Point cloud registration methods, apparatus, equipment and storage media

Abstract

This application provides a point cloud registration method, apparatus, device, and storage medium. In this embodiment, based on door information in a spatial object and combined with door connection information between 3D point cloud datasets, pose estimation is performed on the 3D point cloud datasets of each acquisition point. Specifically, 2D door point information in a 2D real-world image is detected and converted into 3D door point information. Based on the 3D door point information of the 3D point cloud dataset and combined with door connection information, the relative pose information between the 3D point cloud datasets is estimated. Throughout this process, a sufficient number of feature matching pairs are not required; point cloud registration is performed based on the 3D door point information corresponding to the door information, thus improving the accuracy of determining relative pose information.

Description

Technical Field

[0001] This application relates to the field of 3D reconstruction technology, and in particular to a point cloud registration method, apparatus, device and storage medium. Background Technology

[0002] Point cloud registration is the process of calculating the relative pose (rigid transformation or Euclidean transformation) between two point clouds and transforming the source point cloud to the same coordinate system as the target point cloud. Currently, to obtain ideal point cloud registration results, sufficient feature matching pairs can be obtained through feature descriptors such as Scale-invariant Feature Transform (SIFT), Oriented Fast and Rotated BRIEF (ORB), and Signatures of Histograms (SHOT). Point cloud registration is then performed based on these feature matching pairs to determine the relative pose between the two point clouds. However, using feature matching pairs to solve for the relative pose results in low accuracy, affecting the final point cloud registration outcome. Summary of the Invention

[0003] This application provides a point cloud registration method, apparatus, device, and storage medium to improve the accuracy of point cloud registration.

[0004] This application provides a point cloud registration method, comprising: acquiring a 3D point cloud dataset and a 2D real-scene image collected at various acquisition points for multiple spatial objects, wherein each 3D point cloud dataset and each 2D real-scene image contains at least one door information of the spatial object to which it belongs; wherein the multiple spatial objects belong to a target physical space, and each spatial object has one or more acquisition points; according to the transformation relationship between the radar coordinate system and the camera coordinate system, converting the 2D door information in each 2D real-scene image to the corresponding 3D point cloud dataset to obtain the 3D door information of the 3D point cloud dataset, wherein the 2D door information is the intersection information of the corner point and the ground in the door information; based on the 3D door information of each 3D point cloud dataset, combined with the door connection information between the 3D point cloud datasets of each acquisition point, determining the first relative pose information between the 3D point cloud datasets of each acquisition point, so as to achieve point cloud registration between the 3D point cloud datasets of each acquisition point.

[0005] This application embodiment also provides a point cloud registration device, including: an acquisition module, a conversion module, and a determination module; the acquisition module is used to acquire three-dimensional point cloud datasets and two-dimensional real-scene images collected at various acquisition points in multiple spatial objects, each three-dimensional point cloud dataset and each two-dimensional real-scene image containing at least one door information in its respective spatial object; wherein, the multiple spatial objects belong to a target physical space, and each spatial object is provided with one or more acquisition points; the conversion module is used to convert the two-dimensional door point information in each two-dimensional real-scene image to the three-dimensional point cloud dataset corresponding to the two-dimensional real-scene image according to the conversion relationship between the radar coordinate system and the camera coordinate system, to obtain the three-dimensional door point information of the three-dimensional point cloud dataset, wherein the two-dimensional door point information is the intersection information of the corner point and the ground in the door information; the determination module is used to determine the first relative pose information between the three-dimensional point cloud datasets of each acquisition point based on the three-dimensional door point information of each three-dimensional point cloud dataset and the door connection information between the three-dimensional point cloud datasets of each acquisition point, so as to realize point cloud registration between the three-dimensional point cloud datasets of each acquisition point.

[0006] This application also provides a point cloud registration device, including: a memory and a processor; the memory for storing a computer program; and the processor, coupled to the memory, for executing the computer program to implement the steps in the point cloud registration method provided in this application.

[0007] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps in the point cloud registration method provided in this application.

[0008] In this embodiment, based on the door information in the spatial object and the door connection information between the three-dimensional point cloud datasets, pose estimation is performed on the three-dimensional point cloud datasets of each acquisition point. Specifically, two-dimensional door point information in the two-dimensional real-scene image is detected and converted into three-dimensional door point information. Based on the three-dimensional door point information of the three-dimensional point cloud dataset and combined with the door connection information, the relative pose information between the three-dimensional point cloud datasets is estimated. Throughout the process, a sufficient number of feature matching pairs are not required. Point cloud registration is performed based on the three-dimensional door point information corresponding to the door information, which improves the accuracy of determining the relative pose information. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0010] Figure 1 A flowchart illustrating a point cloud registration method provided for an exemplary embodiment of this application;

[0011] Figure 2 A schematic diagram of a point cloud registration device provided for an exemplary embodiment of this application;

[0012] Figure 3 A schematic diagram of a point cloud registration device provided for an exemplary embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] To address the low accuracy of point cloud registration in existing technologies, this application embodiment uses door information from spatial objects, combined with door connection information between 3D point cloud datasets, to perform pose estimation on the 3D point cloud datasets of each acquisition point. Specifically, it detects 2D door point information in 2D real-world images, converts the 2D door point information into 3D door point information, and estimates the relative pose information between 3D point cloud datasets based on the 3D door point information and door connection information. Throughout this process, it eliminates the need for a sufficient number of feature matching pairs, performing point cloud registration based on the 3D door point information corresponding to the door information, thus improving the accuracy of determining relative pose information.

[0015] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating a point cloud registration method provided for an exemplary embodiment of this application. Figure 1 As shown, the method includes:

[0017] 101. Acquire 3D point cloud datasets and 2D real-world images collected at various acquisition points in multiple spatial objects. Each 3D point cloud dataset and each 2D real-world image contains at least one door information in its respective spatial object. Among them, multiple spatial objects belong to the target physical space, and each spatial object has one or more acquisition points.

[0018] 102. Based on the transformation relationship between the radar coordinate system and the camera coordinate system, the two-dimensional door point information in each two-dimensional real scene image is transformed into the three-dimensional point cloud dataset corresponding to the two-dimensional real scene image to obtain the three-dimensional door point information of the three-dimensional point cloud dataset. The two-dimensional door point information is the intersection information of the corner point and the ground in the door information.

