elements in the existing semantic element cluster.
[0037] As an example, for the first observation time, the first storage location can be observed, and the semantic observation data and standardized attributes of the semantic element a1 are obtained. At this time, a semantic element cluster A is newly created according to the semantic element a1. For the second observation time, the first storage location and the second storage location can be observed, and the semantic observation data and standardized attributes of the semantic elements a2 and b2 are obtained. At this time, the semantic element a2 is added to the semantic element cluster A according to the position of the semantic element a2, and a semantic element cluster B is newly created according to the position of the semantic element b2, and the semantic element b2 is added to the semantic element cluster B. In this example, the preset distance threshold is, for example, 2 meters. In this way, the semantic elements are associated according to the positions in the standardized attributes, which can avoid errors introduced by contour point completion and improve the semantic mapping accuracy.
[0038] In step 103, the semantic optimization is performed according to the positions, orientations and sizes of the semantic elements in each semantic element cluster, so as to generate a semantic map according to the optimized semantic element clusters.
[0039] In this embodiment, after multiple observation times, each semantic element cluster usually includes multiple semantic elements. For each semantic element cluster, the optimization processing is performed according to the standardized attributes of the semantic elements in the semantic element cluster, so as to obtain the optimized standardized attributes. Then, the semantic elements represented by the semantic element cluster are added to the semantic map according to the optimized standardized attributes. In this way, the semantic elements are associated according to the standardized attributes, which can avoid errors introduced by contour point completion and improve the semantic mapping accuracy.
[0040] The semantic optimization is exemplified as follows.
[0041] As a possible implementation, for each semantic element cluster, the position residual term, the orientation residual term and the size residual term are determined according to the positions, orientations and sizes of the semantic elements in the semantic element cluster. Then, the semantic optimization processing is performed according to the position residual term, the orientation residual term and the size residual term, so as to obtain the target position, the target orientation and the target size corresponding to the semantic element cluster. The semantic element cluster is added to the semantic map based on the target position, the target orientation and the target size. The trigger condition of the semantic optimization can be set as needed. The trigger condition indicates whether the semantic elements in the sliding time window satisfy the optimization starting standard. For example, the trigger condition is set as that the number of storage locations observed more than N times in the sliding time window is greater than M. N is, for example, 10, and M is, for example, 20. When the semantic optimization is triggered, the residual terms between the vehicle poses and the semantic observations at the observation times in the sliding time window are constructed, and the optimal solution is obtained by using a solving method, such as the ceres method. After the joint optimization is completed, the semantic elements in the sliding time window are removed and added to the semantic map.
[0042] As an example, for the storage location and the column, a residual term is constructed based on standardized attributes of the storage location and the column, including a position distance, an orientation difference, and a size difference, compared to a residual term construction manner through the Euclidean distance of four contour points of the storage location and the column, the residual term construction manner can avoid errors introduced by the contour points lacking actual observation, and improve semantic mapping accuracy.
[0043] In the current semantic map construction method, there is a situation of missing semantic observation data, for example, a column can observe at most three observation contour points, and a storage location may observe two or three observation contour points due to the blocking of the vehicle body, at this time, the missing observation contour points are filled in according to the preset rules or prior information, and the existing method performs semantic mapping according to the filled observation contour points, and the above observation contour points lacking actual observation will introduce errors.
[0044] According to the technical scheme of the embodiments of the present disclosure, by obtaining semantic observation data of a semantic element at each observation time, determining standardized attributes of the semantic element according to the semantic observation data, using position, orientation and size as standardized attributes to represent semantic information, and then performing semantic association on the semantic elements at multiple observation times according to the position, the semantic element cluster is obtained, and the semantic optimization is performed according to the position, the orientation and the size of the semantic element in each semantic element cluster, the semantic map construction is realized, and the errors introduced by the contour points lacking observation are avoided in the semantic association and semantic optimization process, and the semantic mapping accuracy is improved.
