Underground space detecting and positioning method combining laser information and gas concentration information

Abstract

The invention discloses a laser information and gas concentration information combined underground space detection and positioning method, and belongs to the technical field of synchronous positioning and mapping. The method aims at solving the problems that an existing SLAM system is remarkable in accumulative error in an underground closed space, and map consistency is difficult to guarantee. Comprising the following steps: providing an initial value and direction constraint of physical field guidance for laser inter-frame matching by estimating a gas concentration space gradient, and improving tracking robustness in a feature degradation scene; in the local mapping stage, a geometric map and a parameterized gas concentration field model are constructed and jointly optimized, the anchoring attitude is constrained by using concentration consistency, and local drift is inhibited; in a loop detection link, a bimodal descriptor fusing geometry and gas distribution characteristics is generated, high-discrimination scene identification is realized, and geometric similar regions are effectively distinguished. The method can be used for reliable navigation and mapping without satellite signals.

Description

Technical Field

[0001] This invention relates to a method for underground space detection and positioning that combines laser information and gas concentration information, belonging to the field of synchronous positioning and mapping technology. Background Technology

[0002] With the development of underground engineering and emergency rescue, there is an urgent need for high-precision positioning and mapping in enclosed underground spaces. In such environments, satellite signals are completely ineffective, and the structures are repetitive and lighting is insufficient, leading to significant cumulative errors in SLAM systems that rely on lasers or vision. Figure 1 Consistency is difficult to guarantee. Existing methods mostly improve this by fusing inertial or geomagnetic data, but geomagnetic data is susceptible to interference, and inertial data is prone to drift, resulting in insufficient long-term stability. Summary of the Invention

[0003] In view of the significant cumulative error of existing SLAM systems in underground enclosed spaces, Figure 1 To address the issue of inconsistent performance, this invention provides a method for underground space detection and positioning that combines laser information with gas concentration information.

[0004] The present invention provides a method for underground space detection and positioning that combines laser information and gas concentration information, comprising:

[0005] The lidar and gas sensors mounted on the robot were calibrated in a spatiotemporal synchronization manner; at time t, 3D point cloud data of the underground space were collected using lidar. The feature point set at time t is obtained by processing the data. The gas concentration at time t is obtained using a gas sensor. Simultaneously, the initial pose at time t is obtained using an inertial sensor. , In the formula The initial position is in three-dimensional coordinates. The attitude includes pitch, roll, and yaw angles;

[0006] Using a sliding window to select including and The most recent t sets of data, i=1,2,3,...,t-1, The pose correction value at time i; the observed three-dimensional gradient vector of gas concentration at time t is obtained by fitting using the least squares method. ;

[0007] Based on the pose transition matrix at time t Observed values ​​of the three-dimensional gradient vector of gas concentration at time t Calculate the predicted gas concentration change at time t relative to time t-1; establish a gas gradient error term based on the observed and predicted gas concentration changes at time t relative to time t-1; weight and fuse the traditional geometric error term with the gas gradient error term; construct an inter-frame matching pose transfer matrix optimization function with the goal of minimizing the weighted fusion result; and solve the pose transfer matrix at time t using the Gauss-Newton method. ;

[0008] Based on the pose transition matrix at time t , initial pose Geometric feature matching is performed on the map obtained at time t-1, and the geometric feature matching error is calculated. An objective function is established with the goal of minimizing the weighted sum of the geometric feature matching error and the gas concentration error, and the initial pose at time t is obtained by solving the function. pose correction value at time t ;

[0009] Based on pose correction value at time t The feature point set at time t Feature point matching and point cloud fusion are performed with the map obtained at time t-1 before to obtain the local map at time t before;

[0010] In the historical exploration data, 3D point cloud data with geometrically significant features are selected as keyframes to establish a keyframe set. The 3D point cloud data at time t is used as the current frame. Loop closure detection is performed based on the keyframes and the current frame. After confirming the loop closure, the local map at time t is optimized for global consistency to obtain the optimized map.

