Memory parking initialization method and device, electronic equipment and storage medium

Abstract

The application discloses a memory parking initialization method and device, electronic equipment and a storage medium. The method obtains an initialization reference key frame through similarity calculation and a preset threshold value, obtains the pose of a current frame relative to a map according to the pose of the current frame relative to the initialization reference key frame and the pose of the initialization reference key frame in the map, and completes the pose initialization of the current frame, so that the pose of the current frame in the map can be quickly initialized even in the case that there is no positioning signal and the vehicle does not start at the starting point of the map.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically, to a memory parking initialization method, apparatus, electronic device, and storage medium. Background Technology

[0002] Currently, major automakers are fiercely competing in the autonomous driving field, indicating that intelligent driving is becoming increasingly important in people's lives and is also a popular research topic for researchers. One of the more important functions of intelligent driving is Home-zone Parking Assist (HPA). HPA refers to the ability of an autonomous driving system to automatically complete all instructions for parking and exiting a specific parking space in a specific area without human intervention.

[0003] In the current HPA algorithm, for a specific local map, the user needs to drive near the starting point of the target map or start in an area with GPS signal. If it is necessary to locate any point on the target map trajectory without GPS signal, a map index needs to be built or the entire map needs to be traversed before positioning initialization can be performed, which requires a lot of computing resources.

[0004] In existing technologies, there are methods that use navigation information to prompt users to drive to the vicinity of the starting point of the target map, thereby quickly performing positioning initialization. However, this method requires the user to drive the vehicle to the starting area of ​​the target map. It cannot enter the HPA tracking stage from any point in the HPA memory trajectory, nor can it perform positioning initialization during movement.

[0005] Existing technologies also include methods for initializing positioning based on a global coordinate system, but these methods require users to drive to an area with GPS signals to obtain global information beforehand, and cannot be directly initialized in underground parking lots.

[0006] In view of this, the present invention provides a memory parking initialization method, apparatus, electronic device and storage medium to achieve rapid positioning initialization from any point near the HPA target map mapping trajectory. Summary of the Invention

[0007] In view of this, the present invention provides a memory parking initialization method, apparatus, electronic device and storage medium to achieve rapid positioning initialization from any point near the HPA target map mapping trajectory.

[0008] On one hand, the present invention provides a memory parking initialization method, comprising:

[0009] S101: Obtain a map, which includes at least two keyframes;

[0010] S102: After distorting the laser point cloud corresponding to the key frame, extract feature points to obtain the feature description set of the key frame;

[0011] S103: Collect the laser point cloud corresponding to the current frame, and extract feature points after distorting the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame;

[0012] S104: Traverse the feature description sets of all the keyframes, match them with the feature description set of the current frame to obtain the corresponding matching point pair set and record them, and calculate the similarity of all the matching point pair sets to obtain the corresponding similarity value.

[0013] S105: Determine whether the largest similarity value is greater than or equal to a preset threshold;

[0014] If the largest similarity value is greater than or equal to the preset threshold, the keyframe corresponding to the largest similarity value is used as the initialization reference keyframe; the pose of the current frame relative to the map is obtained based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, and the pose initialization of the current frame is completed.

[0015] Optionally, determining whether the largest similarity value is greater than or equal to the preset threshold further includes:

[0016] If the largest similarity value is less than the preset threshold, the next frame is updated to the current frame, and steps S103 to S105 are repeated.

[0017] Optionally, the distortion correction process for the laser point cloud corresponding to the key frame includes: selecting the coordinate system where the laser point cloud corresponding to any key frame is located as the reference coordinate system; and transforming the laser point clouds corresponding to all key frames to the reference coordinate system.

[0018] The distortion correction process for the laser point cloud corresponding to the current frame includes: transforming the laser point cloud corresponding to the current frame to the reference coordinate system.

