SLAM loopback detection method and system based on three-dimensional point cloud height intensity density information

Abstract

The invention relates to the field of laser radar detection, in particular to an SLAM loopback detection method and system based on three-dimensional point cloud height intensity density information. Dividing the laser point cloud into a plurality of three-dimensional grids, dividing each three-dimensional grid to obtain a plurality of sub-three-dimensional grids, and calculating height information, intensity information and density information of each sub-three-dimensional grid to obtain a descriptor of each three-dimensional grid; obtaining a global descriptor according to the descriptor composition; and performing calculation according to the sub-stereo network to obtain grid statistical characteristics, and performing loopback detection according to the global K-dimensional tree to obtain a loopback detection result. By fusing the multi-dimensional features, the loopback detection precision and robustness of the laser radar SLAM technology in a complex environment are significantly improved; the laser point cloud is divided into the multi-layer three-dimensional grids, and the descriptors are generated in combination with the height, strength and density information, so that the problem of insufficient descriptor discrimination caused by single dependence on height or strength characteristics in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of laser radar detection, and in particular to a SLAM loop detection method and system based on three-dimensional point cloud height intensity density information. Background Art

[0002] LiDAR SLAM (Simultaneous Localization and Mapping) technology is one of the core technologies of autonomous mobile robots. It uses LiDAR equipment to obtain three-dimensional point cloud data of the environment, then constructs a map of the surrounding environment in real time and simultaneously estimates the robot's precise position in that environment. This provides a basis for the robot's autonomous navigation in unknown environments, allowing the robot to complete tasks even without a preset map.

[0003] However, lidar SLAM technology also faces challenges in practical applications. Among them, the cumulative error of the lidar odometer is a problem. As the robot's movement time increases and the path becomes more complicated, the cumulative error of the odometer will gradually accumulate, resulting in deviations in the global consistency of the constructed map, which in turn affects the robot's positioning accuracy and navigation effect. In order to suppress this cumulative error, the existing technology proposes a loop detection method based on global descriptors. Through loop detection, the robot can identify whether it has reached a historical position again, thereby correcting the map and posture, and improving the accuracy of mapping and positioning.

[0004] The Chinese patent application publication number CN116679314A discloses a method and system for simultaneous mapping and positioning of three-dimensional laser radars that integrates point cloud intensity. This method fuses the cylindrical projection of point cloud intensity with the spatial density distribution to construct a global descriptor—the point cloud intensity projection shape feature—to improve the effectiveness of loop detection. However, the above method has the following shortcomings: the scanning frequency of the laser radar and the movement speed of the robot will affect the point cloud density, resulting in distortion of the calculated global descriptor, which in turn affects the accuracy of loop detection. Secondly, this method only focuses on the height information and external intensity features of the point cloud, while ignoring the internal geometric structure of the point cloud. In some scenarios, this method will lead to insufficient discrimination, making it easy to generate similar global descriptors in feature-degraded environments or complex indoor environments, thereby causing mismatches. Summary of the Invention

[0005] In response to the problems mentioned in the prior art, the present invention proposes a SLAM loop detection method and system based on the height, intensity and density information of three-dimensional point clouds. By obtaining the height, intensity and density information of three-dimensional point clouds, the loop detection accuracy and robustness of lidar SLAM technology in complex environments are improved.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention proposes a SLAM loop detection method based on three-dimensional point cloud height intensity density information, comprising the following steps: Obtaining a laser point cloud and dividing the laser point cloud into multiple three-dimensional grids; Divide each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; A global descriptor is obtained based on the descriptors of multiple three-dimensional grids; Calculating grid statistical features based on the sub-stereoscopic network, and inserting the grid statistical features and global descriptors into a pre-built global K-dimensional tree; Perform loop detection based on the global K-dimensional tree to obtain the loop detection result.