[0045] Based on the above embodiments, Figure 2 The flowchart of another semantic map construction method provided by the embodiments of the present disclosure is shown in FIG. 2. Figure 2 As shown in the figure, in the method, the standardized attributes of the semantic element are determined according to the semantic observation data, which includes:
[0046] Step 201, for each semantic element, determining a target contour point closest to the vehicle from multiple observation contour points of the semantic element.
[0047] The semantic observation data includes multiple observation contour points.
[0048] As an example, taking a storage location as an example, the geometric shape of the storage location is a rectangle, and the contour points are the vertices of the rectangle. For the four contour points of the storage location, due to the existence of other vehicle body blocking in the storage location, when the semantic observation data is obtained by the ego vehicle, the semantic observation data of the storage location is three observation contour points, and the target contour point closest to the ego vehicle is determined from the three observation contour points.
[0049] The target contour point is determined as follows:
[0050]
[0051] The target contour point is determined as follows: is a target contour point, is a representation of the observation contour point in the global coordinate system, t k is a parameter in the vehicle pose, used to represent the ego position.
[0052] In step 202, the observation edge and the depth edge are determined from the two adjacent edges of the target contour point according to the included angle between the two adjacent edges and the vehicle direction.
[0053] In this embodiment, the two adjacent edges of the target contour point can be determined according to the target contour point and other observation contour points in the plurality of observation contour points. By comparing the included angles between the two adjacent edges of the target contour point and the vehicle direction, the adjacent edge with smaller absolute value of the included angle is selected as the observation edge, and the adjacent edge with larger absolute value of the included angle is selected as the depth edge. The vehicle direction can adopt the vehicle head direction. Thus, since there may be a contour point lacking actual observation in the observation contour points, the observation edge with better observation effect is determined from the adjacent edges formed by the observation contour points, and the standardized attribute is determined based on the observation edge and the depth edge, which can improve the accuracy of the subsequent standardized attribute and further improve the precision of semantic mapping.
[0054] It should be noted that the above determination of the observation edge and the depth edge is an implementation manner, and the longer edge can also be selected as the observation edge and the shorter edge as the depth edge from the two adjacent edges, which is not limited here.
[0055] In step 203, the standardized attribute of the semantic element is determined according to the normal vector of the observation edge in the specified plane, the projection of the depth edge on the normal vector, and the length of the observation edge.
[0056] In this embodiment, the standardized attribute includes position, orientation and size. The position can be represented by a center point, and the size includes length and width. By using the above standardized attributes to represent semantic information, errors introduced by contour points lacking observation can be effectively avoided, and the above standardized attributes have global consistency, so that it is not necessary to unify the order of all observation contour points, and the contour point ordering originally required to be frequently called can be omitted, thereby improving the running efficiency of the semantic mapping project.
[0057] The specific steps of the three standardized attributes will be described below.
[0058] In an embodiment of the present disclosure, the position of the semantic element is determined by the following steps: in the case of a warehouse position, the center point of the semantic element is determined according to the normal vector of the observation edge in the specified plane and the projection of the depth edge on the normal vector; in the case of a column, the center point of the semantic element is determined according to the normal vector of the observation edge in the specified plane and the length of the observation edge. The position of the semantic element is represented by the center point.
[0059] As an example, for a storage location, the center point is determined based on the normal vector of the observed edge in the xy-plane of the global coordinate system and the projection of the depth edge onto the normal vector. For a pillar, the center point is determined based on the normal vector of the observed edge in the xy-plane of the global coordinate system and the length of the observed edge. In this example, the formula is as follows:
[0060]
[0061] Where ν is the normal vector, E best For the observation edge, E deep Let C be the depth edge and C be the center point.
[0062] In one embodiment of this disclosure, the orientation of a semantic element is determined by the following steps: rotating the heading angle corresponding to the normal vector of the observed edge in a specified plane multiple times in the specified plane by a specified angle to obtain multiple candidate angles; determining the target angle with the smallest cosine value from the multiple candidate angles; and determining the orientation of the semantic element based on the target angle.