[0019] Optionally, the pose of the current frame relative to the initialization reference keyframe is obtained in the following manner:

[0020] Obtain the set of matching point pairs corresponding to the initialization reference keyframe;

[0021] The pose of the current frame relative to the initialization reference keyframe is calculated based on the set of matching points corresponding to the initialization reference keyframe.

[0022] Optionally, the feature description sets of all the keyframes are traversed, and each is matched with the feature description set of the current frame to obtain the corresponding set of matching points, including:

[0023] The feature description set of the key frame includes key frame feature points, and the key frame feature points include the corresponding key frame high-dimensional feature description vector set.

[0024] The feature description set of the current frame includes feature points of the current frame, and the feature points of the current frame include the corresponding high-dimensional feature description vector set of the current frame.

[0025] Calculate the maximum bipartite graph matching between all the high-dimensional feature description vector sets of the key frames and all the high-dimensional feature description vector sets of the current frame in the feature description set of the key frames;

[0026] The key frame feature points corresponding to the key frame high-dimensional feature description vector set and the current frame feature points corresponding to the current frame high-dimensional feature description vector set that match the key frame high-dimensional feature description vector set constitute a matching point pair, and all the matching point pairs constitute the matching point pair set.

[0027] Optionally, calculating the pose of the current frame relative to the initialization reference keyframe based on the set of matching point pairs corresponding to the initialization reference keyframe includes:

[0028] Construct an overdetermined equation based on the set of matching point pairs corresponding to the initialization reference keyframe;

[0029] The overdetermined equations are solved using the least squares method to obtain the pose of the current frame relative to the initialization reference keyframe.

[0030] Optionally, the overdetermined equations are calculated in the following manner:

[0031] Q1 = T × P1

[0032] …

[0033] Q i =T×P i

[0034] …

[0035] Q n =T×P n

[0036] Where T is the transformation matrix, (Q i P i ) represents the i-th pair of matching points in the set of matching point pairs corresponding to the initialization reference keyframe, P i For the i-th feature point in the current frame, Q iThe i-th keyframe feature point in the feature description set of the initialization reference keyframe, 1≤i≤n, n≥4 and n is an integer.

[0037] On the other hand, the present invention also provides a memory parking initialization device, comprising:

[0038] The map acquisition module is used to acquire a map, which includes at least two keyframes;

[0039] The data processing module is used to process the laser point cloud corresponding to the key frame to obtain the feature description set of the key frame and to process the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame.

[0040] The similarity calculation module is used to traverse the feature description sets of all the key frames, match them with the feature description set of the current frame to obtain the corresponding matching point pair set and record them, and perform similarity calculation on all the matching point pair sets to obtain the corresponding similarity value.

[0041] An initialization module is used to take the keyframe corresponding to the largest similarity value greater than or equal to a preset threshold as an initialization reference keyframe, obtain the pose of the current frame relative to the map based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe in the map, and complete the pose initialization of the current frame.

[0042] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the memory parking initialization method described in any of the preceding claims.

[0043] In another aspect, the present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the memory parking initialization method as described in any of the preceding claims.

[0044] Compared with the prior art, the memory parking initialization method, device, electronic device and storage medium provided by the present invention achieve at least the following beneficial effects:

[0045] The memory parking initialization method, device, electronic device, and storage medium provided by the present invention obtain an initialization reference keyframe through similarity calculation and a preset threshold. Based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, the pose of the current frame relative to the map is obtained to complete the pose initialization of the current frame. Even in the absence of a positioning signal and when the vehicle has not driven to the starting point of the map, the pose initialization of the current frame on the map can be achieved quickly.

[0046] Of course, any product implementing this invention does not necessarily need to achieve all of the technical effects described above at the same time.

[0047] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0049] Figure 1 This is a flowchart illustrating a memory parking initialization method provided by the present invention;

[0050] Figure 2 This is another flowchart illustrating the memory parking initialization method provided by the present invention.

[0051] Figure 3 This is a schematic diagram of a memory parking initialization device provided by the present invention.