[0011] The underground space detection and positioning method based on the combination of laser information and gas concentration information of the present invention obtains the observed value of the three-dimensional gradient vector of gas concentration at time t. The method is as follows:

[0012] Construct a system of linear equations:

[0013] ,

[0014] In the formula for The pose difference matrix, where the i-th row is the initial position difference between adjacent time steps. , for The concentration difference matrix, with the i-th row representing the observed gas concentration difference. ;

[0015] The three-dimensional gradient vector of gas concentration at time t is obtained by solving the problem. :

[0016] .

[0017] According to the underground space detection and positioning method combining laser information and gas concentration information of the present invention, the predicted value of the gas concentration change at time t relative to time t-1 is:

[0018] ,

[0019] Pose transition matrix It is in the form of a 4×4 homogeneous transformation matrix. To be Initial pose of dimension Convert to Transformation functions in 3D matrix form, To be dimension Convert to The conversion function; As a translation vector, it constitutes the translation part. Transformed into a rotation matrix via Rodrigues transformation Combined in standard form according to homogeneous transformation matrix Dimensional matrix ;

[0020] The corresponding observed change in gas concentration at time t relative to time t-1 is Establish the gas gradient error term for:

[0021] .

[0022] According to the underground space detection and positioning method combining laser information and gas concentration information of the present invention, the pose transfer matrix optimization function is constructed as follows:

[0023] ,

[0024] In the formula For traditional geometric error terms, The gas gradient error weights;

[0025] ,

[0026] In the formula, N represents the number of 3D point cloud matches. Let n be the matching position of the nth 3D point cloud at time t;

[0027] Solve the pose transition matrix optimization function to obtain the pose transition matrix at time t. .

[0028] According to the underground space detection and positioning method of the present invention, which combines laser information and gas concentration information, the gas concentration error is the observed gas concentration value at time t. The difference between the predicted gas concentration at time t and the predicted gas concentration at time t; the predicted gas concentration at time t is expressed as :

[0029] ,

[0030] In the formula These are the three-dimensional coordinate correction values ​​for the position.

[0031] The underground space detection and positioning method of the present invention, which combines laser information and gas concentration information, establishes an objective function with the goal of minimizing the weighted sum of geometric feature matching error and gas concentration error as follows:

[0032] ,

[0033] In the formula, K is the geometric feature matching number. Let be the matching error of the k-th geometric feature, and be the matching position of the k-th 3D point cloud at time t. The distance between the location and the matching location in the local map; Weights are matched for geometric features. This is the gas concentration error weight;

[0034] The objective function is solved using the LM nonlinear least squares algorithm, and the optimal initial pose at time t obtained by iterative solution is then used. As the pose correction value at time t .

[0035] The underground space detection and positioning method based on the combination of laser information and gas concentration information according to the present invention includes loop closure detection, which involves calculating the geometric similarity between the feature point sets of key frames and the current frame. Gas similarity to gas concentration observations corresponding to keyframes and the current frame .

[0036] According to the underground space detection and positioning method combining laser information and gas concentration information of the present invention, the sequence of nearest neighbor gas concentration observation values ​​corresponding to key frames is represented as a feature vector. The sequence of nearest-neighbor gas concentration observations corresponding to the current frame is represented as a feature vector. Gas similarity is calculated using cosine similarity. :

[0037] .

[0038] The underground space detection and positioning method of the present invention, which combines laser information and gas concentration information, further includes calculating loop closure candidate scores for loop closure detection. :

[0039] ,

[0040] In the formula Geometric similarity weights;

[0041] If the loop candidate score If the value exceeds the set loopback threshold, continue with geometric verification of the keyframe and the current frame, ICP matching error verification, and overlap verification. After passing all verifications, the loopback is confirmed.

[0042] The underground space detection and positioning method of the present invention, which combines laser information and gas concentration information, includes a method for optimizing the global consistency of the local map at time t, which includes nonlinear least squares method or graph optimization method.