[0007] As a further improvement of the present invention, a laser point cloud is obtained and divided into a plurality of three-dimensional grids, including: With the laser radar as the center of the circle, let the maximum detection distance and maximum effective height of the laser radar be the scanning radius and height respectively. The scanning area is divided into cylindrical shapes according to the scanning radius and height: The scanning area is divided into a plurality of concentric ring cylinders along the radial direction, and each ring cylinder is evenly divided into a plurality of strip-shaped three-dimensional grids.

[0008] As a further improvement of the present invention, each 3D grid is divided into multiple sub-3D grids, and the height information, intensity information, and density information of each sub-3D grid are calculated respectively, including: The three-dimensional grid is divided into multiple layers of sub-three-dimensional grids of equal size along the vertical direction, and the laser points in each sub-three-dimensional grid are marked. The laser point set is obtained as ,in is the first A laser point, They are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the laser point respectively. is the reflection intensity of the laser point; The height information of each sub-grid includes: If the laser point set is not an empty set, the height information of the sub-3D grid is 1. If each sub-3D grid If the laser point set is an empty set, the height information of the sub-stereo grid is 0. The calculation formula is as follows:

[0009] Where: Indicates height information; The intensity information of each sub-grid includes calculating the average intensity of the laser points in each sub-grid , the calculation formula is as follows:

[0010] Where: Represents a sub-volume mesh The number of laser points in ; Calculate the average intensity of the laser points in all sub-grids at the same height and horizontal angle in the radial direction , the calculation formula is as follows:

[0011] Compare and , determine the intensity information of each sub-three-dimensional grid, the calculation formula is:

[0012] The density information of each sub-grid includes the following calculation formula:

[0013] Where, It represents the median number of point clouds of all sub-grids at different heights within the same radius and horizontal angle range.

[0014] As a further improvement of the present invention, a descriptor of each 3D grid is obtained based on the height information, intensity information, and density information of each sub-3D grid, including: The global information is calculated according to the following formula:

[0015] Where: is global information; For high information; is the intensity information; is the density information; The global information of all sub-3D meshes is summed to obtain the descriptor of the 3D mesh. The calculation formula is as follows:

[0016] Where: is the descriptor; is the number of layers in the vertical direction.

[0017] As a further improvement of the present invention, the descriptors of the plurality of three-dimensional grids obtained by calculation are combined into a global descriptor in the form of a two-dimensional matrix.

[0018] As a further improvement of the present invention, grid statistical features are calculated based on the sub-stereoscopic network, including: The sub-grids with the same radius in any layer are grouped into rings, and the laser points in each ring are counted, and the number of sub-grids with non-zero laser points is counted. , calculate all the The mean and standard deviation of are calculated as follows:

[0019]

[0020] Where: Indicates the number of concentric ring cylinders; represents the average value; represents the standard deviation; Concatenate the mean and standard deviation to obtain the sub-feature signature of this layer; Obtain the sub-feature signatures of all layers, and combine the sub-feature signatures of all layers to obtain the grid statistical features.

[0021] As a further improvement of the present invention, loop detection is performed based on the global K-dimensional tree to obtain loop detection results, including: Perform nearest neighbor search on the global K-dimensional tree to obtain historical grid statistical features similar to the current frame grid statistical features to form candidate historical frames; Calculate the two-dimensional Fourier transform of the global descriptor matrix of the current frame and the two-dimensional Fourier transform of the global descriptor matrix of the candidate historical frame respectively. Through cross power spectrum analysis and inverse Fourier transform, restore the global descriptor matrix of the candidate historical frame to a state similar to the current frame, calculate the normalized difference value, and record the historical frame with the smallest normalized difference value as the suspected loop frame; For the suspected loop frame and the current frame, compare the continuous ring distribution of the sub-three-dimensional grid with a non-zero number of laser points to determine whether the difference in the continuous ring distribution is less than a preset threshold. If so, it is a loop frame; otherwise, it is a non-loop frame.