[0052] In the diagram: 100, Memory Parking Initialization Device; 1, Image Supply Module; 2, Data Processing Module; 3, Similarity Calculation Module; 4, Initialization Module. Detailed Implementation

[0053] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0054] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0055] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0056] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0057] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0058] Example 1

[0059] Combination Figure 1 , Figure 1 This is a flowchart illustrating a memory parking initialization method provided by the present invention, used to explain a specific embodiment of the memory parking initialization method provided by the present invention, including:

[0060] S101: Obtain a map, which includes at least two keyframes, each of which includes the corresponding laser point cloud;

[0061] S102: After distorting the laser point cloud corresponding to the key frame, extract the feature points to obtain the feature description set of the key frame;

[0062] S103: Collect the laser point cloud corresponding to the current frame, and extract feature points after distortion removal of the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame;

[0063] S104: Traverse the feature description sets of all keyframes, match them with the feature description sets of the current frame to obtain the corresponding matching point pairs and record them, and calculate the similarity of each matching point pair set to obtain the corresponding similarity value.

[0064] S105: Determine whether the maximum similarity value is greater than or equal to the preset threshold;

[0065] S1051: If the maximum similarity value is greater than or equal to the preset threshold, the keyframe corresponding to the maximum similarity value is used as the initialization reference keyframe; the pose of the current frame relative to the map is obtained based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, and the pose initialization of the current frame is completed.

[0066] It should be noted that in step S101, when building the map, the movement frequency can be preset, and keyframes can be selected according to the movement frequency interval. Of course, it is not limited to this, and the selection of keyframes can be selected according to actual needs. In step S103, the laser point cloud of the current frame is a set of point cloud data collected by the radar within a preset time period starting from the current moment. In step S105, a preset threshold is set to ensure that the feature description set of the selected keyframes has a high similarity to the feature description set of the current frame, thereby ensuring the reliability of the positioning result and avoiding the feature description set of the keyframe corresponding to the largest similarity value having a low similarity to the feature description set of the current frame, which would lead to errors in subsequent conversion. The value of the preset threshold can be set according to actual needs, and this embodiment does not impose specific limitations on it.

[0067] Understandably, for laser real-time localization and mapping technology, it is necessary to extract the features of laser point clouds. Therefore, the acquisition of the feature description set of key frames does not require additional computing power. Then, the feature description set corresponding to the current frame is obtained and the feature description set of the key frame is quickly matched for similarity to provide the initial pose for localization initialization. The overall computational requirements are relatively small.

[0068] Compared with the prior art, the memory parking initialization method provided by the present invention obtains an initialization reference keyframe through similarity calculation and preset threshold. Based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, the pose of the current frame relative to the map is obtained to complete the pose initialization of the current frame. Even in the absence of a positioning signal (GPS signal) and when the vehicle has not driven to the starting point of the map, the pose initialization of the current frame on the map can be achieved quickly.

[0069] Example 2

[0070] Reference Figure 2 , Figure 2 This is another flowchart illustrating the memory parking initialization method provided by the present invention, to demonstrate another specific embodiment of the memory parking initialization method provided by the present invention, including:

[0071] S101: Obtain a map, which includes at least two keyframes, each of which includes the corresponding laser point cloud.

[0072] S102: After distorting the laser point cloud corresponding to the key frame, extract the feature points to obtain the feature description set of the key frame;

[0073] In step S102, distortion correction is performed on the laser point cloud corresponding to the keyframe, including: selecting the coordinate system of the laser point cloud corresponding to any keyframe as the reference coordinate system; and transforming the laser point cloud corresponding to all keyframes to the reference coordinate system. Feature point extraction is performed on the laser point cloud corresponding to the keyframe, and the extraction methods include, but are not limited to, the Intrinsic Shape Signatures (ISS) method or the Scale Invariant Feature Transform 3D (SIFT3D) method.