[0043] The beneficial effects of this invention are as follows: This invention utilizes the physical characteristic that various gases can form relatively stable and continuous spatial concentration gradient fields in underground spaces, providing unique "fingerprint" information for environmental identification. This invention deeply integrates real-time acquired gas concentration gradient information into the laser SLAM process, using gas distribution characteristics to enhance underground SLAM. It fully leverages the inherent properties of the underground environment, providing a novel technical approach for reliable navigation and mapping in the absence of satellite signals.

[0044] This invention provides a method for laser SLAM enhancement in enclosed underground spaces where satellite signals are denied. Leveraging the relatively stable gas distribution in underground spaces, gas concentration information is used as a novel environmental fingerprint and deeply integrated into the entire laser SLAM chain. This addresses the challenges of cumulative drift and loop closure false detections that traditional geometric SLAM easily encounters in environments with repetitive features and similar structures. At the front end, by estimating the spatial gradient of gas concentration, initial values ​​and directional constraints guided by the physical field are provided for inter-frame laser matching, improving tracking robustness in feature degradation scenarios. In the local mapping stage, a geometric map and a parameterized gas concentration field model are constructed and jointly optimized. Concentration consistency constraints are used to anchor pose and suppress local drift. In the loop closure detection stage, a bimodal descriptor fusing geometric and gas distribution features is generated to achieve highly discriminative scene recognition and effectively distinguish geometrically similar regions. Finally, a globally consistent multimodal environment model tightly coupled with the geometric structure and gas concentration distribution is output. This invention fully utilizes the inherent physical properties of the underground environment, comprehensively enhancing the accuracy, robustness, and scene recognition capabilities of the SLAM system from the perception, estimation, and decision-making levels, providing a more reliable autonomous navigation solution for underground exploration, emergency rescue, and facility operation and maintenance. Attached Figure Description

[0045] Figure 1 This is a flowchart of the underground space detection and positioning method that combines laser information and gas concentration information as described in this invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Specific Implementation Method 1: Combination Figure 1 As shown, this invention provides a method for underground space detection and positioning that combines laser information and gas concentration information, including:

[0048] The lidar and gas sensors mounted on the robot were calibrated in a spatiotemporal synchronization manner; at time t, 3D point cloud data of the underground space were collected using lidar. The feature point set at time t is obtained by processing the data. The gas concentration at time t is obtained using a gas sensor. Simultaneously, the initial pose at time t is obtained using an inertial sensor. , In the formula The initial position is in three-dimensional coordinates. The attitude includes pitch, roll, and yaw angles;

[0049] Using a sliding window to select including and The most recent t sets of data, i=1,2,3,...,t-1, The pose correction value at time i; the observed three-dimensional gradient vector of gas concentration at time t is obtained by fitting using the least squares method. ;

[0050] Based on the pose transition matrix at time t Observed values ​​of the three-dimensional gradient vector of gas concentration at time t Calculate the predicted gas concentration change at time t relative to time t-1; establish a gas gradient error term based on the observed and predicted gas concentration changes at time t relative to time t-1; weight and fuse the traditional geometric error term with the gas gradient error term; construct an inter-frame matching pose transfer matrix optimization function with the goal of minimizing the weighted fusion result; and solve the pose transfer matrix at time t using the Gauss-Newton method. ;

[0051] Based on the pose transition matrix at time t , initial pose Geometric feature matching is performed on the map obtained at time t-1, and the geometric feature matching error is calculated. An objective function is established with the goal of minimizing the weighted sum of the geometric feature matching error and the gas concentration error, and the initial pose at time t is obtained by solving the function. pose correction value at time t ;

[0052] Based on pose correction value at time t The feature point set at time t Feature point matching and point cloud fusion are performed with the map obtained at time t-1 before to obtain the local map at time t before;

[0053] In the historical exploration data, 3D point cloud data with geometrically significant features are selected as keyframes to establish a keyframe set. The 3D point cloud data at time t is used as the current frame. Loop closure detection is performed based on the keyframes and the current frame. After confirming the loop closure, the local map at time t is optimized for global consistency to obtain the optimized map.

