Lightweight solid-state laser radar 3D reconstruction method and system

Abstract

The present invention discloses a lightweight solid-state laser radar three-dimensional reconstruction method and system, including: obtaining laser point cloud information and image information at the current moment, obtaining laser point cloud data with color information through visual projection; extracting its edge features and surface features; using a sliding window method to match feature points, and using edge features and plane features of adjacent frames to construct a local feature mapping map; based on the local feature mapping map, respectively calculating point-to-edge residuals and point-to-plane residuals; with the goal of minimizing point-to-edge residuals and point-to-plane residuals, constructing an odometer estimation model, solving the odometer estimation model, and obtaining a final odometer result; based on the final odometer result, combined with the corresponding point cloud projection, a reconstructed three-dimensional point cloud map is obtained. The present invention extracts edge features and surface features of laser point clouds, which can effectively reduce the computing power consumption of feature matching under dense point cloud input conditions.

Description

Technical Field

[0001] The present invention relates to the field of solid-state laser radar three-dimensional reconstruction technology, and in particular to a lightweight solid-state laser radar three-dimensional reconstruction method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Light detection and ranging (LiDAR) sensors play an important role in simultaneous localization and mapping (SLAM) and have become one of the most important perception devices. Existing SLAM methods mainly target mechanical LiDAR sensors that are often used by large robots. With the continuous development of the semiconductor industry, the cost and volume of high-power and high-beam quality radiation sources have continued to decline, and optical phased array technology has also provided a good technical foundation for the accuracy of radar. The miniaturization of solid-state LiDAR has been successful and is sought after by more and more researchers because it provides a cost-effective and lightweight solution for small robots. Compared with mechanical LiDAR, solid-state LiDAR sensors have higher update frequency and angular resolution, but also have a smaller field of view (FoV), which is very challenging for existing LiDAR SLAM algorithms. Therefore, it is necessary to provide a more robust and computationally efficient SLAM method for this new type of sensing device.

[0004] Existing methods are mainly used for mechanical LiDAR sensors, which collect surrounding information by rotating high-frequency laser arrays. Although they have achieved impressive experimental results on large scene maps, it is not widely used due to its high cost. At the same time, due to the size and weight of mechanical LiDAR, it is difficult to implement on small systems. For example, by carrying a mechanical LiDAR for building inspection, the flight durability of an unmanned aerial vehicle (UAV) is significantly reduced. In addition, due to the large size of mechanical LiDAR, it is also impossible to integrate it into a handheld device.

[0005] Although the performance of mechanical LiDAR and solid-state LiDAR is similar, the implementation or challenges of LiDAR SLAM are different. To illustrate the difference between the two LiDARs, we take Velodyne VLP-16 and Realsense L515 as examples, with specifications in Table 1. Solid-state LiDAR has higher angular resolution, which means a higher density of points in the same scanning area. Therefore, traditional LiDAR odometry methods such as Iterative Closest Point (ICP) may be computationally inefficient because more points need to be calculated. Secondly, solid-state LiDAR has a higher update frequency, and traditional LiDAR SLAM (such as LOAM) is not computationally efficient enough to achieve real-time performance. Another challenge is the pyramid-like coverage view, which may cause severe tracking loss during large rotations.

[0006] Table 1 Differences between mechanical and solid-state LiDAR

[0007] Summary of the invention

[0008] In order to solve the above problems, the present invention proposes a lightweight solid-state lidar 3D reconstruction method and system, which can provide accurate positioning and high-quality mapping through improved feature map extraction, mileage estimation and probability map construction.

[0009] In some embodiments, the following technical solutions are adopted:

[0010] A lightweight solid-state laser radar 3D reconstruction method, comprising:

[0011] Obtain the laser point cloud information and image information at the current moment, and obtain the laser point cloud data with color information through visual projection;

[0012] Based on the laser point cloud data with color information, extracting edge features and surface features thereof;

[0013] Based on the odometer information at the historical moment, a sliding window method is used to match feature points, and a local feature map is constructed using edge features and plane features of adjacent frames; based on the local feature map, point-to-edge residuals and point-to-plane residuals are calculated respectively;

[0014] With the goal of minimizing the point-to-edge residual and the point-to-plane residual, an odometer estimation model is constructed, and the odometer estimation model is solved to obtain a final odometer result;

[0015] Based on the final odometer result and in combination with the corresponding point cloud projection, a reconstructed three-dimensional point cloud map is obtained.

