Lidar tunnel detection method and apparatus, storage medium, and electronic device

Abstract

The application provides a laser radar tunnel detection method and device, a storage medium and an electronic device. The method comprises the following steps: setting a first reference point and a second reference point; scanning to obtain a point cloud, obtaining a first distortion point and a second distortion point in the point cloud, and recording a timestamp; calculating a reference vector and a first distortion vector, and then obtaining a distortion error; recording motion state data of a laser radar vehicle; taking the above data as training data to train a distortion detection model; obtaining a third distortion point, obtaining a second distortion vector, detecting a distortion error of the second distortion vector, and performing distortion compensation to obtain a first compensation vector; setting a matching degree threshold, calculating a matching degree of the first compensation vector and the reference vector, if the matching degree reaches the matching degree threshold, the distortion compensation is completed, otherwise, the first compensation vector is set as a new second distortion vector, and the previous step is returned. The application improves the overall accuracy of the point cloud by using a small number of reference points.

Description

Technical Field

[0001] This invention relates to the field of tunnel detection technology, and in particular to a method, apparatus, storage medium and electronic device for tunnel detection based on lidar. Background Technology

[0002] LiDAR is a distance measurement technology that uses laser beams to measure the distance and shape of target objects. In tunnel inspection systems, LiDAR is often used to acquire structural information about the tunnel's interior, such as its shape, dimensions, cracks, and deformation. Motion compensation algorithms are used to accurately measure target positions in moving or vibrating environments. In LiDAR tunnel inspection systems, motion compensation algorithms are used to correct the relative motion between the LiDAR sensor and the target, ensuring that the acquired point cloud data accurately reflects the actual internal structure of the tunnel.

[0003] In traditional LiDAR-based tunnel detection systems, motion compensation algorithms primarily address the issue of LiDAR data offset caused by factors such as acceleration and inertia during vehicle or robot movement. This offset can lead to inaccurate representation of the position and shape of the tunnel's internal structure in the LiDAR scan data. The goal of motion compensation algorithms is to correct these offsets, enabling the LiDAR-acquired data to more accurately reflect the actual structure and obstacle positions within the tunnel.

[0004] However, in the motion compensation algorithm of traditional tunnel detection systems, a high-precision map of the tunnel interior is required as a reference. However, in practical applications, obtaining a high-precision map is technically very difficult and expensive, making it unsuitable for large-scale applications. Obtaining a small number of reference points is relatively easier. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a LiDAR-based tunnel detection method, which aims to solve the technical problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] A method for tunnel detection based on lidar includes the following steps:

[0008] (1) Establish a tunnel coordinate system and set a first reference point and a second reference point inside the tunnel;

[0009] (2) A point cloud is obtained by moving the lidar vehicle to scan, and a first distortion point corresponding to the first reference point and a second distortion point corresponding to the second reference point are obtained from the point cloud, and a timestamp is recorded.

[0010] (3) Calculate the reference vector using the first reference point and the second reference point, and calculate the first distortion vector using the first distortion point and the second distortion point. Then, obtain the distortion error D using the reference vector and the first distortion vector. i ;

[0011] (4) Based on the tunnel coordinate system and the timestamp, record the motion state data of the lidar vehicle at each time corresponding to the timestamp while the lidar vehicle is scanning;

[0012] (5) The first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data and the distortion error are used as training data to train a distortion detection model, which is used to determine the distortion error of the lidar vehicle.

[0013] (6) Obtain the third distortion point in the point cloud, use the third distortion point to obtain the second distortion vector, use the distortion detection model to detect the distortion error of the second distortion vector and perform distortion compensation to obtain the first compensation vector;

[0014] (7) Set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed. Otherwise, set the first compensation vector as the new second distortion vector and return to step (6).

[0015] Furthermore, in step (1), there are m first reference points and m second reference points. The m first reference points are numbered 1, 2, ..., m in order of position, where the first reference point i represents the i-th first reference point. The m second reference points are numbered 11, 22, ..., mm in order of position, where the second reference point ii represents the i-th second reference point.

[0016] In step (2), there are m first distortion points and m second distortion points. The m first distortion points are numbered 1', 2', ..., m' according to the corresponding first reference points, where the first distortion point i' represents the i-th first distortion point. The m second distortion points are numbered 11', 22', ..., mm' according to the corresponding second reference points, where the second distortion point ii' represents the i-th second distortion point.