[0054] In this implementation, gas sensing is used as a novel modality throughout the gas-sensing-enhanced laser SLAM process. In the front-end odometry stage, the spatial gradient of gas concentration is estimated to provide high-quality initial values ​​for inter-frame point cloud matching guided by the physical field. Gradient direction consistency constraints are added to the optimization objective, significantly improving matching robustness under feature degradation environments. In the local mapping stage, a parameterized local gas concentration field model is constructed, and gas observation consistency is jointly optimized with geometric constraints as a nonlinear optimization term to achieve physical anchoring and correction of accumulated drift. In the loop closure detection stage, gas distribution fingerprints and geometric descriptors are fused to form a dual-modal discrimination criterion, solving the misidentification problem caused by repetitive underground structures. Finally, the system outputs a multimodal environment model that fuses geometric structure and gas concentration distribution. By deeply embedding stable gas physical field information into the entire SLAM perception, estimation, and decision-making chain, a systematic improvement in positioning and mapping accuracy and robustness is achieved in enclosed underground spaces where satellite signals are denied.

[0055] 3D point cloud data Denoising and feature extraction (such as planar and edge features) are performed to obtain a feature point set. Simultaneously, the gas concentration sequence is filtered to eliminate high-frequency fluctuations caused by sensor noise.

[0056] This implementation integrates gas concentration information into the traditional laser SLAM process, including sensor data processing, inter-frame matching, local map maintenance, loop closure detection, and global optimization.

[0057] First, sensor data processing is performed, including the feature point set. It is composed of extracted edge feature points and planar feature points. Feature point set The extraction process includes: estimating the normal vector and calculating the curvature of the input point cloud; setting a curvature threshold based on the calculated curvature values ​​to filter out edge points; and simultaneously extracting planar feature points from the point cloud using the random sample consistency method. The initial pose of the acquired data is also included. Cumulative error with inertial sensors.

[0058] Then, inter-frame matching is performed, which involves matching the laser point cloud data and concentration gradient data at time t-1 with the data at time t to preliminarily determine the pose transfer from time t-1 to time t.

[0059] Inter-frame matching and coarse pose optimization based on gas gradients can be combined with classic inter-frame matching algorithms such as ICP. Introducing the gas gradient as a constraint in the initial estimation reduces the probability of mismatches. Calculation of adjacent time steps... and Concentration change values ​​between And combined with the approximate displacement of the sensor , Let be the three-dimensional vector of the pose at time t. The concentration change mainly originates from spatial changes, and the concentration change is approximately represented as: . The gas concentration gradient is used. A gradient calculation unit based on a sliding window is employed to estimate it. .Will As an error term for verifying frame matching, the optimal pose transformation matrix is ​​obtained to improve matching accuracy.

[0060] Furthermore, the observed three-dimensional gradient vector of gas concentration at time t was obtained. The method is as follows:

[0061] Construct a system of linear equations:

[0062] ,

[0063] In the formula for The pose difference matrix, where the i-th row is the initial position difference between adjacent time steps. , for The concentration difference matrix, with the i-th row representing the observed gas concentration difference. ;

[0064] The three-dimensional gradient vector of gas concentration at time t is obtained by solving the problem. :

[0065] .

[0066] The predicted change in gas concentration at time t relative to time t-1 is:

[0067] ,

[0068] Pose transition matrix It is in the form of a standard 4×4 homogeneous transformation matrix. To be Initial pose of dimension Convert to Transformation functions in 3D matrix form, To be dimension Convert to The conversion function; As a translation vector, it constitutes the translation part. Transformed into a rotation matrix via Rodrigues transformation Combined in standard form according to homogeneous transformation matrix Dimensional matrix ; To Fill in the dimensions to satisfy the matrix multiplication condition.