[0016] In other embodiments, the following technical solutions are adopted:

[0017] A lightweight solid-state laser radar 3D reconstruction system, comprising:

[0018] Point cloud data acquisition module, used to obtain laser point cloud information and image information at the current moment, and obtain laser point cloud data with color information through visual projection;

[0019] A feature extraction module, used to extract edge features and surface features based on the laser point cloud data with color information;

[0020] A feature matching module is used to match feature points based on the odometer information at historical moments by using a sliding window method, and to construct a local feature mapping map by using edge features and plane features of adjacent frames; based on the local feature mapping map, point-to-edge residuals and point-to-plane residuals are calculated respectively;

[0021] The odometer calculation module constructs an odometer estimation model with the goal of minimizing the point-to-edge residual and the point-to-plane residual, solves the odometer estimation model, and obtains the final odometer result;

[0022] The three-dimensional point cloud map reconstruction module is used to obtain a reconstructed three-dimensional point cloud map based on the final odometer result and the corresponding point cloud projection.

[0023] In other embodiments, the following technical solutions are adopted:

[0024] A terminal device includes a processor and a memory, wherein the processor is used to implement various instructions; the memory is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the above-mentioned lightweight solid-state laser radar three-dimensional reconstruction method.

[0025] In other embodiments, the following technical solutions are adopted:

[0026] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned lightweight solid-state laser radar three-dimensional reconstruction method.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] (1) The present invention extracts edge features and surface features of laser point clouds, which can effectively reduce the computing power consumption of feature matching under dense point cloud input conditions.

[0029] (2) The present invention combines odometer historical information and adopts a sliding window method for feature matching, which can limit the least squares residual to a certain limit and prevent the exponential increase in the amount of calculation caused by the increasing size of the information matrix.

[0030] (3) The present invention constructs an odometer estimation model based on point-to-plane residuals and point-to-edge residuals, thereby reducing the probability of odometer degradation / failure caused by feature point loss / mismatching due to fewer feature points or feature planes.

[0031] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of a lightweight solid-state laser radar 3D reconstruction method in an embodiment of the present invention;

[0033] Figure 2 The parameter performance evaluation of the method of this embodiment and the real value and A-LOAM is performed;

[0034] Figure 3 is a schematic diagram of positioning and mapping results in an embodiment of the present invention;

[0035] FIG4(a)-(b) are respectively the rotation test results of the method of this embodiment and the rotation test results using A-LOAM. DETAILED DESCRIPTION

[0036] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0038] Embodiment 1

[0039] This embodiment uses a solid-state laser radar sensor, which has a higher refresh rate and resolution than a traditional mechanical laser radar, so the amount of calculation increases accordingly and cannot accurately match the original point cloud.

[0040] Based on this, in one or more embodiments, a lightweight solid-state laser radar 3D reconstruction method is disclosed, combined with Figure 1 , specifically including the following process:

[0041] (1) Obtain the laser point cloud information and image information at the current moment, and obtain the laser point cloud data with color information through visual projection;

[0042] Specifically, the laser point cloud information at the current moment is obtained through a solid-state laser radar; the image information at the current moment is obtained through a camera, and then the image information is visually projected to finally obtain laser point cloud data with color information.

[0043] (2) extracting edge features and surface features based on the laser point cloud data with color information;

[0044] Since the point cloud data returned by the LiDAR is unordered, in order to calculate the edge and plane features, the point cloud is first projected into a two-dimensional point matrix. th Laser point input P k , split in vertical and horizontal directions respectively.