[0063] As an example, the specified plane is the xy-plane in the global coordinate system, the specified angle is 90°, and the number of rotations is four. Specifically, the heading angle θ of the normal vector ν is rotated 90° within the xy-plane, repeated four times, and the angle with the smallest cosine value is selected as the orientation of the semantic element. In this example, the formula is as follows:
[0064]
[0065] Where D represents the orientation.
[0066] In one embodiment of this disclosure, the length and width of a semantic element are determined by the following steps: when the semantic element is a storage location, the length and width of the semantic element are determined based on the projection of the depth edge onto the normal vector and the length of the observed edge; when the semantic element is a pillar, the length and width of the semantic element are determined based on the length of the observed edge.
[0067] For storage locations, the length and width are determined by the length of the observed edge and the length of the projection of the depth edge onto the normal vector, respectively. For pillars, the length and width of the semantic observation are equal by default, so the length of the observed edge is used as the length and width of the pillar.
[0068] In this embodiment of the disclosure, the use of the above-mentioned standardized attributes to represent semantic information can effectively avoid errors introduced by contour points lacking observation. Furthermore, the above-mentioned standardized attributes have global consistency, which can omit the contour point sorting that originally needed to be called frequently, thereby improving the running efficiency of the semantic mapping project.
[0069] Reference Figure 3 , Figure 3A schematic diagram of a semantic mapping process is shown. Preprocessing involves transforming the semantic observation contour points from the vehicle body coordinate system to the global coordinate system. Semantic association involves calculating the distance and similarity between the semantic observation and the observation data of each cluster in the semantic association container, thus associating the semantic observation with other observation data of the same semantic element. Semantic optimization utilizes the semantic observation data of each cluster for joint optimization to obtain the contour point coordinates of the corresponding semantic elements, and further optimizes the pose at the corresponding observation time. For pillars and storage locations, standardized attributes are used to represent semantic information. By standardizing the contour points of storage locations and pillars, errors introduced by contour points lacking actual observation can be avoided during semantic association and semantic optimization, thereby improving the overall accuracy of semantic mapping.
[0070] In one embodiment of this disclosure, before performing semantic optimization based on the position, orientation, and size of semantic elements in each semantic element cluster, the method further includes: determining the similarity between each semantic element and the semantic element cluster based on the difference between the standardized attribute of each semantic element in the semantic element cluster and the mean of the fitted attribute of the semantic element cluster, and the weight of the standardized attribute; identifying abnormal semantic elements whose similarity does not meet the preset conditions from all semantic elements in the semantic element cluster, and deleting the abnormal semantic elements from the semantic element cluster.
[0071] As an example, in the semantic association step, the distance between the center point of a semantic element and the fitting center of each cluster's observation data is calculated to determine whether they are observations of the same semantic element. If no new semantic elements are added to a semantic element cluster after multiple observation times, the weights s of each standardized attribute are set accordingly. C s D s L s W Based on the center point, orientation, and width and height of the semantic elements, as well as the fitted mean values for the center point, orientation, and width and height of each semantic element in the semantic element cluster, the similarity between each semantic element in the semantic element cluster and the fitted mean value is calculated to screen out abnormal observations and avoid introducing them into the subsequent semantic optimization stage, as shown in the following formula:
[0072]
[0073] in, Fit the mean to the center point. To fit the mean, The length is fitted to the mean. The width is the fitted mean, and j represents the j-th semantic element.
[0074] Figure 4The diagram illustrates a scenario where residual terms are constructed based on standardized attributes of storage locations and pillars. Here, pose0 and pose1 correspond to two observation times, C, D, L, and W represent the center point, orientation, length, and width, respectively; rectangles represent different storage locations, and squares represent different pillars.
[0075] Figure 5 This is a schematic diagram of the structure of a semantic map construction apparatus provided in an embodiment of the present disclosure, as shown below. Figure 5 As shown, the semantic map construction device includes: an acquisition module 51, an association module 52, and a generation module 53.
[0076] The acquisition module 51 is used to acquire semantic observation data of semantic elements at each observation time and determine the standardized attributes of semantic elements based on the semantic observation data; the standardized attributes include position, orientation and size.
