Laser slam positioning method and device for a geometry-deficient scene

Abstract

This invention relates to the field of robot autonomous navigation technology, and provides a laser SLAM localization method and apparatus for scenes lacking geometric features. The method includes: acquiring an original point cloud collected in a target scene; creating a geometric point cloud map of the target scene based on the original point cloud; and creating a marked point cloud map based on the marked point cloud in the original point cloud; matching the real-time point cloud currently scanned by the robot with the geometric point cloud map to determine a geometric residual; matching the real-time marked point cloud in the real-time point cloud currently scanned by the robot with the marked point cloud map to determine a marked residual; determining a total residual based on the geometric residual, the marked residual, and their respective weights; and optimizing the total residual, determining the robot's current position as the position corresponding to the minimum total residual. This invention accurately determines the robot's current position by optimizing the total residual through a weighted average of the geometric residual and the marked residual.

Description

Technical Field

[0001] This invention relates to the field of robot autonomous navigation technology, and in particular to a laser SLAM localization method and apparatus for scenes lacking geometric features. Background Technology

[0002] Simultaneous Localization and Mapping (SLAM) is one of the core foundational technologies in the field of autonomous robot navigation. This technology uses vehicles such as drones, wheeled unmanned vehicles, Automated Guided Vehicles (AGVs), railcars, robotic dogs, and humanoid robots (collectively referred to as robots) to acquire point cloud data of their surroundings in unknown environments via lidar. It then constructs an environmental map in real time and simultaneously estimates its own position. Its basic working principle involves the lidar emitting a laser beam and receiving reflected signals to measure the distance and shape of the surrounding environment, generating high-precision 3D point cloud data. This point cloud data contains geometric information about the environment. The system calculates the robot's displacement by matching point clouds between consecutive frames, gradually constructing an environmental map.

[0003] In structured environments (such as offices or indoor scenes with well-defined geometry), abundant corner, edge, and planar features provide sufficient constraints for point cloud matching, enabling laser SLAM algorithms to achieve centimeter-level positioning accuracy. However, in environments like tunnels, culverts, and corridors, the environmental structure is typically highly uniform and symmetrical, lacking distinct geometric features. The robot's positioning results drift along the longitudinal direction of tunnels, etc. This is because while the cross-sectional structure provides strong geometric constraints (e.g., the arc or square structure formed by the side walls and ceiling), the point cloud features between consecutive frames are highly similar in the longitudinal direction, failing to provide effective constraints for position estimation. As the robot travels further, this drift accumulates, leading to positioning errors of several meters or even tens of meters, failing to meet the meter-level positioning requirements of inspection tasks. Therefore, traditional laser SLAM suffers from inaccurate positioning in environments lacking geometric features. Summary of the Invention

[0004] This invention provides a laser SLAM positioning method and apparatus for scenes lacking geometric features, in order to solve the problem of inaccurate positioning in traditional laser SLAM in environments lacking geometric features.

[0005] This invention provides a laser SLAM localization method for scenes lacking geometric features, comprising the following steps.

[0006] The original point cloud collected in the target scene is acquired, a geometric point cloud map of the target scene is created based on the original point cloud, and a marked point cloud map is created based on the marked point cloud in the original point cloud. The marked point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold. The geometric residual is determined by matching the real-time point cloud currently being scanned by the robot with the geometric point cloud map. The marker residual is determined by matching the real-time labeled point cloud in the real-time point cloud currently being scanned by the robot with the labeled point cloud map. The total residual is determined based on the geometric residual and the labeled residual, and their respective weights. The total residual is optimized, and the robot position corresponding to the minimum total residual is determined as the robot's current position.

[0007] According to the present invention, a laser SLAM positioning method for scenes lacking geometric features is provided. The target scene includes tunnels, culverts, and corridors. The markers are reflective stickers or reflective bodies, and the markers are set at predetermined intervals along the longitudinal direction of the target scene.

[0008] According to the present invention, a laser SLAM localization method for scenes lacking geometric features is provided, wherein the markers are staggered or symmetrically distributed on both sides of the target scene along the longitudinal direction.

[0009] According to the present invention, a laser SLAM localization method for scenes lacking geometric features is provided, wherein the reflectivity threshold is determined based on the average reflectivity in the target scene plus m times the standard deviation of reflectivity, where m is greater than or equal to 2.