[0074] S103: Collect the laser point cloud corresponding to the current frame, and extract feature points after distortion removal of the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame;

[0075] In step S103, distortion correction is performed on the laser point cloud corresponding to the current frame, including transforming the laser point cloud corresponding to the current frame to a reference coordinate system. Similarly, feature points are extracted from the laser point cloud corresponding to the current frame. The extraction methods include, but are not limited to, the Intrinsic Shape Signatures (ISS) method or the Scale Invariant Feature Transform 3D (SIFT3D) method.

[0076] S104: Traverse the feature description sets of all keyframes, match them with the feature description set of the current frame to obtain the corresponding matching point pair set and record them, and calculate the similarity of each matching point pair set to obtain the corresponding similarity value.

[0077] In step S104, the feature description sets of all keyframes are traversed, and each set is matched with the feature description set of the current frame to obtain the corresponding set of matching points, including:

[0078] The feature description set of a keyframe includes keyframe feature points, and each keyframe feature point includes a corresponding high-dimensional feature description vector set of the keyframe.

[0079] The feature description set of the current frame includes the feature points of the current frame, and the feature points of the current frame include the corresponding high-dimensional feature description vector set of the current frame;

[0080] Calculate the maximum bipartite graph matching between the high-dimensional feature description vector sets of all keyframes and the high-dimensional feature description vector sets of all current frames in the feature description set of the keyframes;

[0081] The keyframe feature points corresponding to the keyframe high-dimensional feature description vector set and the current frame feature points corresponding to the current frame high-dimensional feature description vector set of the matching keyframe high-dimensional feature description vector set constitute a matching point pair, and all matching point pairs constitute a matching point pair set.

[0082] Specifically, the Hungarian algorithm can be used to calculate the maximum bipartite graph matching between the high-dimensional feature description vector sets of all keyframes and the high-dimensional feature description vector sets of all current frames in the feature description set. Of course, it is not limited to this.

[0083] S105: Determine whether the maximum similarity value is greater than or equal to the preset threshold.

[0084] S1051: If the maximum similarity value is greater than or equal to the preset threshold, the keyframe corresponding to the maximum similarity value is used as the initialization reference keyframe; the pose of the current frame relative to the map is obtained based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, and the pose initialization of the current frame is completed.

[0085] In step S1051, the pose of the current frame relative to the initialization reference keyframe is obtained as follows: the set of matching point pairs corresponding to the initialization reference keyframe is obtained. Specifically, it can be directly queried from the set of matching point pairs recorded when traversing the feature description set of all keyframes.

[0086] Calculate the pose of the current frame relative to the initial reference keyframe based on the set of matching points corresponding to the initial reference keyframe.

[0087] Calculate the pose of the current frame relative to the initial reference keyframe based on the set of matching point pairs corresponding to the initial reference keyframe, including:

[0088] Construct overdetermined equations based on the set of matching point pairs corresponding to the initial reference keyframe, where the number of matching point pairs in the set of matching point pairs corresponding to the initial reference keyframe is greater than or equal to 4.

[0089] The overdetermined equations are solved using the least squares method to obtain the pose of the current frame relative to the initialization reference keyframe.

[0090] Specifically, the overdetermined equations are calculated in the following manner:

[0091] Q1 = T × P1

[0092] …

[0093] Q i =T×P i

[0094] …

[0095] Q n =T×P n

[0096] Where T is the transformation matrix, (Q i P i P is the i-th pair of matching points in the set of matching point pairs corresponding to the initial reference keyframe. i For the i-th feature point in the current frame, Q i To initialize the feature point of the i-th keyframe in the feature description set of the reference keyframe, 1≤i≤n, n≥4 and n is an integer.

[0097] S1052: If the maximum similarity value is less than the preset threshold, update the next frame to the current frame and repeat steps S103 to S105.