[0022] In a second aspect, the present invention proposes a SLAM loop detection system based on three-dimensional point cloud height intensity density information, comprising: The first division module is used to obtain the laser point cloud and divide the laser point cloud into multiple three-dimensional grids; The second partitioning module is used to partition each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; The composition module obtains a global descriptor based on the descriptors of multiple three-dimensional grids; An insertion module calculates grid statistical features based on the sub-stereoscopic network and inserts the grid statistical features and the global descriptor into a pre-built global K-dimensional tree; The detection module is used to perform loop detection based on the global K-dimensional tree to obtain loop detection results.

[0023] In the third aspect, the present invention proposes a SLAM loop detection device based on three-dimensional point cloud height intensity density information, including a processor and a memory, wherein when the processor executes the computer program stored in the memory, it implements the SLAM loop detection method based on three-dimensional point cloud height intensity density information as mentioned above.

[0024] In a fourth aspect, the present invention proposes a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the SLAM loop detection method based on the height intensity density information of the three-dimensional point cloud is implemented as described above.

[0025] Compared with the prior art, the present invention has achieved the following technical effects: The present invention significantly improves the loop detection accuracy and robustness of LiDAR SLAM technology in complex environments by fusing multi-dimensional features; by dividing the laser point cloud into multi-layer three-dimensional grids and generating descriptors based on height, intensity and density information, the problem of insufficient descriptor discrimination caused by single reliance on height or intensity features in the existing technology is solved. Through the division of the three-dimensional grid, the local geometric structure of the environment can be effectively captured; the obtained global descriptor is combined with Fourier transform and cross power spectrum analysis to quickly match the global features of historical frames and current frames, significantly reducing the computational complexity; in addition, the introduction of the global K-dimensional tree can further improve the efficiency of candidate frame retrieval and greatly reduce the probability of mismatching.

[0026] The concentric ring cylinder division strategy proposed in the present invention divides the scanning area into multiple strip-shaped three-dimensional grids along the radial direction, which not only conforms to the cylindrical scanning characteristics of the lidar, but also retains the continuous spatial distribution information of the environment through radial stratification, making feature extraction more in line with the actual detection scenario; by generating global information through product fusion, multi-feature coupling is realized, further improving the expressive power of the descriptor; by introducing Fourier transform and continuous ring distribution verification, the present invention ensures the reliability of loop detection results and reduces the false detection rate in complex dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0029] It should be noted that traditional LiDAR SLAM loop detection has high detection accuracy in scenes with rich geometric features. However, because it only uses the height information of 3D point cloud data and ignores other geometric information, including the geometric structure within the point cloud and the intensity characteristics outside the point cloud, it can lack discrimination in some scenes. In feature-degraded or complex environments, such as long corridors, similar global descriptors may be generated, leading to mismatches.

[0030] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments: like Figure 1 As shown, the present invention proposes a SLAM loop detection method based on three-dimensional point cloud height intensity density information, comprising the following steps: Obtaining a laser point cloud and dividing the laser point cloud into multiple three-dimensional grids; Divide each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; A global descriptor is obtained based on the descriptors of multiple three-dimensional grids; Calculating grid statistical features based on the sub-stereoscopic network, and inserting the grid statistical features and global descriptors into a pre-built global K-dimensional tree; Perform loop detection based on the global K-dimensional tree to obtain the loop detection result.

[0031] Step 1: Before obtaining the laser point cloud, first build a global K-dimensional tree and initialize the root node of the global K-dimensional tree to an empty node.

[0032] With the laser radar as the center of the circle, the maximum detection distance and maximum effective height of the laser radar are Scan radius and height , divided into cylindrical scanning areas according to scanning radius and height; Divide the scan area into The ring cylinders are in the shape of concentric rings, and the first The circular cylinder is denoted as , divide each ring cylinder into The three-dimensional grid is strip-shaped, starting from the radius on the bottom surface of the cylindrical point cloud parallel to the direction of movement and moving in the counterclockwise direction, marking the The first The three-dimensional grid is recorded as .