[0010] According to the present invention, a laser SLAM localization method for scenes lacking geometric features creates a marked point cloud map based on the marked point cloud in the original point cloud, comprising: Candidate point clouds are formed by extracting point clouds with reflectance greater than the reflectance threshold from the original point cloud; Voxel mesh filtering and outlier removal are performed on the candidate point cloud to obtain the processed candidate point cloud; The processed candidate point cloud is used as the marker point cloud to create the marker point cloud map.

[0011] According to the present invention, a laser SLAM localization method for scenes lacking geometric features determines the marker residual by matching the real-time marked point cloud in the real-time point cloud currently scanned by the robot with the marked point cloud map. The method includes: Calculate the first Euclidean distance from each point in the real-time marked point cloud to the nearest corresponding point in the marked point cloud map, and determine the first Euclidean distance as the marked residual.

[0012] According to the present invention, a laser SLAM localization method for scenes lacking geometric features determines the marker residual by matching the real-time marked point cloud in the real-time point cloud currently scanned by the robot with the marked point cloud map. The method includes: Calculate the second Euclidean distance from each point in the real-time marked point cloud to the plane normal vector formed by the N nearest points in the marked point cloud map, and determine the second Euclidean distance as the marked residual, wherein the plane normal vector is the plane normal vector passing through the centroid of the N nearest points, and N is greater than or equal to 2.

[0013] The present invention also provides a laser SLAM positioning device for scenes lacking geometric features, comprising the following modules: The point cloud map creation module is used to acquire the original point cloud collected in the target scene, create a geometric point cloud map of the target scene based on the original point cloud, and create a marker point cloud map based on the marker point cloud in the original point cloud. The marker point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold. The geometric residual determination module is used to determine the geometric residual based on the matching of the real-time point cloud currently being scanned by the robot and the geometric point cloud map. The marker residual determination module is used to determine the marker residual by matching the real-time marker point cloud in the real-time point cloud currently scanned by the robot with the marker point cloud map. The total residual determination module is used to determine the total residual based on the geometric residual and the labeled residual and their respective weights. The total residual optimization module is used to optimize the total residual and determine the robot's current position as the robot position corresponding to the minimum total residual.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the laser SLAM localization method for scenes lacking geometric features as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the laser SLAM localization method for scenes lacking geometric features as described above.

[0016] The laser SLAM localization method and apparatus for scenes lacking geometric features provided by this invention acquires the original point cloud collected in the target scene, creates a geometric point cloud map of the target scene based on the original point cloud, and creates a marked point cloud map based on the marked point cloud in the original point cloud. The real-time point cloud and real-time marked point cloud currently scanned by the robot are matched with the corresponding point cloud maps to determine the geometric residual and the marked residual. Since the markers produce obvious high reflectivity characteristics in the original point cloud, which form a sharp contrast with the natural features of other areas of the target scene, they provide strong constraints for point cloud matching. Therefore, the current position of the robot is accurately determined by optimizing the total residual obtained by weighting the geometric residual and the marked residual. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the laser SLAM localization method for scenes lacking geometric features provided by the present invention.

[0019] Figure 2 This is one of the schematic diagrams of marker placement in the laser SLAM positioning method for scenes lacking geometric features provided by the present invention.

[0020] Figure 3 This is the second schematic diagram of marker placement in the laser SLAM positioning method for scenes lacking geometric features provided by this invention.

[0021] Figure 4 This is a schematic diagram of the laser SLAM positioning device for scenes lacking geometric features provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] In existing technologies, due to the lack of obvious geometric features in structured environments, robot localization results can drift along longitudinal directions such as tunnels, leading to inaccurate final localization results. The traditional approach to solving this problem is to introduce multi-sensor fusion schemes, such as laser SLAM, visual SLAM combined with an inertial measurement unit (IMU) navigation system. While these schemes do improve system robustness and localization accuracy, IMUs themselves suffer from integration drift, and errors still accumulate during long-distance navigation, especially in linear tunnel structures where the lack of salient features required for loop closure detection makes it difficult to effectively correct accumulated errors through backend optimization, thus affecting localization accuracy. Furthermore, sensor (e.g., camera) calibration is complex, computational resource requirements are high, and system complexity increases. This is particularly problematic for applications like tunnel inspection, where strict limitations on equipment cost, weight, and power consumption often make adding additional sensors like cameras impractical.