[0019] 103. Based on the 3D gate point information of each 3D point cloud dataset, and combined with the gate connection information between the 3D point cloud datasets of each acquisition point, determine the first relative pose information between the 3D point cloud datasets of each acquisition point, so as to achieve point cloud registration between the 3D point cloud datasets of each acquisition point.

[0020] In this embodiment, the target physical space refers to a specific spatial area containing multiple spatial objects; in other words, multiple spatial objects constitute the target physical space. For example, the target physical space refers to a house, and the multiple spatial objects included in the house can be a kitchen, bedroom, living room, or bathroom, etc. One or more sampling points can be set in each spatial object, and the specific number of sampling points depends on the size and shape of the spatial object or the placement of objects in the physical space.

[0021] In this embodiment, a 3D point cloud dataset can be collected at each acquisition point. A laser radar (LiDAR) can be used to collect the 3D point cloud dataset of the spatial object at each acquisition point. A LiDAR is a system that uses laser beams to detect the spatial structure of a target physical space. Its working principle is to emit a detection signal (laser beam) at each acquisition point towards an object in the target physical space (such as a wall, door, or window), and then compare the received signal reflected back from the object (echo) with the emitted signal to obtain relevant information about the object, such as distance, orientation, height, velocity, attitude, and shape parameters. When a laser beam illuminates the surface of an object, the reflected laser carries information such as orientation and distance. If the laser beam is scanned along a certain trajectory, and the information of the reflected laser points is recorded as the scan is performed, a large number of laser points can be obtained due to the extremely fine scanning, thus forming a 3D point cloud dataset.

[0022] This can be achieved by using a camera to capture two-dimensional real-scene images. The implementation of these images varies depending on the type of camera. For example, if the camera is implemented as a panoramic camera, the two-dimensional real-scene image will be a panoramic image; conversely, if the camera is implemented as a fisheye camera, the two-dimensional real-scene image will be a fisheye image.

[0023] The installation positions of the camera and LiDAR are not limited. For example, the camera and LiDAR can be at a certain angle in the horizontal direction, such as 90 degrees, 180 degrees, or 270 degrees, and also at a certain distance in the vertical direction, such as 0 cm, 1 cm, or 5 cm. The camera and LiDAR can also be fixed to a pan-tilt unit on a bracket, rotating with the unit. During rotation, the LiDAR acquires a 3D point cloud dataset corresponding to the spatial object at the acquisition point, while the camera acquires a 2D real-world image of the corresponding spatial object at the acquisition point. The transformation relationship between the LiDAR coordinate system and the camera coordinate system can be obtained based on the installation positions of the camera and LiDAR.

[0024] In this embodiment, each spatial object contains at least one door information. For example, the living room contains three door information entries, sharing the same door information as the master bedroom, secondary bedroom, and bathroom. The master bedroom contains one door information entry, the secondary bedroom contains one door information entry, and the bathroom contains one door information entry. Based on this, the 3D point cloud dataset acquired by LiDAR and the 2D real-world images acquired by the camera contain at least one door information entry for each spatial object.

[0025] Among them, the two-dimensional real scene image contains two-dimensional door point information, which is the intersection information of the corner point of the door body information and the ground. Typically, a door body information has four corner points, and two of the four corner points intersect with the ground. That is, the two-dimensional door point information corresponds to two corner point information in the door body information.

[0026] In this embodiment, door and window detection can be performed on each two-dimensional real-scene image to obtain the two-dimensional door point information contained in each two-dimensional real-scene image. For example, door and window detection can be performed on each two-dimensional real-scene image using an object detection algorithm, which includes, but is not limited to, Fast Region-Convolutional Neural Networks (Fast-R-CNN), such as the You Only Look Once (YOLO) model or the Single Shot MultiBox Detector (SSD), etc.

[0027] After detecting 2D gate information in a 2D real-world image, the 2D gate information can be transformed from the camera coordinate system to the radar coordinate system to obtain the 3D gate information in the corresponding 3D point cloud dataset. The 3D point cloud dataset corresponding to the 2D real-world image is a 3D point cloud dataset acquired at the same acquisition point as the 2D real-world image. Specifically, based on the transformation relationship between the radar coordinate system and the camera coordinate system, the 2D gate information in each 2D real-world image is transformed into the corresponding 3D point cloud dataset to obtain the 3D gate information in the 3D point cloud dataset.

[0028] For example, when a 2D real-world image is converted into a panoramic image, the coordinates of the 2D gate information in the image coordinate system are (c, r). According to the transformation relationship between the image coordinate system and the spherical coordinate system, the 2D gate information in the image coordinate system is transformed to the spherical coordinate system to obtain the 3D gate information Pb = (xb, yb, zb) in the spherical coordinate system. Assuming the height of the camera is hc, the 3D gate information in the spherical coordinate system is transformed to the camera coordinate system to obtain Pc = hc / yb*Pb = hc / yb(xb, yb, zb). Assuming the calibration matrix between the camera coordinate system and the radar coordinate system is Tm, the coordinates of the 3D gate information in the radar coordinate system are Pl = Pc*Tm = hc / yb*Pb*Tm = hc / yb*Rm*Pb+tm, where Rm and tm are the rotation matrix and translation matrix of the calibration matrix Tm, respectively.

[0029] In this embodiment, considering that point cloud registration between two 3D point cloud datasets involves rotation or translation transformations, the point cloud registration problem is actually an optimization problem with nonlinear, non-convex functions and a large number of local extrema. To obtain ideal point cloud registration results, a sufficient number of feature matching pairs can be obtained through feature descriptors such as SIFT, ORB, and SHOT, thereby solving the relative pose between the two 3D point cloud datasets. However, using feature matching pairs to solve for the relative pose may be limited by the small number of feature matching pairs, which can lead to low accuracy in solving the relative pose information.