[0033] Step 2: This embodiment will three-dimensional grid Divide it into 8 layers in the vertical direction, and get 8 sub-grids of the same size , mark the layer height from top to bottom , sub-stereo grid All laser points are represented as a set ,in Represents a sub-volume mesh The A laser point, They are the x-axis coordinate, y-axis coordinate, z-axis coordinate and reflection intensity of the laser point respectively.

[0034] For each sub-grid Calculating height information , if each sub-grid A collection of laser points If it is not an empty set, then the sub-three-dimensional grid Height information is 1, each sub-grid If the laser point set is an empty set, the height information of the sub-stereo grid is 0. The calculation formula is as follows:

[0035] For each sub-grid Computational intensity information : Calculate the average intensity of the laser points in each sub-grid , the calculation formula is:

[0036] Where, Represents a sub-volume mesh The set of all laser points in ; Represents a sub-volume mesh The number of laser points in .

[0037] Calculate the average intensity of the laser points in all sub-grids at the same height and horizontal angle in the radial direction , the calculation formula is:

[0038] Final comparison and Size, determines the intensity information of each sub-grid , the calculation formula is:

[0039] For each sub-grid Calculating density information , the calculation formula is as follows:

[0040] Where, It represents the median number of point clouds of all sub-grids at different heights within the same radius and horizontal angle range.

[0041] Calculate descriptors : Calculate each sub-three-dimensional mesh The global information of is calculated as follows:

[0042] Where: is global information; For high information; is the intensity information; is the density information; The information of all sub-grids at different heights Sum and get the descriptor , the calculation formula is:

[0043] All descriptors are composed OK 2D matrix of global descriptors of columns .

[0044] Step 3: Based on the sub-three-dimensional grid , will Same radius in layers All sub-grids form a ring , for each ring , count the number of sub-grids where the number of laser points is not 0 , calculate the Layer All Average value and standard deviation , the calculation formula is:

[0045]

[0046] Where: Indicates the number of concentric ring cylinders.

[0047] Concatenate the mean and standard deviation to get the concatenated value , the splicing value The sub-feature signature of this layer of point cloud is:

[0048] The sub-feature signatures of all layers constitute the grid statistical features of the scan area ,Right now:

[0049] The grid statistical features and global descriptors are inserted into a pre-initialized global K-dimensional tree.

[0050] Step 4: Perform a three-step search on the global K-dimensional tree.

[0051] The first step of the search is to perform a nearest neighbor search on the global K-dimensional tree to obtain the historical grid statistical features that are most similar to the grid statistical features corresponding to the current frame among all the grid statistical features recorded in the current K-dimensional tree. Based on these historical grid statistical features, the corresponding point cloud frames are obtained to form candidate historical frames.

[0052] The second search step is to calculate the global descriptor matrix of the current frame The two-dimensional Fourier transform of is calculated as follows:

[0053] Where: represents the row index in the frequency domain; represents the column index in the frequency domain; Represents the row index in the time domain; Represents the column index in the time domain; Represents the global descriptor matrix of the current frame; Represents the 2D Fourier transform result of the global descriptor matrix of the current frame.

[0054] Calculate the global descriptor matrix for each candidate history frame The two-dimensional Fourier transform of is calculated as follows:

[0055] Calculate the cross power spectrum of each candidate historical frame and the current frame. The calculation formula is:

[0056] Where: Represents the horizontal displacement of the global descriptor between the current frame and the candidate frame; Indicates the vertical displacement of the global descriptor between the current frame and the candidate frame.

[0057] By performing inverse Fourier transform on the cross power spectrum, we can obtain and , restore the global descriptor matrix of each candidate historical frame to the state most similar to the current frame. The calculation formula is:

[0058] Where: Represents the global descriptor after the candidate historical frame is restored; Calculate the normalized difference between the global descriptor restored from each candidate historical frame and the global descriptor of the current frame. The calculation formula is:

[0059] The historical frame with the smallest normalized difference value is recorded as the suspected loop frame.