[0045] Given a point P i ={x i ,y i , z i}∈P k , vertical angle α i and the horizontal angle θ i The calculation formula is as follows:

[0046]

[0047]

[0048] The laser point cloud is respectively detected by the vertical detection range {α min , α max} and horizontal detection range {θ min ,θ max} is divided into M and N sectors. Among them, α min , α max ,θ min ,θ max They are: the minimum vertical angle, maximum vertical angle, minimum horizontal angle, and maximum horizontal angle specified by the sensor.

[0049] For a vertical resolution of α r The horizontal resolution is θ r Solid-state laser radar, after compromising the computing speed and data accuracy, select M and N as half of the total number of points, that is:

[0050]

[0051]

[0052] In each cell (m,n) (m∈[1,M], n∈[1,N]), the geometric center of the cell is calculated to measure In order to extract line features and plane features, search for points near them and define local smoothness:

[0053]

[0054] in, λ is the predefined search radius, a larger λ means more computing resources; is the geometric center of the block area of ​​size i, j.

[0055] Local smoothness Indicates the clarity of surrounding information. The larger it is, the greater the local curvature is;

[0056] when When it is greater than the set threshold, the point in the unit is used as an edge feature. When it is less than the set threshold, the points in the unit are regarded as surface features.

[0057] (3) Based on the odometer information at the historical moment, a sliding window method is used to match feature points, and a local feature map is constructed using edge features and plane features of adjacent frames; based on the local feature map, point-to-edge residuals and point-to-plane residuals are calculated respectively;

[0058] Since the traditional method is based on scan-to-scan matching, although the current frame can be quickly matched with the previous frame, compared with traditional mechanical radar, the scene field of view of a single solid-state laser scan is limited, which will cause large drift in larger scenes.

[0059] In this embodiment, the current posture of the robot in the global scene is defined as T k ∈SE(3), and record the historical information of the odometer as P1, P2, P3…P k-1 , using scan-to-map (matching lidar scan data directly with the map) matching method to improve performance.

[0060] In this embodiment, in order to reduce the amount of calculation, the sliding window method is used for feature point matching; in the SLAM process, as the number of key frames and landmarks increases, the calculation efficiency of the back-end optimization will continue to decline. In order to avoid this situation, the sliding window method is used to limit the key frames to be optimized to a certain number to control the scale of the back-end optimization. First, add new variables to enter the least squares system optimization. If the number of variables reaches a certain dimension, remove the old variables. Finally, the system continuously loops the previous two steps until the program runs to the end.

[0061] At the same time, the edge features and plane features of adjacent frames are used to construct a local feature map.

[0062] For the current P k Definition M k = {P k-1 ,P k-2 ,…,P k-q}Where q applies to the number of frames required to construct the local feature map.

[0063] As mentioned above, matching on the original point cloud is inefficient and sensitive to noise. Therefore, edge points and plane points in the feature space are used for matching.

[0064] For each edge point p k ∈P k In local map coordinates In , the nearest edge is searched from the local map. Since the local map is divided into edge local map and plane local map, each map is constructed by KD tree to improve the search efficiency.

[0065] Therefore, two nearest edge feature points can be selected for each edge point from the edge local map and Point-to-edge residual Defined as Intersection with edge and Distance between:

[0066]

[0067] Here, the symbol × is the cross product of two vectors.

[0068] For the processing of special features, if the number of nearby points is less than 2, the final cost does not consider the remainder from the point to the edge.

[0069] Similarly, for every plane point p k ∈P k , search for the nearest plane feature from the local map of the plane. To estimate a plane in 3D space, you must have 3 points. Therefore, for a given plane feature point p k and its coordinates on the local map The transformation in finds the three closest points from the local planar map and Point-to-plane residuals Defined as and through and The distance between the planes:

[0070]

[0071] Similar to the edge residuals, when the number of nearby points is less than 3, the point-to-plane residuals are not considered.