[0098] It should be noted that during radar scanning, the vehicle where the radar is located rotates and translates to a certain extent. Therefore, the laser beams emitted at different times are not in a strictly unified coordinate system and need to be distorted. The laser point cloud is transformed into a coordinate system corresponding to a unified timestamp. The coordinate system can be the coordinate system corresponding to the laser point cloud of any key frame. For example, the coordinate system corresponding to the laser point cloud of the first determined key frame can be selected. Of course, it is not limited to this. Existing technology can be used for laser point cloud distortion correction. This embodiment will not elaborate on this.

[0099] The feature description set of a keyframe includes keyframe feature points, and each keyframe feature point includes a corresponding set of high-dimensional feature description vectors. This set of high-dimensional feature description vectors contains N high-dimensional feature description vectors, where N is a positive number. These vectors can represent high-dimensional features such as normals, curvature, and neighborhood characteristics. Each keyframe feature point has the same number of categories in its corresponding high-dimensional feature description vector. Similarly, the feature description set of the current frame includes current frame feature points, and each current frame feature point has a corresponding set of high-dimensional feature description vectors. This set of high-dimensional feature description vectors contains N high-dimensional feature description vectors. Each current frame feature point has the same number of categories in its corresponding high-dimensional feature description vector, which is also the same as the number of categories in the high-dimensional feature description vectors corresponding to the keyframe feature points, facilitating similarity calculations and bipartite graph matching. Specifically, the feature description set of each keyframe includes at least 4 keyframe feature points, and the feature description set of the current frame includes at least 4 current frame feature points, ensuring that the matching point pair set calculated by the Hungarian algorithm includes at least 4 matching point pairs, thereby realizing the construction of the overdetermined equation.

[0100] Understandably, by performing coordinate system transformation calculations on the current frame and the initialization reference keyframe based on the matching point pair set, the pose of the current frame relative to the initialization reference keyframe is obtained. Based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, the pose of the current frame relative to the map is obtained, thus completing the pose initialization of the current frame. Even in the absence of a positioning signal and when the vehicle has not driven to the starting point of the map, the pose initialization of the current frame on the map can be achieved quickly.

[0101] Example 3

[0102] This embodiment describes another specific implementation of the memory parking initialization method provided by the present invention, including:

[0103] A keyframe is determined at regular intervals, and the pose information A corresponding to each keyframe is recorded. i This forms a laser point cloud P.

[0104] The laser point cloud set P is subjected to distortion correction, 3D feature extraction, and feature description. Specifically, for the laser point cloud of the i-th keyframe, several 3D feature points FP are obtained after processing. ij Its corresponding high-dimensional feature description vector FD ij This constitutes the feature description set S corresponding to the i-th keyframe. i The feature description set of all keyframes constitutes a set S = {S1, S2, ..., S...} N} and save it in the system.

[0105] The laser point cloud acquired in the current frame is subjected to the same process of distortion correction, feature point extraction, and feature description, and a feature description set S for the current frame is generated. now .

[0106] Traverse the set S of keyframe feature descriptions, for the i-th feature description set S i High-dimensional feature description vector set FD i , and the feature description set S of the current frame now High-dimensional feature description vector set FD now Bipartite graph matching is performed, and similarity values ​​are calculated based on the matching results. The keyframe with the highest similarity value exceeding a certain threshold is used as the initial reference keyframe. The n pairs of matching points obtained by matching the initial reference keyframe with the current frame's bipartite graph are saved to the matching point pair set Pair, where n is a positive number and greater than or equal to 4. The two points of the matching point pair are located in the local coordinate system of the current frame and the local coordinate system of the initial reference keyframe, respectively. It is necessary to transform the two points to the same coordinate system, so the following equation is constructed:

[0107] Q1 = T × P1

[0108] …

[0109] Q i =T×P i

[0110] …

[0111] Q n =T×P n

[0112] Where T is the transformation matrix, (Q i P i Let P be the i-th matching pair in the set of matching pairs Pair. i For the i-th feature point in the current frame, Q i Let i be the feature point of the i-th keyframe in the feature description set of the initialization reference keyframe, where 1 ≤ i ≤ n, n ≥ 4, and n is an integer. When n ≥ 4, the equation is an overdetermined equation, and Q... i =T×P iThe equations are established from three dimensions: x, y, and z.