[0069] The corresponding observed change in gas concentration at time t relative to time t-1 is Establish the gas gradient error term for:

[0070] .

[0071] The pose transition matrix optimization function is constructed as follows:

[0072] ,

[0073] In the formula For traditional geometric error terms, The gas gradient error weights;

[0074] , indicating the geometric registration error.

[0075] In the formula, N represents the number of 3D point cloud matches. Let n be the matching position of the nth 3D point cloud at time t; that is and This refers to the position of the nth matching point in the point cloud at time t-1 and the point cloud at time t after using point cloud matching algorithms such as the iterative nearest point algorithm. To Perform dimension filling and state transition matrix Dimension matching;

[0076] Solve the pose transition matrix optimization function to obtain the pose transition matrix at time t. .

[0077] Then, local map matching is performed to obtain the precise pose at time t.

[0078] Current frame features The pose is then finely matched with a local point cloud map constructed from recent frames, and further optimized through nonlinear optimization. At this point, the optimization backend not only includes the reprojection error constraint of laser feature points, but also introduces the observation model constraint of gas concentration.

[0079] The gas concentration error is the observed gas concentration value at time t. The difference between the predicted gas concentration at time t and the predicted gas concentration at time t; the predicted gas concentration at time t is expressed as :

[0080] ,

[0081] In the formula These are the correction values ​​for the three-dimensional coordinates of the position.

[0082] Furthermore, the objective function is established with the goal of minimizing the weighted sum of geometric feature matching error and gas concentration error as follows:

[0083] ,

[0084] In the formula, K is the geometric feature matching number. Let be the matching error of the k-th geometric feature, and be the matching position of the k-th 3D point cloud at time t. The distance between the location and the matching location in the local map; Weights are matched for geometric features. For example, for planar features, the gas concentration error weights are used. Defined as the distance from a point to its corresponding plane in the local map; for edge features, This is the distance from the point to the corresponding edge line in the local map. , This is used to characterize the uncertainty in geometric matching and gas concentration matching. The matching location is searched in the local map using a nearest neighbor association method. If the feature is a planar feature, the error... The vertical distance from the point to the matching plane; if the feature is an edge feature, the error is... The perpendicular distance from the point to the matching line.

[0085] As the initial value for the iterative solution, the objective function is solved using the LM nonlinear least squares algorithm, and the optimal initial pose at time t obtained from the iterative solution is then used. As the pose correction value at time t .

[0086] After obtaining the optimal pose, the coordinates of the point cloud in the feature point set at time t are determined globally. The point cloud in the local map is merged with the local map point cloud using nearest neighbor correlation. If a very close neighbor is found, the position of that map point is updated (e.g., by averaging). If no neighbor is found, the point is inserted into the map as a new map point, forming a new local map.

[0087] Then perform loop closure detection:

[0088] In this embodiment, loop closure detection includes calculating the geometric similarity between the feature point sets of the keyframe and the current frame. Gas similarity to gas concentration observations corresponding to keyframes and the current frame .

[0089] Maintain a set of keyframes in the history. Keyframes are selected based on the movement distance being higher than a threshold, the inter-frame overlap being lower than a threshold, or the gas concentration change being higher than a threshold.

[0090] The sequence of nearest-neighbor gas concentration observations corresponding to the keyframe is represented as a feature vector. The sequence of nearest-neighbor gas concentration observations corresponding to the current frame is represented as a feature vector. Gas similarity is calculated using cosine similarity. :

[0091] .

[0092] For keyframes and the current frame, a feature vector is formed by extracting the gas concentration observation sequence of the surrounding local area. The feature vector is a local gas data vector formed by continuously recording the gas concentration of several consecutive frames synchronously collected before the keyframe or the current frame. For example, a vector is formed by five gas concentration data collected between the keyframe and the previous four frames.