[0030] Considering that the target physical space contains multiple gates, and each gate can connect two spatial objects, if every gate of a spatial object is already connected to other spatial objects, then it is considered that the spatial object does not need to be connected to other spatial objects. That is, the 3D point cloud dataset corresponding to the spatial object does not need to be registered with the 3D point cloud datasets corresponding to other spatial objects. Based on this, in this embodiment, the relative pose between 3D point cloud datasets is estimated according to the 3D gate point information corresponding to the gate information and combined with the gate connection relationship between 3D point cloud datasets. This is no longer limited by the number of feature matching pairs, thus improving the accuracy of pose information estimation. Specifically, based on the 3D gate point information of each 3D point cloud dataset and combined with the gate connection information between 3D point cloud datasets, the first relative pose information between the 3D point cloud datasets of each acquisition point is determined.

[0031] The door connection information reflects the connection relationship between doors within a spatial object and other spatial objects. Each spatial object contains one or more door information entries. For example, the target physical space contains doors A1, A2, and A3. The living room contains three door information entries: A1, A2, and A3. The master bedroom contains door A1, the secondary bedroom contains door A2, and the bathroom contains door A3. The door connection information could be that the living room and master bedroom are connected via door A1, the living room and secondary bedroom via door A2, the living room and bathroom via door A3, and so on. Furthermore, the door connection information can also represent each spatial object. Alternatively, the door connection information could be that the 3D point cloud dataset E1 corresponding to the living room and the 3D point cloud dataset E2 corresponding to the master bedroom are connected via door A1, the 3D point cloud dataset E1 corresponding to the living room and the 3D point cloud dataset E3 corresponding to the secondary bedroom are connected via door A2, and the 3D point cloud dataset E1 corresponding to the living room and the 3D point cloud dataset E4 corresponding to the bathroom are connected via door A3.

[0032] The gate connection information can be pre-defined, for example, it can include the connection relationships between gates in the target physical space; or, it can be established while performing point cloud registration on the 3D point cloud datasets. For example, after each point cloud registration of two 3D point cloud datasets, gate connection information between the two datasets is established. If subsequent gates in 3D point cloud dataset M1 establish gate connection information with other 3D point cloud datasets, then 3D point cloud dataset M1 will no longer participate in the estimation of subsequent relative pose information, reducing the computational load of pose estimation. For example, based on the gate connection information between 3D point cloud datasets, the 3D point cloud datasets that can participate in pose estimation can be determined; based on the 3D gate point information of the 3D point cloud datasets that can participate in pose estimation, the first relative pose information between the 3D point cloud datasets can be determined.

[0033] For example, if the gate connection information between 3D point cloud datasets indicates that spatial objects F1 and F2 have the same gate information, then the 2D real-world image B1 corresponding to spatial object F1 and the 2D real-world image B2 corresponding to spatial object F2 can be determined. The gate X in the 2D real-world image B1 corresponds to the gate Y in the 2D real-world image B2. The 2D gate point information in the 2D real-world images B1 and B2 is obtained, and the 2D gate point information in the 2D real-world images B1 and B2 is converted into 3D gate point information in the 3D point cloud datasets C1 and C2, respectively. Based on the 3D gate point information in the 3D point cloud datasets C1 and C2, the first relative pose information between the 3D point cloud datasets C1 and C2 is determined.

[0034] For example, if the target physical space includes three-dimensional point cloud datasets G1, G2, and G3, and the gate connection information indicates that three-dimensional point cloud datasets G1 and G2 have the same gate information, and three-dimensional point cloud datasets G1 and G3 also have the same gate information, and all gate information of three-dimensional point cloud dataset G2 is connected to the gates in other three-dimensional point cloud datasets, then the first relative pose information between three-dimensional point cloud datasets G1 and G3 can be determined based on the three-dimensional gate information of three-dimensional point cloud dataset G1 and the three-dimensional gate information of three-dimensional point cloud dataset G3.

[0035] Optionally, after obtaining the first relative pose information between the 3D point cloud datasets at each acquisition point, point cloud fusion can be performed on the 3D point cloud datasets at each acquisition point based on the first relative position information to obtain the 3D point cloud dataset corresponding to the target physical space. Alternatively, after obtaining the first relative pose information between the 3D point cloud datasets at each acquisition point, the first relative position information can be used as the initial relative pose information between the 3D point cloud datasets. The Iterative Closest Point (ICP) algorithm or the Normal Distribution Transform (NDT) algorithm can be used to perform fine registration on the 3D point cloud datasets at each acquisition point. Based on the pose information of the 3D point cloud datasets at each acquisition point obtained from the fine registration, the 3D point cloud datasets at each acquisition point can be fused to obtain the 3D point cloud dataset corresponding to the target physical space.

[0036] In this embodiment, based on the door information in the spatial object and the door connection information between the three-dimensional point cloud datasets, pose estimation is performed on the three-dimensional point cloud datasets of each acquisition point. Specifically, two-dimensional door point information in the two-dimensional real-scene image is detected and converted into three-dimensional door point information. Based on the three-dimensional door point information of the three-dimensional point cloud dataset and combined with the door connection information, the relative pose information between the three-dimensional point cloud datasets is estimated. Throughout the process, a sufficient number of feature matching pairs are not required. Point cloud registration is performed based on the three-dimensional door point information corresponding to the door information, which improves the accuracy of determining the relative pose information.

[0037] In one optional embodiment, an implementation method for determining the first relative pose information between 3D point cloud datasets at each acquisition point based on the 3D gate information of each 3D point cloud dataset and the gate connection information between the 3D point cloud datasets at each acquisition point includes: sequentially determining the target point cloud dataset according to a set point cloud registration order; the set point cloud registration order can be the order in which the 3D point cloud datasets were acquired, or it can be the point cloud registration order between the 3D point cloud datasets based on the relative positional relationship of multiple spatial objects; and determining at least one candidate point cloud dataset based on the gate connection information between the 3D point cloud datasets at each acquisition point. Each 3D point cloud dataset may contain one gate information or multiple gate information, and each gate information maintains gate connection information, indicating that the gate information has established a connection relationship with gate information in other 3D point cloud datasets. The at least one candidate point cloud dataset is a 3D point cloud dataset whose gate information has not yet established a connection relationship with the gate information of other 3D point cloud datasets. Based on the 3D gate point information of the target point cloud dataset and each candidate point cloud dataset, the second relative pose information corresponding to each candidate point cloud dataset is estimated. Based on the second relative pose information corresponding to each candidate point cloud dataset, a first candidate point cloud dataset is selected as the source point cloud dataset from at least one candidate point cloud dataset. The second relative pose information corresponding to the first candidate point cloud dataset is used as the first relative pose information between the source point cloud dataset and the target point cloud dataset. Specifically, the second relative pose information corresponding to the first candidate point cloud dataset is the second relative pose information that minimizes the point cloud error between the candidate point cloud dataset and the target point cloud dataset.