[0025] To address the issue of inaccurate localization using traditional SLAM-based positioning techniques in scenarios lacking geometric features, this invention provides a laser SLAM localization method for such scenarios. The specific process is as follows: Figure 1 As shown, it includes steps S110 to S150.

[0026] Step S110: Obtain the original point cloud collected in the target scene, create a geometric point cloud map of the target scene based on the original point cloud, and create a marker point cloud map based on the marker point cloud in the original point cloud. The marker point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold.

[0027] Since the target scene lacks geometric features, such as tunnels, culverts, and corridors, markers with reflectivity greater than a reflectivity threshold are pre-placed in the target scene. These markers produce a significant high reflectivity in the original point cloud, creating a sharp contrast with the natural features of other areas of the target scene and providing strong constraints for point cloud matching. In this step, a geometric point cloud map of the target scene is created from the original point cloud (which includes the marker point cloud), and a marked point cloud map of the target scene is created from the marked point cloud. Both point cloud maps are saved as independent layers for subsequent robot localization.

[0028] Step S120: Based on the matching of the real-time point cloud currently scanned by the robot and the geometric point cloud map, determine the geometric residual. The geometric residual reflects the distance deviation between the points in the real-time point cloud and the corresponding points in the geometric point cloud map after rotation and translation transformation, when the robot is in its current pose (including position).

[0029] Step S130: Based on the matching of the real-time labeled point cloud in the real-time point cloud currently scanned by the robot and the labeled point cloud map, the labeled residual is determined. The labeled residual reflects the distance deviation between the points in the real-time labeled point cloud and the corresponding points in the labeled point cloud map after rotation and translation transformation, when the robot is in its current pose (including position).

[0030] Step S140: Determine the total residual based on the geometric residual and the labeled residual and their respective weights. Specifically, the formula for calculating the total residual r_total is as follows: r_total = w_geom × r_geom + w_intensity × r_intensity.

[0031] Here, w_geom and w_intensity are the weights of the geometric residual r_geom and the labeled residual r_intensity, respectively, with a sum of 1. Their respective magnitudes can be adaptively adjusted according to the richness of environmental features. In regions with degraded geometric features, the weight of the labeled residual is increased, allowing it to play a dominant role in the optimization. Preferably, since the geometric environmental features in the target scene are relatively poor, the weight of the labeled residual can be directly set to be greater than the weight of the geometric residual, thereby more accurately locating the robot's position.

[0032] Step S150: Optimize the total residual, and determine the robot's current position as the position corresponding to the minimum total residual. Since the residual reflects the degree of distance deviation between the points in the real-time point cloud and the corresponding points in the point cloud map after rotation and translation transformation, the smaller the total residual, the smaller the degree of distance deviation, and the more accurate the robot's positioning in the target scene.

[0033] This embodiment of the laser SLAM localization method for scenes lacking geometric features acquires the original point cloud collected in the target scene, creates a geometric point cloud map of the target scene based on the original point cloud, and creates a marked point cloud map based on the marked point cloud in the original point cloud. The robot's current scanned real-time point cloud and real-time marked point cloud are matched with the corresponding point cloud maps to determine the geometric residual and the marked residual. Since the markers are manually added, they are discontinuous and separate, and have strong geometric features compared to the target scene. Based on their high reflectivity, they can be easily extracted from the original scanned point cloud, thus providing strong constraints for point cloud matching. Therefore, by optimizing the total residual obtained by weighting the geometric residual and the marked residual, the robot's current position can be accurately determined.

[0034] In some embodiments, the target scene includes tunnels, culverts, and corridors. The markers are reflective stickers or reflective bodies. Preferably, the markers should have regular geometric shapes (e.g., circles, squares) and standardized dimensions to facilitate algorithm recognition and matching. Different shapes can be used to encode different segments, that is, different shapes of markers are placed in different segments of the target scene, thereby forming a correspondence between the shape of the markers and the location segment. The point cloud shape of the markers can preliminarily determine the current location segment of the robot in the scene, thus providing additional pose constraints.

