CN-122015817-A - Laser radar-IMU tight coupling SLAM method and system for metal cavity

CN122015817ACN 122015817 ACN122015817 ACN 122015817ACN-122015817-A

Abstract

A laser radar-IMU tight coupling SLAM method and system for a metal cavity comprise the steps of obtaining point cloud data of a laser radar in the metal cavity, preprocessing the point cloud data based on an improved laser radar data processing algorithm to obtain preprocessed point cloud data, obtaining IMU data, carrying out double drift correction of fusion reflection intensity and IMU pre-integration on the IMU data and the preprocessed point cloud data to obtain corrected IMU pose, extracting and matching features based on deep learning on the preprocessed point cloud data to obtain an optimized feature matching set, obtaining pose observation values according to the optimized feature matching set, carrying out incremental local optimization and global map optimization on the corrected IMU pose and the pose observation values, and outputting the optimized global pose and a constructed map. The invention can improve the autonomous positioning and map construction capability of the robot in the metal cavity degradation scene.

Inventors

Lang yexing
SHI ZHENG
LIU JUNTONG
WANG GUANYU
HUANG FUCUN
DU YANQIANG
ZHU MINGJIANG
YU HAICHUAN
BAO RUI
LI SHUANG
LIU JIAXIN
DIAN SONGYI
Zhou Fanwen
WANG SHUAI
XIAO QUAN
LI QIANG
ZHAO ZIJIAN
Zhong Xuke

Assignees

国网辽宁省电力有限公司电力科学研究院
四川乐成电气科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (12)