[0038] Optionally, an implementation method for selecting a first candidate point cloud dataset as a source point cloud dataset from at least one candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset includes: performing pose transformation on each candidate point cloud dataset according to the second relative pose information corresponding to each candidate point cloud dataset, and calculating first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset. The method for calculating the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset is not limited; for example, it can be calculated for each 3D point p in the pose-transformed candidate point cloud dataset. i Obtain the three-dimensional point p i The nearest neighbor 3D point q in the target point cloud dataset i The three-dimensional point q i The normal vector at point n i Therefore, the three-dimensional point p can be calculated. i With three-dimensional point q i Point-to-plane distance on the plane Considering that there are often many outliers between the candidate point cloud dataset and the target point cloud dataset for registration, a third distance threshold d is set to enhance robustness against outliers. m If the point-to-surface distance d exceeds the third distance threshold, then the third distance threshold d will be set. m As the distance d between points and surfaces, then calculate multiple 3D points p. i With three-dimensional point q i The average of the point-to-surface distances between the candidate and target point cloud datasets is used as the first distance information between them. This first distance information can be used as the registration quality score for the second relative pose information corresponding to the candidate point cloud dataset. Where, m i This score represents the number of 3D points in the registered point clouds of the candidate point cloud dataset and the target point cloud dataset. This score reflects the degree of fit between the candidate point cloud dataset and the target point cloud dataset.

[0039] After calculating the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset, a first candidate point cloud dataset can be selected as the source point cloud dataset from at least one candidate point cloud dataset based on the first distance information between the target point cloud dataset and each candidate point cloud dataset after pose transformation. For example, the candidate point cloud dataset corresponding to the smallest first distance information can be used as the first candidate point cloud dataset, and the first candidate point cloud dataset can be used as the source point cloud dataset. Alternatively, multiple candidate point cloud datasets corresponding to first distance information exceeding a set first distance threshold can be identified, and the first candidate point cloud dataset can be selected as the source point cloud dataset from multiple 3D point cloud datasets based on the relative positional relationship of spatial objects; wherein the relative positional relationship of spatial objects is obtained through other sensors, such as a GPS positioning module, a WiFi positioning module, or even a Simultaneous Localization and Mapping (SLAM) module. In one optional embodiment, after point cloud registration of the 3D point cloud dataset, gate connection information between the registered 3D point clouds can be established. Specifically, based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset, a second distance information between the 3D gate information in the source point cloud dataset and the destination point cloud dataset is calculated. For example, the 3D gate information in the source point cloud dataset is pose transformed based on the first relative pose information, and the second distance information between the pose-transformed 3D gate information and the 3D gate information in the destination point cloud dataset is calculated. If the second distance information is less than a set second distance threshold, the 3D gate information in the source point cloud dataset and the destination point cloud dataset are considered to be successfully matched, and gate connection information between the 3D gate information in the source point cloud dataset and the destination point cloud dataset can be established.

[0040] Optionally, each 3D gate point information includes: two 3D corner point information. The center point information of the two 3D corner points in each 3D gate point information can be calculated for both the source and destination point cloud datasets, respectively, to obtain source center point information and destination center point information. The source point cloud dataset contains one or more 3D gate point information. For each 3D gate point information, the center point information of the two 3D corner points in that 3D gate point information is calculated, referred to as source center point information. Similarly, the destination point cloud dataset may also contain one or more 3D gate point information. For each 3D gate point information, the center point information of the two 3D corner points in that 3D gate point information is calculated, referred to as destination center point information. A third distance information between the source center point information and the destination center point information is calculated based on the first relative pose information between the source and destination point cloud datasets. For example, the pose of the source center point information can be transformed using the first relative pose information, and the third distance information between the transformed source center point information and the destination center point information can be calculated. This third distance information is used as the second distance information between the 3D gate point information in the source and destination point cloud datasets.

[0041] Optionally, if the second distance information is greater than or equal to a set second distance threshold, it is considered that the 3D gate point information in the source point cloud dataset and the target point cloud dataset has not been matched successfully, indicating that the accuracy of point cloud registration between the target point cloud dataset and the source point cloud dataset is low. The pose information of the target point cloud dataset and at least one candidate point cloud dataset provided by other sensors can be obtained. The other sensors include at least: a WIFI sensor, a GPS sensor, or a SLAM module. Based on the relative positional relationship of multiple spatial objects, a source point cloud dataset corresponding to the target point cloud data is selected from at least one candidate point cloud dataset. Based on the pose information of the target point cloud dataset and the source point cloud dataset provided by other sensors, the first relative pose information between the target point cloud dataset and the source point cloud dataset is determined.

[0042] In one optional embodiment, an implementation method for determining at least one candidate point cloud dataset based on gate connection information between 3D point cloud datasets includes: When performing point cloud registration on a 3D point cloud dataset of a target spatial object for the first time, if no connection relationship is established between the 3D gate point information of the 3D point cloud datasets, then the 3D point cloud datasets corresponding to each acquisition point, excluding the target point cloud dataset, can be used as at least one candidate point cloud dataset. When performing point cloud registration on a 3D point cloud dataset of a target spatial object for subsequent times, the previous point cloud registration process has already established connection relationships between the 3D gate point information that participated in the point cloud registration. Therefore, gate connection information between the 3D gate point information that participated in the point cloud registration can be obtained. If the gate connection information indicates that the 3D gate point information contained in the first 3D point cloud dataset has already established connection relationships with other 3D point cloud datasets, it means that the first 3D point cloud dataset cannot be further registered with other 3D point cloud datasets. In this case, the 3D point cloud datasets corresponding to each acquisition point, excluding the target point cloud dataset and the first 3D point cloud dataset, can be used as at least one candidate point cloud dataset.