[0017] The distortion error D iThe error between the reference vector of the first reference point i and the second reference point ii and the first distortion vector of the first distortion point i′ and the second distortion point ii′ is represented. The error includes: rotation angle θ, scaling factor λ, and translation distance ξ.

[0018] Furthermore, the specific steps of step (3) include:

[0019] Let the three-dimensional coordinates of the first reference point i be represented as, (x i y i , z i The three-dimensional coordinates of the second reference point ii are represented as follows: Therefore, the formula for calculating the reference vector is obtained:

[0020] The three-dimensional coordinates of the first distortion point i′ are represented as (x' i y' i , z' i The three-dimensional coordinates of the second distortion point ii′ are represented as follows: Therefore, the formula for calculating the first distortion vector is obtained:

[0021] The formula for calculating the rotation angle θ is as follows: The scaling factor λ is calculated using the following formula: The formula for calculating the translation distance ξ is as follows: The distortion error D is obtained. i =(θ i , λ i ξ i ), where θ i , λ i ξ i These represent the first distortion vectors respectively. The rotation angle, scaling factor, and translation distance.

[0022] Furthermore, in step (4), the motion state data includes the displacement, velocity, acceleration, attitude information and angular velocity of the lidar vehicle, and the attitude information includes the pitch angle, yaw angle and roll angle of the lidar vehicle.

[0023] Furthermore, the specific steps of step (5) include:

[0024] The first reference point (x) i y i , z i The first reference point sequence {(x1, y1, z1), (x2, y2, z2), ..., (x...} is obtained by arranging the sequence according to the timestamps. my m , z m )}, the second reference point The second reference point sequence is obtained by assembling a sequence according to the order of the timestamps. This leads to the corresponding reference vector sequence;

[0025] The first distortion point (x') i y' i , z' i The first distortion point sequence is obtained by assembling the sequence according to the timestamps: {((x'1, y'1, z'1), (x'2, y'2, z'2), ..., (x'... m y' m , z' m )}, the second distortion point The second distortion point sequence is obtained by assembling the sequence according to the timestamps. This leads to the corresponding first distortion vector sequence;

[0026] The motion state data of the lidar vehicle are arranged in the order of the timestamps to obtain a motion state sequence (V1, V2, ..., V...). m );

[0027] The distortion error sequence {D1, D2, ..., D} is calculated based on the reference vector sequence and the first distortion vector sequence. m};

[0028] Set the sliding window length, sliding step size, and prediction time step size, and obtain training samples from the motion state sequence and the distortion error sequence using the sliding window method. Use the training samples as training data for training the distortion detection model.

[0029] The first reference point, the second reference point, the first distortion point, the second distortion point, and the motion state data are used as input data for training the distortion detection model. The distortion error is used as output data for training the distortion detection model. The model aims to accurately determine the distortion error within a short prediction time as the prediction target, and to minimize the loss function between the actual distortion error and the predicted distortion error. As a training target, a final distortion detection model is trained to determine the distortion error of the lidar vehicle;

[0030] Where D i This represents the actual distortion error at the i-th timestamp. The loss function represents the distortion error predicted at the i-th timestamp. The calculation formula is:

[0031] Furthermore, the specific steps of step (6) include:

[0032] Using any two points from the third distortion points as the starting and ending points of the second distortion vector, the second distortion vector (Δx') is obtained. i ,Δy' i , Δz' i );

[0033] Using the second distortion vector and the translation distance ξ i The formula for calculating translation compensation is as follows:

[0034]

[0035] The second distortion vector is rotated and compensated using the rotation angle θ, and the calculation formula is as follows:

[0036]

[0037] The second distortion vector is scaled and compensated using the scaling factor λ, and the calculation formula is as follows:

[0038]

[0039] Finally, the first compensation vector (L″) is obtained. x , L″ y , L″ z ).

[0040] Furthermore, in step (7), the specific steps for calculating the matching degree E between the first compensation vector and the reference vector include:

[0041] Feature points are extracted from the first and second reference points on the known reference vector to obtain reference point features;

[0042] The first compensation point on the first compensation vector is matched with the known first reference point and second reference point using feature points to calculate the Euclidean distance between the first compensation point and the matched first reference point or second reference point. The Euclidean distance is the matching degree E between the first compensation point and the first reference point or second reference point.