[0093] Geometric similarity Geometric descriptive feature vectors are obtained through classic methods such as fast point feature histograms or Scan Context, and the similarity of their geometric descriptive feature vectors is calculated using cosine similarity.

[0094] Loop closure detection also includes calculating loop closure candidate scores. :

[0095] ,

[0096] Final loop candidate score Geometric similarity Similarity to gases Jointly decided, in the formula Geometric similarity weights;

[0097] If the loop candidate score If the value exceeds the set loopback threshold, continue with geometric verification of the keyframe and the current frame, ICP matching error verification, and overlap verification. After passing all verifications, the loopback is confirmed.

[0098] Loop closure candidate scores are calculated based on the similarity. A high score indicates a high probability of loop closure, but does not guarantee a loop closure; further verification is required.

[0099] When the similarity between two poses is higher than the threshold, geometric verification is performed to verify the ICP matching error and overlap of the two frames of laser point clouds. When both conditions meet the threshold, loop closure is confirmed.

[0100] After confirming the loop closure, the system adopts existing standard frameworks based on pose graph optimization (such as g2o, GTSAM, or CeresSolver) to perform global consistency optimization on the local map at time t using nonlinear least squares or graph optimization methods, thereby correcting the global pose and point cloud position.

[0101] Global pose graph optimization employs either the classic Levenberg-Marquardt (LM) or Gauss-Newton algorithm. This results in a globally consistent map. .in For geometric point cloud maps, The gas concentration distribution field is fused.

[0102] Online updating and representation of gas concentration field maps:

[0103] As mapping progresses, the gas concentration field map will be continuously updated. Using Gaussian processes or lighter spatial interpolation grid models, discrete gas concentration observations can be... The data is fused into a continuous or semi-continuous spatial concentration distribution estimate. This map can be used not only for subsequent navigation, but its gradient information can also serve as prior information for subsequent positioning stages, forming a perception loop.

[0104] This invention aims to address the problem of improving positioning accuracy in underground spaces where absolute location information is unavailable. By leveraging the limited air circulation in underground spaces and the relatively stable and continuous spatial concentration gradient fields formed by various gases, this invention proposes an underground space detection and positioning method that combines laser information with gas concentration information.

[0105] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for underground space detection and positioning combining laser information and gas concentration information, characterized in that... include, Spatiotemporal synchronization calibration of the lidar and gas sensors mounted on the robot; 3D point cloud data of the underground space was collected using lidar at time t. The feature point set at time t is obtained by processing the data. The gas concentration at time t is obtained using a gas sensor. Simultaneously, the initial pose at time t is obtained using an inertial sensor. , In the formula The initial position is in three-dimensional coordinates. The attitude includes pitch, roll, and yaw angles; Using a sliding window to select including and The most recent t sets of data, i=1,2,3,...,t-1, The pose correction value at time i; the observed three-dimensional gradient vector of gas concentration at time t is obtained by fitting using the least squares method. ; Based on the pose transition matrix at time t Observed values ​​of the three-dimensional gradient vector of gas concentration at time t Calculate the predicted gas concentration change at time t relative to time t-1; establish a gas gradient error term based on the observed and predicted gas concentration changes at time t relative to time t-1; weight and fuse the traditional geometric error term with the gas gradient error term; construct an inter-frame matching pose transfer matrix optimization function with the goal of minimizing the weighted fusion result; and solve the pose transfer matrix at time t using the Gauss-Newton method. ; Based on the pose transition matrix at time t , initial pose Perform geometric feature matching with the map obtained at time t-1 before, and calculate the geometric feature matching error; An objective function is established with the goal of minimizing the weighted sum of geometric feature matching error and gas concentration error, and the initial pose at time t is obtained by solving the function. pose correction value at time t ; Based on pose correction value at time t The feature point set at time t Feature point matching and point cloud fusion are performed with the map obtained at time t-1 before to obtain the local map at time t before; In the historical exploration data, 3D point cloud data with geometrically significant features are selected as keyframes to establish a keyframe set. The 3D point cloud data at time t is used as the current frame. Loop closure detection is performed based on the keyframes and the current frame. After confirming the loop closure, the local map at time t is optimized for global consistency to obtain the optimized map.