[0077] The association module 52 is used to perform semantic association on semantic elements at multiple observation times based on location, so as to obtain at least one semantic element cluster;
[0078] The generation module 53 is used to perform semantic optimization based on the position, orientation and size of the semantic elements in each semantic element cluster, so as to generate a semantic map based on the optimized semantic element clusters.
[0079] In one embodiment of this disclosure, the semantic observation data includes multiple observation contour points, and the acquisition module 51 is specifically used for:
[0080] For each semantic element, determine the target contour point closest to the vehicle from multiple observed contour points of the semantic element;
[0081] Based on the angle between the two adjacent sides of the target contour point and the vehicle direction, determine the observation edge and the depth edge from the two adjacent sides;
[0082] Based on the normal vector of the observed edge in the specified plane, the projection of the depth edge onto the normal vector, and the length of the observed edge, the standardized attributes of the semantic elements are determined.
[0083] In one embodiment of this disclosure, the location of a semantic element is determined through the following steps:
[0084] When the semantic element is a location, the center point of the semantic element is determined based on the normal vector of the observed edge in the specified plane and the projection of the depth edge onto the normal vector; whereby the center point represents the position of the semantic element.
[0085] When the semantic element is a pillar, the center point of the semantic element is determined based on the normal vector of the observed edge in the specified plane and the length of the observed edge.
[0086] In one embodiment of this disclosure, the orientation of a semantic element is determined through the following steps:
[0087] The heading angle corresponding to the normal vector of the observed edge in the specified plane is rotated multiple times in the specified plane according to a specified angle to obtain multiple candidate angles;
[0088] The target angle with the smallest cosine value is determined from multiple candidate angles, and the orientation of the semantic element is determined based on the target angle.
[0089] In one embodiment of this disclosure, the association module 52 is specifically used for:
[0090] For any target semantic element at the current observation time, calculate the distance between the position of the target semantic element and the fitted position of the existing semantic element cluster;
[0091] Determine a first semantic element cluster whose distance is less than or equal to a preset distance threshold from the existing semantic element clusters, and add the target semantic element to the first semantic element cluster;
[0092] If the first semantic element cluster does not exist in the existing semantic element cluster, then a second semantic element cluster is created based on the target semantic element.
[0093] In one embodiment of this disclosure, the device further includes:
[0094] The filtering module is used to determine the similarity between each semantic element and the semantic element cluster based on the difference between the standardized attribute of each semantic element in the semantic element cluster and the mean of the fitted attribute of the semantic element cluster, and the weight of the standardized attribute.
[0095] Identify anomalous semantic elements from all semantic elements in the semantic element cluster whose similarity does not meet the preset conditions, and delete the anomalous semantic elements from the semantic element cluster.
[0096] In one embodiment of this disclosure, the generation module 53 is specifically used for:
[0097] For each semantic feature cluster, determine the position residual, orientation residual, and size residual based on the position, orientation, and size of the semantic features in the semantic feature cluster;
[0098] Semantic optimization is performed based on the location residual, orientation residual, and size residual to obtain the target location, target orientation, and target size corresponding to the semantic feature cluster.
[0099] Semantic element clusters are added to the semantic map based on target location, target orientation, and target size.
[0100] The semantic map construction apparatus provided in this disclosure can execute any semantic map construction method provided in this disclosure, and has the corresponding functional modules and beneficial effects of the method execution. Content not described in detail in the apparatus embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.
[0101] This disclosure also provides an electronic device including one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the methods of the embodiments of this disclosure above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.
[0102] In one example, the electronic device may also include input and output devices, which are interconnected via a bus system and / or other forms of connection. Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc. In addition, depending on the specific application, the electronic device may include any other suitable components such as a bus, input / output interfaces, etc.
[0103] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.
[0104] Computer program products can be written in any combination of one or more programming languages to perform the operations of embodiments of this disclosure. The programming languages include object-oriented programming languages such as Java and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0105] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.