[0043] In one optional embodiment, an implementation of estimating second relative pose information between a target point cloud dataset and each candidate point cloud dataset based on 3D gate point information of the target point cloud dataset and each candidate point cloud dataset includes: matching at least two gate point pairs based on the respective 3D gate point information in the target point cloud dataset and each candidate point cloud dataset; and estimating the second relative pose information between the target point cloud dataset and each candidate 3D point cloud dataset based on the at least two gate point pairs.

[0044] For example, the target point cloud dataset may contain one or more 3D gate information, each of which includes two 3D corner information. Similarly, the candidate point cloud dataset may also contain one or more 3D gate information. For example, a 3D gate information H0 in the target point cloud dataset includes 3D corner information H1 and H2, and a 3D gate information J0 in the candidate point cloud dataset includes 3D corner information J1 and J2. Matching the 3D gate information H0 with the 3D gate information J0 yields two gate pair information, namely: gate pair information K1(H1, J1) and gate pair information K2(H2, J2). If the target point cloud dataset or the candidate point cloud dataset contains multiple 3D gate information, then multiple (even number) gate pair information can be obtained.

[0045] In this design, the lidar can be fixed to the rotation axis of the support or a gimbal device, rotating with the rotation axis or gimbal device to acquire a 3D point cloud dataset at the acquisition point. Rotation of the 3D point cloud dataset involves rotating it around the vertical axis, while translation involves translating the acquisition point on the horizontal plane. The relative pose information between the 3D point cloud datasets is a two-dimensional rigid body transformation with three degrees of freedom (y-axis during rotation and x- and z-axis during translation). Each gate point pair obtained through matching can provide two linear equations. Based on each gate point pair, a second relative pose information corresponding to the candidate point cloud dataset can be estimated. In other words, for one 3D gate point information in the target point cloud dataset and one in the candidate point cloud dataset, two gate point pairs can be matched, thus obtaining two second relative pose information corresponding to the candidate point cloud dataset.

[0046] Accordingly, for the two second relative pose information corresponding to the candidate point cloud dataset, pose transformation can be performed on the candidate point cloud dataset respectively, and the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset can be calculated; wherein, the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset can be calculated based on multiple candidate point cloud datasets, and the candidate point cloud dataset corresponding to the smallest first distance information is taken as the source point cloud dataset, and the second relative pose information used to calculate the smallest first distance information is taken as the first relative pose information between the source point cloud dataset and the target point cloud dataset.

[0047] In one optional embodiment, an outlier may exist in the 3D point cloud dataset. For example, 3D point cloud datasets are collected in the living room and dining room respectively. The 3D points shared by these two collection points are mainly within the space of the living room and dining room. 3D points outside the space of the living room and dining room are likely outliers, such as 3D points outside the doors of the living room or dining room. Based on this, 3D points outside the door information corresponding to the 3D door point information can be filtered out according to the 3D door point information in the 3D point cloud dataset. Specifically, before estimating the first relative pose information between 3D point cloud datasets based on the 3D door point information of each 3D point cloud dataset and the door connection information between the 3D point cloud datasets, the plane where the door corresponding to the 3D door point information is located can also be determined according to the 3D door point information of each 3D point cloud dataset; based on the positional relationship between the 3D points in the 3D point cloud dataset and the plane where the door is located, 3D points outside the space object corresponding to the 3D point cloud dataset can be filtered out. For example, the 3D door point information in the 3D point cloud dataset and its vertical direction can be used to estimate the plane where the door is located. Assume the estimated plane where the door is located is: Where n is the normal vector of the plane containing the door, ρ is a constant representing the position of the plane containing the door, and x is a 3D point in the 3D point cloud dataset. If the following conditions are met... If so, the 3D point is considered to be outside the door and needs to be filtered out; if If a 3D point is located on the plane of the door or inside the door, then there is no need to filter out that 3D point.

[0048] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 101 to 103 can be device A; or the execution subject of steps 101 and 102 can be device A, and the execution subject of step 103 can be device B; and so on.

[0049] Furthermore, some processes described in the above embodiments and accompanying drawings include multiple operations appearing in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0050] Figure 2 A schematic diagram of a point cloud registration device provided for an exemplary embodiment of this application is shown below. Figure 2 The point cloud registration device shown includes: an acquisition module 21, a conversion module 22, and a determination module 23.

[0051] The acquisition module 21 is used to acquire three-dimensional point cloud datasets and two-dimensional real-scene images collected at various acquisition points in multiple spatial objects. Each three-dimensional point cloud dataset and each two-dimensional real-scene image contains at least one door information in its respective spatial object. Among them, multiple spatial objects belong to the target physical space, and each spatial object has one or more acquisition points.

[0052] The conversion module 22 is used to convert the two-dimensional door point information in each two-dimensional real scene image to the three-dimensional point cloud dataset corresponding to the two-dimensional real scene image according to the conversion relationship between the radar coordinate system and the camera coordinate system, so as to obtain the three-dimensional door point information of the three-dimensional point cloud dataset. The two-dimensional door point information is the intersection information of the corner point and the ground in the door information.

[0053] The determination module 23 is used to determine the first relative pose information between the three-dimensional point cloud datasets of each acquisition point based on the three-dimensional gate point information of each three-dimensional point cloud dataset and the gate connection information between the three-dimensional point cloud datasets of each acquisition point, so as to realize the point cloud registration between the three-dimensional point cloud datasets of each acquisition point.

[0054] In one optional embodiment, the determining module is specifically configured to: sequentially determine the target point cloud dataset according to a set point cloud registration order; determine at least one candidate point cloud dataset based on the gate connection information between the 3D point cloud datasets of each acquisition point; estimate the second relative pose information corresponding to each candidate point cloud dataset based on the 3D gate point information of the target point cloud dataset and each candidate point cloud dataset; select a first candidate point cloud dataset as the source point cloud dataset from at least one candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset; and use the second relative pose information corresponding to the first candidate point cloud dataset as the first relative pose information between the source point cloud dataset and the target point cloud dataset.