[0060] The third step is to compare the suspected loop frame and the current frame, layer by layer and radius by radius, the continuous ring distribution of the sub-stereo grid with non-zero laser point number, and determine whether the difference in the continuous ring distribution is less than the preset threshold. If so, it is a loop frame, otherwise it is a non-loop frame.

[0061] Based on the same inventive concept, an embodiment of the present invention also provides a SLAM loop detection system based on three-dimensional point cloud height intensity density information. Since the principle of solving the problem by the SLAM loop detection system based on three-dimensional point cloud height intensity density information is similar to the aforementioned SLAM loop detection method based on three-dimensional point cloud height intensity density information, the implementation of the SLAM loop detection system based on three-dimensional point cloud height intensity density information can refer to the implementation of the SLAM loop detection method based on three-dimensional point cloud height intensity density information, and the repeated parts will not be repeated.

[0062] In specific implementation, the SLAM loop detection system based on three-dimensional point cloud height intensity density information provided by the embodiment of the present invention specifically includes: The first division module is used to obtain the laser point cloud and divide the laser point cloud into multiple three-dimensional grids; The second partitioning module is used to partition each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; The composition module obtains a global descriptor based on the descriptors of multiple three-dimensional grids; An insertion module calculates grid statistical features based on the sub-stereoscopic network and inserts the grid statistical features and the global descriptor into a pre-built global K-dimensional tree; The detection module is used to perform loop detection based on the global K-dimensional tree to obtain loop detection results.

[0063] Correspondingly, an embodiment of the present invention also provides a SLAM loop detection device based on three-dimensional point cloud height intensity density information, including a processor and a memory, wherein when the processor executes the computer program stored in the memory, it implements the SLAM loop detection method based on three-dimensional point cloud height intensity density information provided by the embodiment of the present invention.

[0064] For more specific details about the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be described again here.

[0065] Accordingly, an embodiment of the present invention also provides a computer-readable storage medium for storing a computer program, wherein, when the computer program is executed by a processor, the SLAM loop detection method based on three-dimensional point cloud height intensity density information provided in the embodiment of the present invention is implemented.

[0066] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. References to the same or similar portions of the various embodiments will be sufficient. The systems, devices, and storage media disclosed in the embodiments are described briefly because they correspond to the methods disclosed in the embodiments. For relevant details, refer to the method descriptions.

[0067] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0068] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0069] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0070] The above is a detailed introduction to the SLAM loop detection method, system, device and storage medium based on the height intensity density information of three-dimensional point clouds provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A SLAM loop detection method based on three-dimensional point cloud height intensity density information, characterized in that: The following steps are involved: Obtaining a laser point cloud and dividing the laser point cloud into multiple three-dimensional grids; Divide each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; A global descriptor is obtained based on the descriptors of multiple three-dimensional grids; Calculating grid statistical features based on the sub-stereoscopic network, and inserting the grid statistical features and global descriptors into a pre-built global K-dimensional tree; Perform loop detection based on the global K-dimensional tree to obtain the loop detection result.

2. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: Obtain a laser point cloud and divide it into multiple three-dimensional grids, including: With the laser radar as the center of the circle, let the maximum detection distance and maximum effective height of the laser radar be the scanning radius and height respectively. The scanning area is divided into cylindrical shapes according to the scanning radius and height: The scanning area is divided into a plurality of concentric ring cylinders along the radial direction, and each ring cylinder is evenly divided into a plurality of strip-shaped three-dimensional grids.

3. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: Each 3D grid is divided into multiple sub-3D grids, and the height information, intensity information, and density information of each sub-3D grid are calculated separately, including: The three-dimensional grid is divided into multiple layers of sub-three-dimensional grids of equal size along the vertical direction, and the laser points in each sub-three-dimensional grid are marked. The laser point set is obtained as ,in The first A laser point, They are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the laser point respectively. is the reflection intensity of the laser point; The height information of each sub-grid includes: If the laser point set is not an empty set, the height information of the sub-3D grid is 1. If each sub-3D grid If the laser point set is an empty set, the height information of the sub-stereo grid is 0. The calculation formula is as follows: Where: Indicates height information; The intensity information of each sub-grid includes calculating the average intensity of the laser points in each sub-grid , the calculation formula is as follows: Where: Represents a sub-cube mesh The number of laser points in ; Calculate the average intensity of the laser points in all sub-grids at the same height and horizontal angle in the radial direction , the calculation formula is as follows: Compare and , determine the intensity information of each sub-three-dimensional grid, the calculation formula is: The density information of each sub-grid includes the following calculation formula: Where, It represents the median number of point clouds of all sub-grids at different heights within the same radius and horizontal angle range.

4. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: According to the height information, intensity information and density information of each sub-3D grid, the descriptor of each 3D grid is obtained, including: The global information is calculated according to the following formula: Where: is global information; For high information; is the intensity information; is the density information; The global information of all sub-3D meshes is summed to obtain the descriptor of the 3D mesh. The calculation formula is as follows: Where: is the descriptor; is the number of layers in the vertical direction.

5. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: The descriptors of the calculated multiple three-dimensional grids are combined into a global descriptor in the form of a two-dimensional matrix.

6. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: The grid statistical features are calculated based on the sub-stereoscopic network, including: The sub-grids with the same radius in any layer are grouped into rings, and the laser points in each ring are counted, and the number of sub-grids with non-zero laser points is counted. , calculate all the The mean and standard deviation of are calculated as follows: Where: Indicates the number of concentric ring cylinders; represents the average value; represents the standard deviation; Concatenate the mean and standard deviation to obtain the sub-feature signature of this layer; Obtain the sub-feature signatures of all layers, and combine the sub-feature signatures of all layers to obtain the grid statistical features.

7. The SLAM loop detection method based on three-dimensional point cloud height intensity density information according to claim 1, characterized in that: Perform loop detection based on the global K-dimensional tree to obtain loop detection results, including: Perform nearest neighbor search on the global K-dimensional tree to obtain historical grid statistical features similar to the current frame grid statistical features to form candidate historical frames; Calculate the two-dimensional Fourier transform of the global descriptor matrix of the current frame and the two-dimensional Fourier transform of the global descriptor matrix of the candidate historical frame respectively. Through cross power spectrum analysis and inverse Fourier transform, restore the global descriptor matrix of the candidate historical frame to a state similar to the current frame, calculate the normalized difference value, and record the historical frame with the smallest normalized difference value as the suspected loop frame; For the suspected loop frame and the current frame, compare the continuous ring distribution of the sub-three-dimensional grid with a non-zero number of laser points to determine whether the difference in the continuous ring distribution is less than a preset threshold. If so, it is a loop frame; otherwise, it is a non-loop frame.

8. A SLAM loop detection system based on three-dimensional point cloud height intensity density information, characterized in that: include: The first division module is used to obtain the laser point cloud and divide the laser point cloud into multiple three-dimensional grids; The second partitioning module is used to partition each 3D grid into multiple sub-3D grids, calculate the height information, intensity information, and density information of each sub-3D grid, and obtain a descriptor for each 3D grid based on the height information, intensity information, and density information of each sub-3D grid; The composition module obtains a global descriptor based on the descriptors of multiple three-dimensional grids; An insertion module calculates grid statistical features based on the sub-stereoscopic network and inserts the grid statistical features and the global descriptor into a pre-built global K-dimensional tree; The detection module is used to perform loop detection based on the global K-dimensional tree to obtain loop detection results.

9. SLAM loop detection equipment based on three-dimensional point cloud height intensity density information, characterized in that: The method comprises a processor and a memory, wherein when the processor executes the computer program stored in the memory, the method realizes the SLAM loop detection method based on three-dimensional point cloud height intensity density information as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, it implements the SLAM loop detection method based on three-dimensional point cloud height intensity density information as described in any one of claims 1 to 7.

Images