[0072] (4) constructing an odometer estimation model with the goal of minimizing the point-to-edge residual and the point-to-plane residual, solving the odometer estimation model, and obtaining the final odometer result;

[0073] In this embodiment, the final mileage is estimated by minimizing the point-to-plane residual and the point-to-edge residual:

[0074]

[0075] This nonlinear optimization problem can be solved using Gauss-Newton optimization. We use the left-perturbation formulation and apply increments over the Lie group. There are several advantages over the differential model in LOAM: (1) the rotations or attitudes are stored in a singularity-free format; (2) unconstrained optimization is performed at each iteration; and (3) operations occur at the matrix level, so there is no need to worry about taking derivatives of a set of scalar trigonometric functions.

[0076] Definition k =[ρ,φ]∈se(3) and the transformation matrix

[0077] It represents the conversion of 6D pose into a 4*4 matrix in the following way.

[0078]

[0079] in[·] × is the skew matrix of the 3D vector.

[0080] The left perturbation model can be calculated by the following formula:

[0081]

[0082] Among them, [T k p k ] is to convert the 4D point expression {x, y, z, 1} to a 3D point expression {x, y, z} before calculating the tilt matrix

[0083] The Jacobian matrix of the point-to-edge residual is defined as follows:

[0084]

[0085] in,

[0086]

[0087] The Jacobian matrix of the point-to-plane residual is defined as follows:

[0088]

[0089] However, due to the lack of historical information at the initial moment, the alignment between the current scan and the local map may not be ideal. Therefore, a better method needs to be found to assist feature matching. Since feature matching can be found through iteration, that is, the initial pose and based on The initial correspondence of . From this, the odometer can be derived for estimation Finally, it converges to the current optimal estimate. Although the iterative calculation is computationally inefficient, a good initial pose can speed up the convergence of the pose estimate. To find a better initial alignment, we assume that the angular velocity is constant and the translation is in a straight line:

[0090]

[0091] The process of iterative mileage estimation is listed in Algorithm 1:

[0092]

[0093]

[0094] (5) Based on the final odometer result and in combination with the corresponding point cloud projection, a reconstructed three-dimensional point cloud map is obtained.

[0095] Since the global map is much larger than the local map, and it cannot be updated based on each frame due to the amount of computation, this solution only uses key frames to update and rebuild the map. The key frames are selected based on the following criteria:

[0096] 1. The robot displacement is significant enough (greater than a predetermined threshold);

[0097] 2. The robot's own rotation angle is large enough (including pitch angle and heading angle);

[0098] 3. The robot spends more time at a certain place than a certain period of time;

[0099] In reality, the thresholds for rotation and translation are obtained from the sensor's FoV, and the minimum update frequency is based on the processor's computing power. Therefore, in order to improve the search efficiency, an octree is used to build a global map. Searching for a specific node from an octree with a depth of n only requires The computational complexity of , can significantly reduce the mapping cost. For each unit in the octree, we use P(n|z 1:t ) to represent the probability of the object existing:

[0100]

[0101] where z t is the measured value at the current moment, z 1:t-1 is the historical measurement value of the key frame, P(n) is the prior probability, and the prior probability of the unknown area is preset to 0.5.

[0102] To further illustrate the robustness of the proposed method, it is also integrated into a handheld device for 3D scanning. In our experiments, Intel Realsense L515 is used for demonstration. This is a small field-of-view solid-state LiDAR with a 70°×55° field of view and a 30 Hz update frequency. It is smaller and lighter than a smartphone, so it can be used on many mobile robotic platforms. The algorithm is written in C++ and implemented on Ubuntu 18.04 and ROS Melodic. In the first experiment, the proposed method was tested on a desktop PC with an Intel 6-core i5-9400 processor CPU. For the handheld device mapping and rotation tests, a dell micro computing platform with an i5-7200U processor was used.

[0103] ①Parameter performance evaluation

[0104] To evaluate the localization results, we compared the experimental results with the ground truth provided by the system. The robot was manually controlled and moved in a 4m × 4m environment. Figure 2 As shown, the average computation time is 31 milliseconds per frame. Our method achieves a translation error of 5 cm. We also compare our method with the LOAM algorithm, which is widely used in LiDAR SLAM. Due to the requirements of the LOAM algorithm, the horizontal sensor input angle is modified accordingly according to the sensor properties of the L515. The results of LOAM are drawn in orange. Compared with LOAM, the tracking effect will be reduced accordingly at higher rotation speeds, and tracking will be lost under long-term large-scale mapping, while our method can track accurately.