[0113] The transformation matrix is ​​converted to a homogeneous matrix, calculated as follows:

[0114]

[0115] Among them, R 11 R 12 and R 13 For P i The rotation parameter about the x-direction, R 21 R 22 and R 23 For P i The rotation parameter about the y-direction, R 31 R 32 and R 33 For P i The rotation parameters about the z-direction, T1 is P i The variable shifted along the x-direction, T2 is P i The variable shifted along the y-direction, T3 is P i The variable shifted along the z-direction.

[0116] Substituting the transformation matrix into the overdetermined equations for solution, only R is available at this point. 11 R 12 R 13 R 21 R 22 R 23 R 31 R 32 R 33 T1, T2, and T3 are unknowns, and the least squares method is used to solve them to obtain the pose of the current frame relative to the initialization reference keyframe. Since the pose of the initialization reference keyframe on the map is known, the pose of the current frame relative to the map can be obtained by combining the pose of the current frame relative to the initialization reference keyframe, thus completing the pose initialization of the current frame. This does not require the vehicle to be in a position with a positioning signal, nor does it require the vehicle to drive to the starting point of the map.

[0117] Example 4

[0118] Based on the same inventive concept, and referring to Figure 3 , Figure 3 This is a schematic diagram of a memory parking initialization device provided by the present invention, illustrating a specific embodiment of the memory parking initialization device 100 provided by the present invention, including:

[0119] Mapping module 1, coupled to data processing module 2, is used to acquire a map, which includes at least two keyframes;

[0120] Data processing module 2, coupled to similarity calculation module 3, is used to process the laser point cloud corresponding to the key frame to obtain the feature description set of the key frame and to process the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame.

[0121] The similarity calculation module 3, coupled to the initialization module 4, is used to traverse the feature description sets of all keyframes and calculate the similarity value with the feature description set of the current frame.

[0122] Initialization module 4, coupled to keyframe determination module 1, is used to take the keyframe corresponding to the largest similarity value greater than or equal to a preset threshold as the initialization reference keyframe. Based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, the pose of the current frame relative to the map is obtained to complete the initialization.

[0123] Example 5

[0124] Based on the same inventive concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the memory parking initialization method as described in any of the above embodiments.

[0125] Example 6

[0126] Based on the same inventive concept, the present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the memory parking initialization method as described in any of the above embodiments.

[0127] As can be seen from the above embodiments, the memory parking initialization method, device, electronic device, and storage medium provided by the present invention achieve at least the following beneficial effects:

[0128] The memory parking initialization method, device, electronic device, and storage medium provided by the present invention obtain an initialization reference keyframe through similarity calculation and a preset threshold. Based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe on the map, the pose of the current frame relative to the map is obtained to complete the initialization. Even in the absence of a positioning signal and when the vehicle has not driven to the starting point of the map, the pose of the current frame on the map can be quickly initialized.