[0043] The present invention also provides a lidar-based tunnel detection device, comprising:

[0044] Reference point setting module: used to establish the tunnel coordinate system and set the first and second reference points within the tunnel;

[0045] The reference point setting module is specifically used to: set m reference points for both the first reference point and the second reference point, number the m first reference points in order of position as 1, 2, ..., m, where the first reference point i represents the i-th first reference point, and number the m second reference points in order of position as 11, 22, ..., mm, where the second reference point ii represents the i-th second reference point;

[0046] Distortion point acquisition module: used to obtain point cloud by moving and scanning with lidar vehicle, acquire the first distortion point corresponding to the first reference point and the second distortion point corresponding to the second reference point in the point cloud, and record the timestamp;

[0047] The distortion point acquisition module is specifically used to: set m first distortion points and m second distortion points, number the m first distortion points as 1', 2', ..., m' according to the corresponding first reference points, wherein the first distortion point i' represents the i-th first distortion point, and number the m second distortion points as 11', 22', ..., mm' according to the corresponding second reference points, wherein the second distortion point ii' represents the i-th second distortion point;

[0048] Distortion error calculation module: used to calculate a reference vector using the first reference point and the second reference point, and to calculate a first distortion vector using the first distortion point and the second distortion point, and then to obtain the distortion error D using the reference vector and the first distortion vector. i ;

[0049] The distortion error calculation module is specifically used to: represent the three-dimensional coordinates of the first reference point i as, (x... i y i , z i The three-dimensional coordinates of the second reference point ii are represented as follows: Therefore, the formula for calculating the reference vector is obtained:

[0050] The three-dimensional coordinates of the first distortion point i′ are represented as (x' i y' i , z' i The three-dimensional coordinates of the second distortion point ii′ are represented as follows: Therefore, the formula for calculating the first distortion vector is obtained:

[0051] The formula for calculating the rotation angle θ is as follows: The scaling factor λ is calculated using the following formula: The formula for calculating the translation distance ξ is as follows: The distortion error D is obtained. i =(θ i , λ i ξ i ), where θ i , λ i ξ i These represent the first distortion vectors respectively. The rotation angle, scaling factor, and translation distance;

[0052] Motion data acquisition module: used to record the motion state data of the lidar vehicle at each time corresponding to the timestamp, based on the tunnel coordinate system and the timestamp;

[0053] Model training module: used to train a distortion detection model by using the first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data and the distortion error as training data. The distortion detection model is used to determine the distortion error of the lidar vehicle.

[0054] The model training module is specifically used to: train the first reference point (x) i y i , z i The first reference point sequence {(x1, y1, z1), (x2, y2, z2), ..., (x...} is obtained by arranging the sequence according to the timestamps. m y m , z m )}, the second reference point The second reference point sequence is obtained by assembling a sequence according to the order of the timestamps. This leads to the corresponding reference vector sequence;

[0055] The first distortion point (x') i y' i , z' i The first distortion point sequence {(x'1, y'1, z'1), (x'2, y'2, z'2), ..., (x'1, y'1, z'1)} is obtained by arranging the sequences according to the timestamps. m y' m , z' m )}, the second distortion point The second distortion point sequence is obtained by assembling the sequence according to the timestamps. This leads to the corresponding first distortion vector sequence;

[0056] The motion state data of the lidar vehicle are arranged in the order of the timestamps to obtain the motion state sequence {V1, V2, ..., V...}. m};

[0057] The distortion error sequence {D1, D2, ..., D...} is calculated based on the reference vector sequence and the first distortion vector sequence. m};

[0058] Set the sliding window length, sliding step size, and prediction time step size, and obtain training samples from the motion state sequence and the distortion error sequence using the sliding window method. Use the training samples as training data for training the distortion detection model.

[0059] The first reference point, the second reference point, the first distortion point, the second distortion point, and the motion state data are used as input data for training the distortion detection model. The distortion error is used as output data for training the distortion detection model. The model aims to accurately determine the distortion error within a short prediction time as the prediction target, and to minimize the loss function between the actual distortion error and the predicted distortion error. As a training target, a final distortion detection model is trained to determine the distortion error of the lidar vehicle;

[0060] Where D i This represents the actual distortion error at the i-th timestamp. The loss function represents the distortion error predicted at the i-th timestamp. The calculation formula is:

[0061] Distortion compensation module: used to obtain the third distortion point in the point cloud, obtain the second distortion vector using the third distortion point, detect the distortion error of the second distortion vector using the distortion detection model and perform distortion compensation to obtain the first compensation vector;