[0035] The spacing between markers should be flexibly designed according to the target scenario, such as variations in tunnel length and curvature. Figure 2 As shown, markers are placed at predetermined distances (e.g., 10 meters) along the longitudinal direction of the target scene, with adjacent markers spaced at the same predetermined distance. This predetermined distance also provides additional constraints. Furthermore, at curves or structural changes in the target scene, the predetermined distance can be appropriately shortened, meaning the markers are placed more densely to provide more constraints at curves or structural changes, resulting in more accurate robot localization. For example, markers are placed at predetermined distances L along the length (longitudinal) direction of at least one side wall of a tunnel, and in the lateral direction of the target scene, such as the tunnel's height, adjacent markers are staggered to provide even more constraints.

[0036] Furthermore, such as Figure 3 As shown, the markers are staggered or symmetrically distributed on both sides of the target scene along the longitudinal direction to ensure that a sufficient number of marker points can be captured from different angles, thus providing more constraints for point cloud matching.

[0037] In some embodiments, the reflectivity threshold is determined based on the average reflectivity in the target scene plus m times the standard deviation of reflectivity, where m is greater than or equal to 2, to clearly distinguish it from the reflectivity in the target scene. That is, reflectivity threshold = average reflectivity + (m × standard deviation of reflectivity), preferably, m is 2 or 3.

[0038] If reflectance follows a normal distribution, the threshold can be set to "average reflectance + 2 times the standard deviation of reflectance". Theoretically, only about 2.5% of the data will exceed this threshold. If it is "average reflectance + 3 times the standard deviation of reflectance", only about 0.15% of the data will exceed this threshold. Targets that exceed this threshold have extremely high reflectance and do not belong to the background or ordinary objects of the target scene. They can be considered as significant markers.

[0039] Taking a tunnel as an example, the layout parameters of the markers are shown in Table 1 below: Table 1. Marker Layout Parameters

[0040] In some embodiments, step S110, creating a marked point cloud map based on the marked point cloud in the original point cloud, includes: Step S111: Extract point clouds with reflectance greater than the reflectance threshold from the original point cloud to form candidate point clouds. By analyzing the reflectance of each point in the original point cloud, select points with reflectance greater than the reflectance threshold to form candidate point clouds.

[0041] Step S112: Perform voxel mesh filtering and outlier removal on the candidate point cloud to obtain the processed candidate point cloud. In this step, voxel mesh filtering is applied to the candidate point cloud to reduce point cloud density, avoid overfitting local features, and improve subsequent matching efficiency. A statistical outlier removal algorithm can be used to remove isolated outliers with excessively high reflectivity caused by noise, ensuring the purity of the marked point cloud map, thereby enabling accurate robot localization during subsequent point cloud matching.

[0042] Step S113: Create the marked point cloud map using the processed candidate point cloud as the marked point cloud.

[0043] In some embodiments, step S130, which involves matching the real-time labeled point cloud in the real-time point cloud currently scanned by the robot with the labeled point cloud map, to determine the labeled residual, specifically includes: Calculate the first Euclidean distance from each point in the real-time marked point cloud to the nearest corresponding point in the marked point cloud map, and determine the first Euclidean distance as the marked residual. That is, in this embodiment, the point-to-point residual is used as the marked residual, and the formula for calculating the first Euclidean distance is as follows: r_point = ||T × p_current – ​​p_map|| 2 .

[0044] Where r_point represents the first Euclidean distance, also known as the marker residual; T represents the robot's current pose transformation matrix; p_current represents the coordinates of a point in the real-time marker point cloud; p_map represents the coordinates of the nearest point in the marker point cloud map corresponding to p_current; and || represents the calculation of the L2 norm, i.e., the calculation of the Euclidean distance. It can be understood that before matching, the real-time marker point cloud needs to be transformed to the world coordinate system corresponding to the marker point cloud map.

[0045] In this embodiment, the nearest point between the marker point cloud map and the real-time marker point cloud can be obtained by searching the KDTree. Specifically, the marker point cloud map can be added to the KDTree, which is similar to the spatial binary search method for finding the nearest point in the real-time marker point cloud. This can greatly improve the effect of finding the nearest point, without having to calculate the distance with every point in the marker point cloud.