2. The underground space detection and positioning method combining laser information and gas concentration information according to claim 1, characterized in that, Obtain the observed three-dimensional gradient vector of gas concentration at time t. The method is as follows: Construct a system of linear equations: ， In the formula for The pose difference matrix, where the i-th row is the initial position difference between adjacent time steps. , for The concentration difference matrix, with the i-th row representing the observed gas concentration difference. ; The three-dimensional gradient vector of gas concentration at time t is obtained by solving the problem. : 。 3. The underground space detection and positioning method combining laser information and gas concentration information according to claim 2, characterized in that, The predicted change in gas concentration at time t relative to time t-1 is: ， Pose transition matrix It is in the form of a 4×4 homogeneous transformation matrix. To be Initial pose of dimension Convert to Transformation functions in 3D matrix form, To be dimension Convert to The conversion function; As a translation vector, it constitutes the translation part. Transformed into a rotation matrix via Rodrigues transformation Combined in standard form according to homogeneous transformation matrix Dimensional matrix ; The corresponding observed change in gas concentration at time t relative to time t-1 is Establish the gas gradient error term for: 。 4. The underground space detection and positioning method combining laser information and gas concentration information according to claim 3, characterized in that, The pose transition matrix optimization function is constructed as follows: ， In the formula For traditional geometric error terms, The gas gradient error weights; ， In the formula, N represents the number of 3D point cloud matches. Let n be the matching position of the nth 3D point cloud at time t; Solve the pose transition matrix optimization function to obtain the pose transition matrix at time t. .

5. The underground space detection and positioning method combining laser information and gas concentration information according to claim 4, characterized in that, The gas concentration error is the observed gas concentration value at time t. The difference between the predicted gas concentration at time t and the predicted gas concentration at time t; the predicted gas concentration at time t is expressed as : ， In the formula These are the three-dimensional coordinate correction values ​​for the position.

6. The underground space detection and positioning method combining laser information and gas concentration information according to claim 5, characterized in that, The objective function is established with the goal of minimizing the weighted sum of geometric feature matching error and gas concentration error as follows: ， In the formula, K is the geometric feature matching number. Let be the matching error of the k-th geometric feature, and be the matching position of the k-th 3D point cloud at time t. The distance between the location and the matching location in the local map; Weights are matched for geometric features. This is the gas concentration error weight; The objective function is solved using the LM nonlinear least squares algorithm, and the optimal initial pose at time t obtained by iterative solution is then used. As the pose correction value at time t .

7. The underground space detection and positioning method combining laser information and gas concentration information according to claim 6, characterized in that, Loop closure detection involves calculating the geometric similarity between the feature point sets of keyframes and the current frame. Gas similarity to gas concentration observations corresponding to keyframes and the current frame .

8. The method for underground space detection and positioning combining laser information and gas concentration information according to claim 7, characterized in that, The sequence of nearest-neighbor gas concentration observations corresponding to the keyframe is represented as a feature vector. The sequence of nearest-neighbor gas concentration observations corresponding to the current frame is represented as a feature vector. Gas similarity is calculated using cosine similarity. : 。 9. The method for underground space detection and positioning combining laser information and gas concentration information according to claim 8, characterized in that, Loop closure detection also includes calculating loop closure candidate scores. : ， In the formula Geometric similarity weights; If the loop candidate score If the value exceeds the set loopback threshold, continue with geometric verification of the keyframe and the current frame, ICP matching error verification, and overlap verification. After passing all verifications, the loopback is confirmed.

10. The method for underground space detection and positioning combining laser information and gas concentration information according to claim 9, characterized in that, Methods for optimizing the global consistency of a local map at time t include nonlinear least squares or graph optimization.

Images