[0106] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0108] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A semantic map construction method, characterized in that, The method includes: Obtain semantic observation data of semantic elements at each observation time, and determine the standardized attributes of the semantic elements based on the semantic observation data; the standardized attributes include position, orientation, and size. Based on the location, semantic elements at multiple observation times are semantically correlated to obtain at least one semantic element cluster; Semantic optimization is performed based on the position, orientation, and size of the semantic elements in each semantic element cluster to generate a semantic map based on the optimized semantic element clusters.
2. The method as described in claim 1, characterized in that, The semantic observation data includes multiple observation contour points, and determining the standardized attributes of the semantic elements based on the semantic observation data includes: For each of the semantic elements, determine the target contour point closest to the vehicle from the plurality of observed contour points of the semantic element; Based on the angle between the two adjacent sides of the target contour point and the vehicle direction, determine the observation edge and the depth edge from the two adjacent sides; The standardized attributes of the semantic element are determined based on the normal vector of the observed edge in the specified plane, the projection of the depth edge onto the normal vector, and the length of the observed edge.
3. The method as described in claim 2, characterized in that, The location of the semantic element is determined by the following steps: When the semantic element is a location, the center point of the semantic element is determined based on the normal vector of the observed edge in the specified plane and the projection of the depth edge onto the normal vector; wherein, the center point represents the position of the semantic element; When the semantic element is a column, the center point of the semantic element is determined based on the normal vector of the observed edge in the specified plane and the length of the observed edge.
4. The method as described in claim 2, characterized in that, The orientation of the semantic element is determined by the following steps: The heading angle corresponding to the normal vector of the observed edge in the specified plane is rotated multiple times in the specified plane according to a specified angle to obtain multiple candidate angles; The target angle with the smallest cosine value is determined from the plurality of candidate angles, and the orientation of the semantic element is determined based on the target angle.
5. The method as described in claim 1, characterized in that, The step of semantically associating semantic elements from multiple observation times based on the location to obtain at least one semantic element cluster includes: For any target semantic element at the current observation time, calculate the distance between the position of the target semantic element and the fitted position of the existing semantic element cluster; From the existing semantic element clusters, determine the first semantic element cluster whose distance is less than or equal to a preset distance threshold, and add the target semantic element to the first semantic element cluster; If the first semantic element cluster does not exist in the existing semantic element cluster, then a second semantic element cluster is created based on the target semantic element.
6. The method as described in claim 1, characterized in that, Before performing semantic optimization based on the position, orientation, and size of the semantic elements in each semantic element cluster, the method further includes: The similarity between each semantic element and the semantic element cluster is determined based on the difference between the standardized attribute of each semantic element in the semantic element cluster and the mean of the fitted attribute of the semantic element cluster, and the weight of the standardized attribute. Identify abnormal semantic elements from all semantic elements in the semantic element cluster whose similarity does not meet the preset conditions, and delete the abnormal semantic elements from the semantic element cluster.
7. The method as described in claim 1, characterized in that, The step of semantic optimization based on the position, orientation, and size of semantic elements in each semantic element cluster to generate a semantic map based on the optimized semantic element clusters includes: For each semantic element cluster, based on the position, orientation, and size of the semantic elements in the semantic element cluster, determine the position residual, orientation residual, and size residual. Semantic optimization is performed based on the position residual, orientation residual, and size residual to obtain the target position, target orientation, and target size corresponding to the semantic element cluster. The semantic element clusters are added to the semantic map based on the target location, target orientation, and target size.
8. A semantic map construction device, characterized in that, include: The acquisition module is used to acquire semantic observation data of semantic elements at each observation time and determine the standardized attributes of the semantic elements based on the semantic observation data. The standardized attributes include location, orientation, and size; The association module is used to perform semantic association on semantic elements at multiple observation times based on the location, so as to obtain at least one semantic element cluster; The generation module is used to perform semantic optimization based on the position, orientation, and size of the semantic elements in each semantic element cluster, so as to generate a semantic map based on the optimized semantic element clusters.
9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the semantic map construction method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the semantic map construction method according to any one of claims 1-7.