1. A laser radar-IMU tight coupling SLAM method for a metal cavity, comprising the steps of: acquiring point cloud data of a laser radar in a metal cavity, and preprocessing the point cloud data based on an improved laser radar data processing algorithm to obtain preprocessed point cloud data; Acquiring IMU data, and performing double drift correction of fusion reflection intensity and IMU pre-integration on the IMU data and the preprocessed point cloud data to obtain corrected IMU pose; Extracting and matching the preprocessed point cloud data by adopting features based on deep learning to obtain an optimized feature matching set, and obtaining a pose observation value according to the optimized feature matching set; and performing incremental local optimization and global map optimization on the corrected IMU pose and pose observation values, and outputting the optimized global pose and the constructed map.
2. The laser radar-IMU tight coupling SLAM method for metal cavities of claim 1, Preprocessing point cloud data based on an improved laser radar data processing algorithm, and specifically comprises the following steps: performing self-adaptive dynamic threshold filtering on the reflection intensity in the point cloud data based on the dynamic reflectivity to obtain abnormal point cloud data; Judging whether the point cloud data acquired by the laser radar have multipath interference or not, if so, judging the point cloud data to be abnormal point cloud data; And eliminating abnormal point cloud data in the point cloud data to obtain preprocessed point cloud data.
3. The laser radar-IMU tight coupling SLAM method for metal cavities of claim 2, The self-adaptive dynamic threshold filtering is carried out on the reflection intensity in the point cloud data based on the dynamic reflectivity, and the method is specifically as follows: Calculating a signal intensity mean I mean and a standard deviation σ I within a local time window: Wherein N represents the total number of point clouds in a local time window, and I i represents the reflection intensity value of the ith laser point; construction of dynamic threshold from signal intensity mean I mean and standard deviation sigma I : Wherein, the And Is an empirical parameter; and marking the point cloud data points with the reflection intensity I raw meeting I raw >T dynamic in the point cloud data in the corresponding local time window as abnormal point cloud data.
4. The laser radar-IMU tight coupling SLAM method for metal cavities of claim 1, The dual drift correction specifically includes: acquiring state variable data of the IMU, and carrying out prediction division on the state variable data of the IMU to obtain an accumulated error model; obtaining reflection intensity according to the preprocessed point cloud data, and calculating a reflection intensity correction factor; And constructing a joint optimization objective function based on the laser radar reflection intensity, the reflection intensity correction factor and the accumulated error model to obtain the corrected IMU pose.
5. The laser radar-IMU tight coupling SLAM method for a metal cavity according to claim 4, wherein, Acquiring state variable data of the IMU, and carrying out prediction and division on the state variable data of the IMU to obtain an accumulated error model, wherein the accumulated error model is specifically as follows: the acquisition of the state variable data of the IMU comprises the acceleration a and the angular velocity of the IMU at the current moment Pre-integrating the state variable data of the IMU to obtain the displacement of the IMU data at the moment And rotating And based on acceleration a and angular velocity Calculating the speed ; Combined with displacement Speed and velocity of And rotating R to build an accumulated error model as follows: Wherein, the Is a cumulative error model.
6. The laser radar-IMU tight coupling SLAM method for a metal cavity according to claim 4, wherein, Extracting environmental characteristics and reflection intensity from the preprocessed point cloud data, and calculating a reflection intensity correction factor, wherein the method comprises the following steps of: Reflection intensity correction factor The formula of (2) is as follows: wherein I measured is the reflection intensity of the preprocessed point cloud data, The point cloud data after preprocessing corresponds to the point cloud reflection intensity known a priori.
7. The laser radar-IMU tight coupling SLAM method for a metal cavity according to claim 6, wherein, Constructing a joint optimization objective function based on the laser radar reflection intensity, the reflection intensity correction factor and the accumulated error model to obtain corrected IMU state variables, wherein the method specifically comprises the following steps of: Constructing a joint optimization objective function based on the accumulated error model and the reflection intensity correction factor: Wherein, the For the pose estimated from lidar data, For the pose estimated by the laser radar data, deltax is an IMU correction increment, W is a weight matrix, and gamma is a weight factor of the reflection intensity residual error; minimizing the joint optimization objective function, which is obtained at this time Is the corrected state variable; accumulating and compensating the corrected state variables to obtain corrected IMU pose: In the formula, Is that The IMU pose after the moment correction, Is that The state of the IMU at the moment, The delta is corrected for the IMU.
8. The laser radar-IMU tight coupling SLAM method for metal cavities of claim 1, The method comprises the steps of extracting and matching the preprocessed point cloud data by adopting features based on deep learning to obtain an optimized feature matching set, and obtaining a pose observation value according to the optimized feature matching set, and specifically comprises the following steps: performing format conversion on the preprocessed point cloud data, and performing feature extraction on the preprocessed point cloud data after format conversion based on a deep learning network to obtain a candidate matching feature pair set; Constructing a matching cost function, and repeatedly sampling on the candidate matching feature pair set through a random sampling consistency algorithm to obtain an optimized feature matching set after mismatching is removed; And performing secondary optimization on the optimized feature matching set with the mismatching removed through a matching cost function to obtain an optimized feature matching set as a pose observation value.
9. The laser radar-IMU tight coupling SLAM method for metal cavities of claim 1, The incremental local optimization and global map optimization are performed on the corrected IMU pose and the pose observed value, and specifically comprise the following steps: performing fusion processing on the corrected IMU pose and the pose observed value to obtain fusion data; incremental local optimization is carried out on the fusion data, and the current frame state and pose estimation are updated in real time; Constructing a global pose graph according to the current frame state and pose estimation, forming a key frame graph, and performing multi-scale global graph optimization on the key frame graph; And obtaining the global pose after global map optimization and the constructed map.
10. A laser radar-IMU tight coupling SLAM system for a metal cavity, configured to implement the laser radar-IMU tight coupling SLAM method for a metal cavity according to any one of claims 1-9, comprising a data acquisition module, a preprocessing module, a correction module, a feature extraction and matching module, and an optimization module; the data acquisition module is used for acquiring point cloud data and IMU data of the laser radar in the metal cavity; the preprocessing module is used for preprocessing the point cloud data based on an improved laser radar data processing algorithm to obtain preprocessed point cloud data; The correction module is used for carrying out double drift correction of fusion reflection intensity and IMU pre-integration on the IMU data and the preprocessed point cloud data to obtain corrected IMU pose; the feature extraction and matching module is used for extracting and matching the preprocessed point cloud data by deep learning-based features to obtain an optimized feature matching set, and obtaining a pose observation value according to the optimized feature matching set; and the optimization module performs incremental local optimization and global map optimization on the corrected IMU pose and the pose observed value, and outputs the optimized global pose and the constructed map.
11. A terminal comprises a processor and a storage medium, and is characterized in that: The storage medium is used for storing instructions; The processor being operative according to the instructions to perform the steps of the method according to any one of claims 1-9.
12. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1-9.