[0046] In some embodiments, step S130, which involves matching the real-time labeled point cloud in the real-time point cloud currently scanned by the robot with the labeled point cloud map, to determine the labeled residual, specifically includes: Calculate the second Euclidean distance from each point in the real-time marked point cloud to the plane normal vector formed by the N nearest points in the marked point cloud map, and determine the second Euclidean distance as the marker residual. Here, the plane normal vector is the plane normal vector passing through the centroids of the N nearest points, where N is greater than or equal to 2. That is, in this embodiment, the point-to-plane normal residual is used as the marker residual. The formula for calculating the second Euclidean distance is as follows: r_plane = ||(T × p_current – ​​c_map) × n||.

[0047] Where r_plane is the second Euclidean distance, also known as the label residual, p_current represents the coordinates of points in the real-time labeled point cloud, c_map is the centroid of the N nearest points in the labeled point cloud map, and n is the normal vector of the plane passing through c_map. Specifically, n can be obtained by fitting using Principal Component Analysis (PCA). PCA estimates the normal vector from a set of coplanar points, essentially utilizing the variance distribution characteristics of the data: the direction of the normal vector is the direction of least data variation. The specific steps for fitting n using PCA are as follows: Step 1: Calculation N The closest points that are coplanar (or approximately coplanar) p 1, p 2…, p N center of mass c (i.e., the center point), the calculation formula is as follows: .

[0048] Where p1, p2, ..., p N These represent the three-dimensional coordinates of N points.

[0049] Step 2: Subtract the centroid from each of the nearest points to obtain the centered point set: q i = p i - c .

[0050] Step 3: Calculate the 3×3 covariance matrix based on the centered point set: .

[0051] in, express The transpose of the matrix describes the variance and covariance of the data points in three dimensions (x, y, and z).

[0052] Step 4: Adjust the covariance matrix C Eigenvalue decomposition yields three eigenvalues. λ 1. λ 2. λ 3 ( λ 1≥ λ 2≥ λ (3≥0) and the unit eigenvectors corresponding to the three eigenvalues. v 1. v 2. v 3: .

[0053] Step 5: Determine the unit eigenvector corresponding to the smallest eigenvalue as the plane normal vector n. Specifically, for an ideal coplanar set of points, the data exhibits the greatest variation (largest variance) in two mutually perpendicular directions within the plane, and the corresponding eigenvalues... λ 1 and λ 2 is relatively large, but its variation is minimal (small variance) in the direction perpendicular to the plane, corresponding to the eigenvalue. λ 3 is close to zero. Therefore, it is close to the smallest eigenvalue. λ 3 corresponding eigenvectors v 3 is the unit normal vector n of the plane in question.

[0054] It should be noted that the present invention is not limited to the point-to-point residual and point-to-plane normal residual mentioned above. Other residual calculation methods that can achieve the same effect and realize accurate matching of marked point clouds are also used in the present invention.

[0055] The laser SLAM positioning device for scenes lacking geometric features provided by the present invention will be described below. The laser SLAM positioning device for scenes lacking geometric features described below can be referred to in correspondence with the laser SLAM positioning method for scenes lacking geometric features described above.

[0056] The laser SLAM positioning device for scenes lacking geometric features according to embodiments of the present invention, such as Figure 4 As shown, it includes: The point cloud map creation module 410 is used to acquire the original point cloud collected in the target scene, create a geometric point cloud map of the target scene based on the original point cloud, and create a marker point cloud map based on the marker point cloud in the original point cloud. The marker point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold.

[0057] The geometric residual determination module 420 is used to determine the geometric residual based on the matching of the real-time point cloud currently scanned by the robot and the geometric point cloud map.

[0058] The marker residual determination module 430 is used to determine the marker residual by matching the real-time marker point cloud in the real-time point cloud currently scanned by the robot with the marker point cloud map.

[0059] The total residual determination module 440 is used to determine the total residual based on the geometric residual and the labeled residual and their respective weights.

[0060] The total residual optimization module 450 is used to optimize the total residual and determine the robot's current position as the robot position corresponding to the minimum total residual.