[0105] ② Actual mapping evaluation on handheld devices

[0106] To further demonstrate the robustness of this method, this embodiment also evaluates this method on a handheld device. With the development of virtual reality (VR), augmented reality (AR) and the gaming industry, SLAM has been implemented on various mobile devices such as smartphones and virtual reality glasses. However, the computing resources of most mobile platforms are limited. Compared with the stable movement of the experimental platform, handheld devices have vibrations and large viewing angle changes, which may lead to tracking loss and positioning failure.

[0107] 1) 3D mapping test: In the experiment, we used a 3D scanner to scan the indoor environment at a normal walking speed. The positioning and mapping results are shown in Figure 2. Figure 3 As shown, our method is able to accurately localize itself and perform mapping in real time.

[0108] 2) Rotation test: Handheld devices usually have higher rotation changes, which may cause tracking loss. To demonstrate the performance of the method under large rotation, we placed the solid-state lidar in the horizontal direction and randomly rotated the solid-state lidar at a maximum rotation speed of 1.57 rad / s. The solid-state lidar returned to the horizontal direction and recorded the angle deviation. When the final angle deviation was greater than 10 degrees, tracking loss was considered. We compared our method with A-LOAM. Our method was able to locate and perform mapping in 30 consecutive tests in different positions and environments. However, A-LOAM only successfully mapped 3 times. The results are shown in Table 2. The positioning and mapping results are shown in Figure 4(a) and (b).

[0109] Table 2: Rotation test results

[0110]

[0111] It can be seen that compared with A-LOAM, our method has a higher success rate and can maintain better positioning accuracy during unstable movement and has a higher success rate.

[0112] Embodiment 2

[0113] In one or more embodiments, a lightweight solid-state laser radar 3D reconstruction system is disclosed, comprising:

[0114] Point cloud data acquisition module, used to obtain laser point cloud information and image information at the current moment, and obtain laser point cloud data with color information through visual projection;

[0115] A feature extraction module, used to extract edge features and surface features based on the laser point cloud data with color information;

[0116] A feature matching module is used to match feature points based on the odometer information at historical moments by using a sliding window method, and to construct a local feature mapping map by using edge features and plane features of adjacent frames; based on the local feature mapping map, point-to-edge residuals and point-to-plane residuals are calculated respectively;

[0117] The odometer calculation module constructs an odometer estimation model with the goal of minimizing the point-to-edge residual and the point-to-plane residual, solves the odometer estimation model, and obtains the final odometer result;

[0118] The three-dimensional point cloud map reconstruction module is used to obtain a reconstructed three-dimensional point cloud map based on the final odometer result and the corresponding point cloud projection.

[0119] The specific implementation methods of the above modules have been described in Example 1 and will not be described in detail here.

[0120] Embodiment 3

[0121] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the lightweight solid-state laser radar 3D reconstruction method in Embodiment 1 when executing the program. For the sake of brevity, it will not be described in detail here.

[0122] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0123] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0124] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0125] Embodiment 4

[0126] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the lightweight solid-state lidar three-dimensional reconstruction method described in Example 1.