Description

Laser radar-IMU tight coupling SLAM method and system for metal cavity Technical Field The invention relates to the technical field of binocular vision, in particular to a laser radar-IMU tight coupling SLAM method and system for a metal cavity. Background With the rapid development of the fields of intelligent robots, unmanned driving, industrial automation, etc., SLAM (Simultaneous Localization AND MAPPING, real-time localization and mapping) technology is increasingly important in these applications. SLAM technology enables robots to perform self-localization and mapping simultaneously in an unknown environment, which is critical for autonomous navigation, path planning and environmental awareness. Currently, the research of SLAM technology can be roughly divided into three types of visual SLAM, laser radar SLAM and IMU (inertial measurement unit) auxiliary SLAM, wherein the laser radar-IMU tight coupling SLAM algorithm gradually becomes the focus of research due to strong environmental adaptability and good robustness. In SLAM systems, lidar is the dominant environmental-aware sensor because of its ability to provide accurate distance measurement information. By means of the laser radar, the SLAM system can obtain three-dimensional point cloud data about the environment, and can effectively identify and construct structures such as obstacles, channels and walls in the environment. The IMU can provide high-frequency attitude estimation for the robot by measuring the acceleration and the angular velocity of the robot in the three-dimensional space, and can effectively make up for the deficiency of the laser radar especially under the condition that the motion state is dynamic or visual information is limited. Although the existing laser radar-IMU tight coupling SLAM algorithm (such as LIO-SAM, fastLIO, LOAM and the like) has good effects in various application scenes, significant technical bottlenecks still exist in a metal cavity degradation environment with the characteristics of complex electromagnetic interference, strong metal reflection, sparse characteristics and the like. The specific problems are as follows: 1. The quality of laser radar point cloud data is reduced, interference is difficult to effectively inhibit, and problems of overexposure, distortion, multipath reflection and the like are easily generated on a laser radar echo signal due to the fact that a metal surface with high reflection is commonly existing in a metal cavity, so that a large number of false points or false structures are formed, and the characteristic extraction and registration process in SLAM is obviously interfered. The existing method mostly adopts a fixed threshold value or a filtering strategy, cannot dynamically adapt to complex interference characteristics, and therefore the reliability of the point cloud is reduced, and the graph construction precision is affected. 2. The IMU has obvious drift in a strong interference scene, and the existing compensation mechanism has insufficient robustness, namely zero bias fluctuation and error accumulation are easy to occur to the output of the IMU due to strong electromagnetic interference in a metal cavity environment. Although the existing SLAM system usually introduces IMU pre-integration or factor graph optimization for drift suppression, in a closed space which cannot rely on external information such as GPS, the algorithms are difficult to stably restrict drift for a long time, and finally the stability and accuracy of overall pose estimation are affected. 3. The matching is difficult in a sparse characteristic environment, the traditional extraction algorithm has poor adaptability, the environment structure of the metal cavity is regular, the surface is mostly smooth, and the obvious geometric characteristic is lacked, so that the algorithm for manually extracting the characteristic based on angular points, edges and the like is difficult to obtain enough registration constraint. The traditional feature matching algorithm is easy to have feature repetition or matching errors in the weak texture environment, and further aggravates pose drift risk. 4. The map optimization method has large calculation cost and difficult real-time performance, and in order to improve global consistency, the conventional SLAM system usually adopts the map optimization method to carry out global error correction. However, in the scene that the laser data is severely interfered and the environment is severely changed, the side weight calculation in the graph is easily influenced by abnormal data, meanwhile, the calculation complexity is high, the optimization convergence is difficult, and the deployment capability of the system in the industrial application with high dynamic and strong real-time performance is limited. In summary, the main shortcomings of the existing lidar-IMU tight coupling SLAM algorithm in the metal cavity degradation scenario can be summarized as follows: 1. The laser