[0061] This embodiment of the laser SLAM positioning device for scenes lacking geometric features acquires the original point cloud collected in the target scene, creates a geometric point cloud map of the target scene based on the original point cloud, and creates a marked point cloud map based on the marked point cloud in the original point cloud. The robot's current real-time point cloud and real-time marked point cloud are matched with the corresponding point cloud maps to determine the geometric residual and the marked residual. Since the markers produce obvious high reflectivity in the original point cloud, which contrasts sharply with the natural features of other areas of the target scene, it provides strong constraints for point cloud matching. Therefore, the robot's current position is accurately determined by optimizing the total residual obtained by weighting the geometric residual and the marked residual.

[0062] In some embodiments, the target scene includes: tunnels, culverts and corridors, and the markers are reflective stickers or reflective bodies, and the markers are set at predetermined intervals along the longitudinal direction of the target scene.

[0063] In some embodiments, the markers are staggered or symmetrically distributed on both sides of the target scene along the longitudinal direction.

[0064] In some embodiments, the reflectivity threshold is determined based on the average reflectivity in the target scene plus m times the standard deviation of reflectivity, where m is greater than or equal to 2.

[0065] In some embodiments, the point cloud map creation module 410 is specifically used for: Points with reflectance greater than the reflectance threshold are extracted from the original point cloud to form candidate point clouds; voxel mesh filtering and outlier removal are performed on the candidate point clouds to obtain processed candidate point clouds; the processed candidate point clouds are used as the marked point clouds to create the marked point cloud map.

[0066] In some embodiments, the marker residual determination module 430 is specifically used to calculate the first Euclidean distance from each point in the real-time marker point cloud to the corresponding nearest point in the marker point cloud map, and determine the first Euclidean distance as the marker residual.

[0067] In some embodiments, the marker residual determination module 430 is specifically used to calculate the second Euclidean distance from each point in the real-time marker point cloud to the plane normal vector formed by the corresponding N nearest points in the marker point cloud map, and determine the second Euclidean distance as the marker residual, wherein the plane normal vector is the plane normal vector passing through the centroid of the N nearest points, and N is greater than or equal to 2.

[0068] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a laser SLAM localization method for scenes lacking geometric features. This method includes: The original point cloud collected in the target scene is acquired. A geometric point cloud map of the target scene is created based on the original point cloud. A marked point cloud map is also created based on the marked point cloud in the original point cloud. The marked point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features. The reflectivity of the markers is greater than the reflectivity threshold.

[0069] The geometric residual is determined by matching the real-time point cloud currently being scanned by the robot with the geometric point cloud map.

[0070] The marker residual is determined by matching the real-time marked point cloud in the real-time point cloud currently being scanned by the robot with the marked point cloud map.

[0071] The total residual is determined based on the geometric residual and the labeled residual, and their respective weights.

[0072] The total residual is optimized, and the robot position corresponding to the minimum total residual is determined as the robot's current position.

[0073] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the laser SLAM localization method for scenes lacking geometric features provided by the above methods, the method comprising: The original point cloud collected in the target scene is acquired. A geometric point cloud map of the target scene is created based on the original point cloud. A marked point cloud map is also created based on the marked point cloud in the original point cloud. The marked point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features. The reflectivity of the markers is greater than the reflectivity threshold.

[0075] The geometric residual is determined by matching the real-time point cloud currently being scanned by the robot with the geometric point cloud map.

[0076] The marker residual is determined by matching the real-time marked point cloud in the real-time point cloud currently being scanned by the robot with the marked point cloud map.

[0077] The total residual is determined based on the geometric residual and the labeled residual, and their respective weights.

[0078] The total residual is optimized, and the robot position corresponding to the minimum total residual is determined as the robot's current position.

[0079] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the laser SLAM localization method for scenes lacking geometric features provided by the methods described above, the method comprising: The original point cloud collected in the target scene is acquired. A geometric point cloud map of the target scene is created based on the original point cloud. A marked point cloud map is also created based on the marked point cloud in the original point cloud. The marked point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features. The reflectivity of the markers is greater than the reflectivity threshold.

[0080] The geometric residual is determined by matching the real-time point cloud currently being scanned by the robot with the geometric point cloud map.

[0081] The marker residual is determined by matching the real-time marked point cloud in the real-time point cloud currently being scanned by the robot with the marked point cloud map.

[0082] The total residual is determined based on the geometric residual and the labeled residual, and their respective weights.