[0055] In an optional embodiment, the determining module is specifically used to: perform pose transformation on each candidate point cloud dataset according to the second relative pose information corresponding to each candidate point cloud dataset, and calculate the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset; and select a first candidate point cloud dataset as the source point cloud dataset from at least one candidate point cloud dataset according to the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset.

[0056] In an optional embodiment, the determining module is specifically used to: when performing point cloud registration on a 3D point cloud dataset in the target physical space for the first time, take a 3D point cloud dataset other than the target point cloud dataset as at least one candidate point cloud dataset; when performing point cloud registration on a 3D point cloud dataset in the target physical space for a subsequent time, if the gate connection information between the 3D point cloud datasets that have participated in point cloud registration indicates that the 3D gate point information contained in the first 3D point cloud dataset has established a connection relationship with other 3D point cloud datasets, then take a 3D point cloud dataset other than the target point cloud dataset and the first 3D point cloud dataset as at least one candidate point cloud dataset.

[0057] In one optional embodiment, the determining module is specifically configured to: match at least two gate point pairs based on the 3D gate point information in the target point cloud dataset and each candidate point cloud dataset; and estimate the second relative pose information corresponding to each candidate 3D point cloud dataset based on the at least two gate point pairs.

[0058] In an optional embodiment, the point cloud registration device further includes: a calculation module and an establishment module; the calculation module is used to calculate second distance information between the three-dimensional gate information in the source point cloud dataset and the target point cloud dataset based on the first relative pose information between the source point cloud dataset and the target point cloud dataset; the establishment module is used to establish gate connection information between the three-dimensional gate information in the source point cloud dataset and the target point cloud dataset if the second distance information is less than a set second distance threshold.

[0059] In an optional embodiment, the 3D gate point information includes: two 3D corner point information; the calculation module is specifically used to: calculate the center point information of the two 3D corner point information in each 3D gate point information for the source point cloud dataset and the destination point cloud dataset respectively, to obtain the source center point information and the destination center point information respectively; calculate the third distance information between the source center point information and the destination center point information based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset; and use the third distance information as the second distance information between the 3D gate point information in the source point cloud dataset and the destination point cloud dataset.

[0060] In an optional embodiment, the point cloud registration device further includes: a selection module; the acquisition module is further configured to: if the second distance information is greater than or equal to a set second distance threshold, acquire the pose information of the target point cloud dataset and at least one candidate point cloud dataset provided by other sensors; the other sensors include at least: a wireless communication sensor, a positioning sensor, or a real-time positioning and mapping module; the selection module is configured to select a source point cloud dataset corresponding to the target point cloud data from at least one candidate point cloud dataset according to the relative positional relationship of multiple spatial objects; the determination module is further configured to: determine the first relative pose information between the target point cloud dataset and the source point cloud dataset according to the pose information of the target point cloud dataset and the source point cloud dataset provided by other sensors.

[0061] In an optional embodiment, the point cloud registration device further includes: a filtering module; before estimating the first relative pose information between the three-dimensional point cloud datasets based on the three-dimensional gate point information of each three-dimensional point cloud dataset and the gate connection information between the three-dimensional point cloud datasets, the determining module is further configured to: determine the plane where the gate corresponding to the three-dimensional gate point information is located based on the three-dimensional gate point information of each three-dimensional point cloud dataset; the filtering module is configured to: filter out three-dimensional points outside the spatial object corresponding to the three-dimensional point cloud dataset based on the positional relationship between the three-dimensional points in the three-dimensional point cloud dataset and the plane where the gate is located.

[0062] For details regarding the implementation of the point cloud registration device, please refer to the aforementioned embodiments, which will not be repeated here.

[0063] The point cloud registration device provided in this application embodiment estimates the pose of the 3D point cloud datasets of each acquisition point based on the door information in the spatial object and the door connection information between the 3D point cloud datasets. Specifically, it detects the 2D door point information in the 2D real scene image, converts the 2D door point information into 3D door point information, and estimates the relative pose information between the 3D point cloud datasets based on the 3D door point information and the door connection information. In the whole process, it does not require a sufficient number of feature matching pairs. Point cloud registration is performed based on the 3D door point information corresponding to the door information, which improves the accuracy of determining the relative pose information.

[0064] Figure 3 This is a schematic diagram of a point cloud registration device provided for an exemplary embodiment of this application. Figure 3 As shown, the device includes a memory 34 and a processor 35.

[0065] Memory 34 is used to store computer programs and can be configured to store various other data to support operations on the point cloud registration device. Examples of this data include instructions for any application or method used to operate on the point cloud registration device.

[0066] The memory 34 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0067] The processor 35, coupled to the memory 34, executes a computer program in the memory 34 to: acquire 3D point cloud datasets and 2D real-world images collected at various acquisition points in multiple spatial objects, wherein each 3D point cloud dataset and each 2D real-world image contains at least one door information of its respective spatial object; wherein the multiple spatial objects belong to a target physical space, and each spatial object has one or more acquisition points; according to the transformation relationship between the radar coordinate system and the camera coordinate system, convert the 2D door information in each 2D real-world image to the corresponding 3D point cloud dataset to obtain the 3D door information of the 3D point cloud dataset, wherein the 2D door information is the intersection information of the corner point and the ground in the door information; based on the 3D door information of each 3D point cloud dataset, combined with the door connection information between the 3D point cloud datasets of each acquisition point, determine the first relative pose information between the 3D point cloud datasets of each acquisition point to achieve point cloud registration between the 3D point cloud datasets of each acquisition point.

[0068] In an optional embodiment, when the processor 35 determines the first relative pose information between the three-dimensional point cloud datasets at each acquisition point based on the three-dimensional gate point information of each three-dimensional point cloud dataset and the gate connection information between the three-dimensional point cloud datasets at each acquisition point, it specifically performs the following steps: sequentially determining the target point cloud dataset according to a set point cloud registration order; determining at least one candidate point cloud dataset based on the gate connection information between the three-dimensional point cloud datasets at each acquisition point; estimating the second relative pose information corresponding to each candidate point cloud dataset based on the three-dimensional gate point information of the target point cloud dataset and each candidate point cloud dataset; selecting a first candidate point cloud dataset as the source point cloud dataset from at least one candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset; and using the second relative pose information corresponding to the first candidate point cloud dataset as the first relative pose information between the source point cloud dataset and the target point cloud dataset.