[0127] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A lightweight solid-state laser radar 3D reconstruction method, characterized in that: include: Obtain the laser point cloud information and image information at the current moment, and obtain the laser point cloud data with color information through visual projection; Based on the laser point cloud data with color information, extracting edge features and surface features thereof; Based on the odometer information at the historical moment, a sliding window method is used to match feature points, and a local feature map is constructed using edge features and plane features of adjacent frames; based on the local feature map, point-to-edge residuals and point-to-plane residuals are calculated respectively; With the goal of minimizing the point-to-edge residual and the point-to-plane residual, an odometer estimation model is constructed, and the odometer estimation model is solved to obtain a final odometer result; Based on the final odometer result, combined with the corresponding point cloud projection, a reconstructed three-dimensional point cloud map is obtained; Wherein, based on the laser point cloud data with color information, edge features and surface features are extracted, specifically including: The laser point cloud data is divided into M and N sectors according to the vertical detection range and the horizontal detection range; Calculate the geometric center of the cell (m,n) m∈[1,M], n∈[1,N]; Based on the geometric center, calculate the smoothness corresponding to the unit (m, n); If the smoothness is greater than a set threshold, the point in the unit is an edge feature; if the smoothness is less than the set threshold, the point in the unit is a surface feature; Based on the local feature map, the point-to-edge residual is calculated, specifically: Select the two nearest edge feature points for each edge point from the edge local map and Point-to-edge residual Defined as edge points With edge feature points and The distance between Based on the local feature map, the point-to-surface residual is calculated, specifically: For a given plane feature point p k , whose coordinates in the local plane map are Find the three closest points from a planar local map and Point-to-plane residual Defined as and through and The distance between the planes; With the goal of minimizing the point-to-edge residual and the point-to-plane residual, an odometer estimation model is constructed, specifically: Among them, T k is the current position of the robot, is the point-to-edge residual, is the point-to-plane residual.

2. A lightweight solid-state laser radar 3D reconstruction method as claimed in claim 1, characterized in that: The sliding window method is used for feature point matching, which includes: Add a set number of new variables to enter the least squares system optimization. If the number of variables reaches a certain dimension, remove the set number of old variables; repeat this process until the program ends.

3. A lightweight solid-state laser radar 3D reconstruction method as claimed in claim 1, characterized in that: The local feature map is constructed using the edge features and plane features of adjacent frames, including: Constructing a local edge map based on edge features in the current frame and a set number of historical frames; Constructs a local planar map based on planar features in the current frame and a set number of historical frames.

4. A lightweight solid-state laser radar 3D reconstruction system, characterized in that: include: Point cloud data acquisition module, used to obtain laser point cloud information and image information at the current moment, and obtain laser point cloud data with color information through visual projection; A feature extraction module, used to extract edge features and surface features based on the laser point cloud data with color information; A feature matching module is used to match feature points based on the odometer information at historical moments by using a sliding window method, and to construct a local feature mapping map by using edge features and plane features of adjacent frames; based on the local feature mapping map, point-to-edge residuals and point-to-plane residuals are calculated respectively; The odometer calculation module constructs an odometer estimation model with the goal of minimizing the point-to-edge residual and the point-to-plane residual, solves the odometer estimation model, and obtains the final odometer result; A three-dimensional point cloud map reconstruction module, used to obtain a reconstructed three-dimensional point cloud map based on the final odometer result and the corresponding point cloud projection; Wherein, based on the laser point cloud data with color information, edge features and surface features are extracted, specifically including: The laser point cloud data is divided into M and N sectors according to the vertical detection range and the horizontal detection range; Calculate the geometric center of the cell (m,n) m∈[1,M], n∈[1,N]; Based on the geometric center, calculate the smoothness corresponding to the unit (m, n); If the smoothness is greater than a set threshold, the point in the unit is an edge feature; if the smoothness is less than the set threshold, the point in the unit is a surface feature; Based on the local feature map, the point-to-edge residual is calculated, specifically: Select the two nearest edge feature points for each edge point from the edge local map and Point-to-edge residual Defined as edge points With edge feature points and The distance between Based on the local feature map, the point-to-surface residual is calculated, specifically: For a given plane feature point p k , whose coordinates in the local plane map are Find the three closest points from a planar local map and Point-to-plane residual Defined as and through and The distance between the planes; With the goal of minimizing the point-to-edge residual and the point-to-plane residual, an odometer estimation model is constructed, specifically: Among them, T k is the current position of the robot, is the point-to-edge residual, is the point-to-plane residual.

5. A terminal device, comprising a processor and a memory, wherein the processor is used to implement each instruction; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the lightweight solid-state lidar three-dimensional reconstruction method described in any one of claims 1-3.

6. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the lightweight solid-state lidar three-dimensional reconstruction method described in any one of claims 1-3.

Images