[0083] The total residual is optimized, and the robot position corresponding to the minimum total residual is determined as the robot's current position.

[0084] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0085] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0086] All actions involving the acquisition of signal information or data in this invention are carried out in compliance with the relevant data protection laws and policies of the country where the device is located, and with the authorization granted by the owner of the device.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A laser SLAM localization method for scenes lacking geometric features, characterized in that, include: The original point cloud collected in the target scene is acquired, a geometric point cloud map of the target scene is created based on the original point cloud, and a marked point cloud map is created based on the marked point cloud in the original point cloud. The marked point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold. The geometric residual is determined by matching the real-time point cloud currently being scanned by the robot with the geometric point cloud map. The marker residual is determined by matching the real-time labeled point cloud in the real-time point cloud currently being scanned by the robot with the labeled point cloud map. The total residual is determined based on the geometric residual and the labeled residual, and their respective weights. The total residual is optimized, and the robot position corresponding to the minimum total residual is determined as the robot's current position.

2. The laser SLAM localization method for scenes lacking geometric features according to claim 1, characterized in that, The target scene includes tunnels, culverts and corridors, and the markers are reflective stickers or reflective bodies, which are set at predetermined intervals along the longitudinal direction of the target scene.

3. The laser SLAM localization method for scenes lacking geometric features according to claim 2, characterized in that, The markers are staggered or symmetrically distributed on both sides of the target scene along the longitudinal direction.

4. The laser SLAM localization method for scenes lacking geometric features according to claim 1, characterized in that, The reflectivity threshold is determined based on the average reflectivity in the target scene plus m times the standard deviation of reflectivity, where m is greater than or equal to 2.

5. The laser SLAM localization method for scenes lacking geometric features according to any one of claims 1 to 4, characterized in that, Creating a marked point cloud map based on the marked point cloud in the original point cloud includes: Candidate point clouds are formed by extracting point clouds with reflectance greater than the reflectance threshold from the original point cloud; Voxel mesh filtering and outlier removal are performed on the candidate point cloud to obtain the processed candidate point cloud; The processed candidate point cloud is used as the marker point cloud to create the marker point cloud map.

6. The laser SLAM localization method for scenes lacking geometric features according to any one of claims 1 to 4, characterized in that, Based on the matching of the real-time labeled point cloud in the real-time point cloud currently scanned by the robot and the labeled point cloud map, the labeled residual is determined, including: Calculate the first Euclidean distance from each point in the real-time marked point cloud to the nearest corresponding point in the marked point cloud map, and determine the first Euclidean distance as the marked residual.

7. The laser SLAM localization method for scenes lacking geometric features according to any one of claims 1 to 4, characterized in that, Based on the matching of the real-time labeled point cloud in the real-time point cloud currently scanned by the robot and the labeled point cloud map, the labeled residual is determined, including: Calculate the second Euclidean distance from each point in the real-time marked point cloud to the plane normal vector formed by the N nearest points in the marked point cloud map, and determine the second Euclidean distance as the marked residual, wherein the plane normal vector is the plane normal vector passing through the centroid of the N nearest points, and N is greater than or equal to 2.

8. A laser SLAM positioning device for scenes lacking geometric features, characterized in that, include: The point cloud map creation module is used to acquire the original point cloud collected in the target scene, create a geometric point cloud map of the target scene based on the original point cloud, and create a marker point cloud map based on the marker point cloud in the original point cloud. The marker point cloud is the point cloud corresponding to the markers deployed in the target scene to enhance geometric features, and the reflectivity of the markers is greater than the reflectivity threshold. The geometric residual determination module is used to determine the geometric residual based on the matching of the real-time point cloud currently being scanned by the robot and the geometric point cloud map. The marker residual determination module is used to determine the marker residual by matching the real-time marker point cloud in the real-time point cloud currently scanned by the robot with the marker point cloud map. The total residual determination module is used to determine the total residual based on the geometric residual and the labeled residual and their respective weights. The total residual optimization module is used to optimize the total residual and determine the robot's current position as the robot position corresponding to the minimum total residual.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the laser SLAM localization method for scenes lacking geometric features as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the laser SLAM localization method for scenes lacking geometric features as described in any one of claims 1 to 7.

Images