[0069] In an optional embodiment, when the processor 35 selects a first candidate point cloud dataset as a source point cloud dataset from at least one candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset, it is specifically configured to: perform pose transformation on each candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset, and calculate the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset; and select the first candidate point cloud dataset as a source point cloud dataset from at least one candidate point cloud dataset based on the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset.

[0070] In an optional embodiment, when the processor 35 determines at least one candidate point cloud dataset based on the gate connection information between the three-dimensional point cloud datasets of each acquisition point, it specifically performs the following: when performing point cloud registration on the three-dimensional point cloud dataset in the target physical space for the first time, it uses the three-dimensional point cloud dataset other than the target point cloud dataset as at least one candidate point cloud dataset; when performing point cloud registration on the three-dimensional point cloud dataset in the target physical space for a subsequent time, if the gate connection information between the three-dimensional point cloud datasets that have participated in point cloud registration indicates that the three-dimensional gate point information contained in the first three-dimensional point cloud dataset has established a connection relationship with other three-dimensional point cloud datasets, it uses the three-dimensional point cloud dataset other than the target point cloud dataset and the first three-dimensional point cloud dataset as at least one candidate point cloud dataset.

[0071] In an optional embodiment, when the processor 35 estimates the second relative pose information corresponding to each candidate point cloud dataset based on the 3D gate information of the target point cloud dataset and each candidate point cloud dataset, it is specifically used to: match at least two gate point pairs based on the respective 3D gate information in the target point cloud dataset and each candidate point cloud dataset; and estimate the second relative pose information corresponding to each candidate 3D point cloud dataset based on the at least two gate point pairs.

[0072] In an optional embodiment, the processor 35 is further configured to: calculate second distance information between the three-dimensional gate information in the source point cloud dataset and the destination point cloud dataset based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset; and if the second distance information is less than a set second distance threshold, establish gate connection information between the three-dimensional gate information in the source point cloud dataset and the destination point cloud dataset.

[0073] In an optional embodiment, the 3D gate point information includes: two 3D corner point information; when the processor 35 calculates the second distance information between the 3D gate point information in the source point cloud dataset and the target point cloud dataset based on the first relative pose information between the source point cloud dataset and the target point cloud dataset, it specifically performs the following: for the source point cloud dataset and the target point cloud dataset, calculate the center point information of the two 3D corner point information in each 3D gate point information, respectively, to obtain the source center point information and the target center point information; calculate the third distance information between the source center point information and the target center point information based on the first relative pose information between the source point cloud dataset and the target point cloud dataset; and use the third distance information as the second distance information between the 3D gate point information in the source point cloud dataset and the target point cloud dataset.

[0074] In an optional embodiment, the processor 35 is further configured to: if the second distance information is greater than or equal to a set second distance threshold, acquire the pose information of the target point cloud dataset and at least one candidate point cloud dataset provided by other sensors; the other sensors include at least: a wireless communication sensor, a positioning sensor, or a real-time positioning and mapping module; select a source point cloud dataset corresponding to the target point cloud data from at least one candidate point cloud dataset according to the relative positional relationship of multiple spatial objects; and determine the first relative pose information between the target point cloud dataset and the source point cloud dataset according to the pose information of the target point cloud dataset and the source point cloud dataset provided by other sensors.

[0075] In an optional embodiment, before estimating the first relative pose information between the three-dimensional point cloud datasets based on the three-dimensional gate point information of each three-dimensional point cloud dataset and the gate connection information between the three-dimensional point cloud datasets, the processor 35 is further configured to: determine the plane where the gate corresponding to the three-dimensional gate point information is located based on the three-dimensional gate point information of each three-dimensional point cloud dataset; and filter out three-dimensional points outside the spatial object corresponding to the three-dimensional point cloud dataset based on the positional relationship between the three-dimensional points in the three-dimensional point cloud dataset and the plane where the gate is located.

[0076] For details regarding the implementation of the point cloud registration equipment, please refer to the aforementioned embodiments, which will not be repeated here.

[0077] The point cloud registration device provided in this application embodiment estimates the pose of the 3D point cloud datasets of each acquisition point based on the door information in the spatial object and the door connection information between the 3D point cloud datasets. Specifically, it detects the 2D door point information in the 2D real scene image, converts the 2D door point information into 3D door point information, and estimates the relative pose information between the 3D point cloud datasets based on the 3D door point information and the door connection information. In the whole process, it does not require a sufficient number of feature matching pairs. Point cloud registration is performed based on the 3D door point information corresponding to the door information, which improves the accuracy of determining the relative pose information.

[0078] Furthermore, such as Figure 3 As shown, the point cloud registration device also includes other components such as a communication component 36, a display 37, a power supply component 38, and an audio component 39. Figure 3 The diagram only shows a portion of the components and does not imply that the point cloud registration device only includes... Figure 3 The components shown. It should be noted that... Figure 3 The components within the dashed box are optional, not mandatory, and their specific requirements depend on the product type of the point cloud registration device.

[0079] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the functions provided in embodiments of this application. Figure 1 The steps in the method shown.

[0080] The above Figure 3The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA) technology, Ultra-Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0081] The above Figure 3 The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions, but also the duration and pressure associated with the touch or swipe operation.

[0082] The above Figure 3 The power supply component provides power to the various components of the device in which it resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which it resides.

[0083] The above Figure 3 The audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0084] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0089] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0090] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0091] It should also be noted that 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 limitation, 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.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A point cloud registration method, characterized in that, include: Acquire 3D point cloud datasets and 2D real-scene images collected at various acquisition points in multiple spatial objects. Each 3D point cloud dataset and each 2D real-scene image contains at least one door information in its respective spatial object. The multiple spatial objects belong to the target physical space, and each spatial object has one or more acquisition points. Based on the transformation relationship between the radar coordinate system and the camera coordinate system, the two-dimensional door point information in each two-dimensional real scene image is transformed into the three-dimensional point cloud dataset corresponding to the two-dimensional real scene image to obtain the three-dimensional door point information of the three-dimensional point cloud dataset. The two-dimensional door point information is the intersection information of the corner point and the ground in the door information. The target point cloud dataset is determined sequentially according to the set point cloud registration order; Based on the gate connection information between the three-dimensional point cloud datasets of each acquisition point, at least one candidate point cloud dataset is determined. Based on the target point cloud dataset and the 3D gate point information of each candidate point cloud dataset, estimate the second relative pose information corresponding to each candidate point cloud dataset; Based on the second relative pose information corresponding to each candidate point cloud dataset, a first candidate point cloud dataset is selected as the source point cloud dataset from at least one candidate point cloud dataset. The second relative pose information corresponding to the first candidate point cloud dataset is used as the first relative pose information between the source point cloud dataset and the destination point cloud dataset.