[0129] While specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. In the above embodiments of the present invention, the descriptions of each embodiment have different focuses; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A memory parking initialization method, characterized in that, include: S101: Obtain a map, which includes at least two keyframes; S102: After distorting the laser point cloud corresponding to the key frame, feature points are extracted to obtain the feature description set of the key frame. The distortion-removal process of the laser point cloud corresponding to the key frame includes: selecting the coordinate system where the laser point cloud corresponding to any key frame is located as the reference coordinate system, and transforming the laser point clouds corresponding to all key frames to the reference coordinate system. S103: Acquire the laser point cloud corresponding to the current frame, extract feature points after distortion correction of the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame, and perform distortion correction of the laser point cloud corresponding to the current frame, including: transforming the laser point cloud corresponding to the current frame to the reference coordinate system; S104: Traverse the feature description sets of all the keyframes, and match them with the feature description sets of the current frame based on maximum bipartite graph matching to obtain the corresponding matching point pair sets and record them. Calculate the similarity of all the matching point pair sets to obtain the corresponding similarity values. S105: Determine whether the largest similarity value is greater than or equal to a preset threshold; If the largest similarity value is greater than or equal to the preset threshold, the keyframe corresponding to the largest similarity value is used as the initialization reference keyframe; an overdetermined equation is constructed based on the set of matching point pairs, and the pose of the current frame relative to the initialization reference keyframe is solved using the least squares method. The pose of the current frame relative to the map is obtained and based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe in the map, thus completing the pose initialization of the current frame. The step of traversing the feature description sets of all the keyframes and matching them with the feature description set of the current frame to obtain the corresponding set of matching points includes: The feature description set of the key frame includes key frame feature points, and the key frame feature points include a corresponding set of high-dimensional feature description vectors of the key frame; wherein the high-dimensional feature description vectors of the key frame include normal, curvature, and neighborhood characteristics. The feature description set of the current frame includes feature points of the current frame, and the feature points of the current frame include the corresponding high-dimensional feature description vector set of the current frame. Using the Hungarian algorithm, the maximum bipartite graph matching of all high-dimensional feature description vector sets of the keyframes and all high-dimensional feature description vector sets of the current frame is calculated. The key frame feature points corresponding to the key frame high-dimensional feature description vector set and the current frame feature points corresponding to the current frame high-dimensional feature description vector set that match the key frame high-dimensional feature description vector set constitute a matching point pair, and all the matching point pairs constitute the matching point pair set.

2. The memory parking initialization method according to claim 1, characterized in that, Determining whether the largest similarity value is greater than or equal to the preset threshold further includes: If the largest similarity value is less than the preset threshold, the next frame is updated to the current frame, and steps S103 to S105 are repeated.

3. The memory parking initialization method according to claim 1, characterized in that, The pose of the current frame relative to the initialization reference keyframe is obtained in the following manner: Obtain the set of matching point pairs corresponding to the initialization reference keyframe; The pose of the current frame relative to the initialization reference keyframe is calculated based on the set of matching points corresponding to the initialization reference keyframe.

4. The memory parking initialization method according to claim 3, characterized in that, Calculating the pose of the current frame relative to the initialization reference keyframe based on the set of matching point pairs corresponding to the initialization reference keyframe includes: Construct an overdetermined equation based on the set of matching point pairs corresponding to the initialization reference keyframe; The overdetermined equations are solved using the least squares method to obtain the pose of the current frame relative to the initialization reference keyframe.

5. The memory parking initialization method according to claim 4, characterized in that, The overdetermined equations are calculated in the following manner: in, For the transformation matrix, ( , ) is the first in the set of matching point pairs corresponding to the initialization reference keyframe. For matching point pairs, For the first Each feature point in the current frame The first feature description set of the initialization reference keyframe Keyframe feature points, , and It is an integer.

6. A memory parking initialization device for implementing the memory parking initialization method as described in any one of claims 1-5, characterized in that, include: The map acquisition module is used to acquire a map, which includes at least two keyframes; The data processing module is used to process the laser point cloud corresponding to the key frame to obtain the feature description set of the key frame and to process the laser point cloud corresponding to the current frame to obtain the feature description set of the current frame. The similarity calculation module is used to traverse the feature description sets of all the key frames, match them with the feature description set of the current frame to obtain the corresponding matching point pair set and record them, and perform similarity calculation on all the matching point pair sets to obtain the corresponding similarity value. An initialization module is used to take the keyframe corresponding to the largest similarity value greater than or equal to a preset threshold as an initialization reference keyframe, obtain the pose of the current frame relative to the map based on the pose of the current frame relative to the initialization reference keyframe and the pose of the initialization reference keyframe in the map, and complete the pose initialization of the current frame.

7. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic device implements the memory parking initialization method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the memory parking initialization method as described in any one of claims 1 to 5.

Images