2. The method according to claim 1, characterized in that, Based on the second relative pose information corresponding to each candidate point cloud dataset, a first candidate point cloud dataset is selected as the source point cloud dataset from at least one candidate point cloud dataset, including: Based on the second relative pose information corresponding to each candidate point cloud dataset, pose transformation is performed on each candidate point cloud dataset, and the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset is calculated. Based on the first distance information between each candidate point cloud dataset after pose transformation and the target point cloud dataset, a first candidate point cloud dataset is selected from at least one candidate point cloud dataset as the source point cloud dataset.

3. The method according to claim 1, characterized in that, Based on the gate connection information between the 3D point cloud datasets of each acquisition point, at least one candidate point cloud dataset is determined, including: In the case of performing point cloud registration on a 3D point cloud dataset in the target physical space for the first time, a 3D point cloud dataset other than the target point cloud dataset is regarded as at least one candidate point cloud dataset. If, in cases where point cloud registration is not performed on a 3D point cloud dataset in the target physical space for the first time, the gate connection information between the 3D point cloud datasets that have participated in point cloud registration indicates that the 3D gate point information contained in the first 3D point cloud dataset has established connection relationships with other 3D point cloud datasets, then the 3D point cloud datasets other than the target point cloud dataset and the first 3D point cloud dataset will be regarded as at least one candidate point cloud dataset.

4. The method according to claim 1, characterized in that, Based on the target point cloud dataset and the 3D gate point information of each candidate point cloud dataset, the second relative pose information corresponding to each candidate point cloud dataset is estimated, including: Based on the target point cloud dataset and the 3D gate point information in each candidate point cloud dataset, at least two gate point pairs are matched. Based on the at least two gate point pairs, estimate the second relative pose information corresponding to each candidate 3D point cloud dataset.

5. The method according to claim 1, characterized in that, Also includes: Based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset, calculate the second distance information between the three-dimensional gate point information in the source point cloud dataset and the destination point cloud dataset; If the second distance information is less than the set second distance threshold, then establish gate connection information between the three-dimensional gate point information in the source point cloud dataset and the destination point cloud dataset.

6. The method according to claim 5, characterized in that, The 3D gate information includes: information on two 3D corner points; based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset, a second distance information is calculated between the 3D gate information in the source point cloud dataset and the destination point cloud dataset, including: For the source point cloud dataset and the destination point cloud dataset, calculate the center point information of the two three-dimensional corner points in each three-dimensional gate point information to obtain the source center point information and the destination center point information respectively; Based on the first relative pose information between the source point cloud dataset and the destination point cloud dataset, calculate the third distance information between the source center point information and the destination center point information; The third distance information is used as the second distance information between the three-dimensional gate point information in the source point cloud dataset and the destination point cloud dataset.

7. The method according to claim 5, characterized in that, Also includes: If the second distance information is greater than or equal to the set second distance threshold, then the pose information of the target point cloud dataset and at least one candidate point cloud dataset provided by other sensors is obtained; The other sensors include at least: wireless communication sensors, positioning sensors, or real-time positioning and mapping modules; Based on the relative positional relationships of multiple spatial objects, a source point cloud dataset corresponding to the target point cloud data is selected from at least one candidate point cloud dataset. Based on the pose information of the target point cloud dataset and the source point cloud dataset provided by the other sensors, a first relative pose information between the target point cloud dataset and the source point cloud dataset is determined.

8. The method according to claim 1, characterized in that, Before determining the target point cloud dataset according to the set point cloud registration order, the method further includes: determining the plane where the gate body corresponding to the three-dimensional gate point information is located based on the three-dimensional gate point information of each three-dimensional point cloud dataset; Based on the positional relationship between the 3D points in the 3D point cloud dataset and the plane where the door is located, 3D points outside the spatial object corresponding to the 3D point cloud dataset are filtered out.

9. A point cloud registration device, characterized in that, include: Acquisition module, conversion module, and determination module; The acquisition module is used to acquire three-dimensional point cloud datasets and two-dimensional real-scene images collected at various acquisition points in multiple spatial objects. Each three-dimensional point cloud dataset and each two-dimensional real-scene image contains at least one door information in its respective spatial object. The multiple spatial objects belong to the target physical space, and each spatial object has one or more acquisition points. The conversion module is used to convert the two-dimensional door point information in each two-dimensional real scene image to the three-dimensional point cloud dataset corresponding to the two-dimensional real scene image according to the conversion relationship between the radar coordinate system and the camera coordinate system, so as to obtain the three-dimensional door point information of the three-dimensional point cloud dataset. The two-dimensional door point information is the intersection information of the corner point and the ground in the door information. The determining module is configured to: sequentially determine the target point cloud dataset according to a set point cloud registration order; determine at least one candidate point cloud dataset based on the gate connection information between the 3D point cloud datasets of each acquisition point; estimate the second relative pose information corresponding to each candidate point cloud dataset based on the 3D gate point information of the target point cloud dataset and each candidate point cloud dataset; select a first candidate point cloud dataset as the source point cloud dataset from at least one candidate point cloud dataset based on the second relative pose information corresponding to each candidate point cloud dataset; and use the second relative pose information corresponding to the first candidate point cloud dataset as the first relative pose information between the source point cloud dataset and the target point cloud dataset.

10. A point cloud registration device, characterized in that, include: Memory and processor; The memory is used to store a computer program; the processor, coupled to the memory, is used to execute the computer program to implement the steps of the method according to any one of claims 1-8.

11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method according to any one of claims 1-8.

Images