[0062] The distortion compensation module is specifically used to: take any two points among the third distortion points as the starting and ending points of the second distortion vector, so as to obtain the second distortion vector (Δx'). i ,Δy' i , Δz' i );

[0063] Using the second distortion vector and the translation distance ξ i The formula for calculating translation compensation is as follows:

[0064]

[0065] The second distortion vector is rotated and compensated using the rotation angle θ, and the calculation formula is as follows:

[0066]

[0067] The second distortion vector is scaled and compensated using the scaling factor λ, and the calculation formula is as follows:

[0068]

[0069] Finally, the first compensation vector (L″) is obtained. x , L″ y , L″ z );

[0070] Iterative optimization module: used to set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed; otherwise, the first compensation vector is set as the new second distortion vector, and the process is returned to the distortion compensation module for iterative optimization.

[0071] The iterative optimization module is specifically used to: extract feature points from the first and second reference points on the known reference vector to obtain reference point features;

[0072] The first compensation point on the first compensation vector is matched with the known first reference point and second reference point using feature points to calculate the Euclidean distance between the first compensation point and the matched first reference point or second reference point. The Euclidean distance is the matching degree E between the first compensation point and the first reference point or second reference point.

[0073] The present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the lidar-based tunneling detection method as described above.

[0074] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the lidar-based tunneling detection method as described above.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0076] By setting a first reference point and a second reference point, obtaining a high-precision map of the entire tunnel is avoided, reducing the complexity of obtaining distortion reference objects. A point cloud is obtained by moving the lidar vehicle to scan, acquiring the first distortion point corresponding to the first reference point and the second distortion point corresponding to the second reference point, and recording the timestamps. A reference vector is calculated using the first and second reference points, and a distortion vector is calculated using the first and second distortion points. The distortion error is then calculated using the reference vector and the distortion vector. During the lidar scanning process, the motion state data of the lidar vehicle at the corresponding timestamps is recorded. The first and second reference points, the first and second distortion points, the motion state data, and the distortion error are used as training data to train a distortion detection model that judges the distortion error of the lidar vehicle. The model is then used to learn the distortion error in the point cloud. The distortion patterns of some points are analyzed to obtain a distortion detection model for analyzing the overall distortion patterns of the point cloud. A third distortion point is obtained from the point cloud, and a second distortion vector is derived using this third distortion point. The distortion error of the second distortion vector is detected using the distortion detection model, and distortion compensation is performed to obtain a first compensation vector. The matching degree between the first compensation vector and the reference vector is calculated. If the matching degree reaches a matching degree threshold, the distortion compensation is complete; otherwise, the first compensation vector is set as a new distortion vector for iterative optimization until the matching degree reaches the matching degree threshold. Ultimately, the overall motion distortion compensation effect is achieved, improving the accuracy of the point cloud. By using a small number of reference points and distortion points to learn the motion distortion patterns of the point cloud, the complex steps of obtaining a high-precision tunnel map in the early stages are avoided, and the overall accuracy of the point cloud is improved through a small number of reference points. Attached Figure Description

[0077] Figure 1 This is a flowchart of the lidar-based tunnel detection method in the first embodiment of the present invention;

[0078] Figure 2 for Figure 1 A schematic diagram of steps S102 and S103;

[0079] Figure 3 This is a structural block diagram of the motion compensation device for the tunnel detection system in the third embodiment of the present invention;

[0080] Figure 4 This is a structural block diagram of the computer device in the fourth embodiment of the present invention;

[0081] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0082] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0083] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0084] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0085] Example 1

[0086] Please see Figure 1 The image shows a lidar-based tunnel detection method in the first embodiment of the present invention, comprising the following steps S101 to S107:

[0087] S101, Establish a tunnel coordinate system and set the first and second reference points inside the tunnel;

[0088] S102, a point cloud is obtained by moving the lidar vehicle to scan, and a first distortion point corresponding to the first reference point and a second distortion point corresponding to the second reference point are obtained from the point cloud, and a timestamp is recorded;

[0089] S103, calculate the reference vector using the first reference point and the second reference point, and calculate the first distortion vector using the first distortion point and the second distortion point. Then, obtain the distortion error D using the reference vector and the first distortion vector. i ;

[0090] S104, Based on the tunnel coordinate system and the timestamp, while the lidar vehicle is scanning, record the motion state data of the lidar vehicle at each time corresponding to the timestamp;

[0091] S105, the first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data, and the distortion error are used as training data to train a distortion detection model, which is used to determine the distortion error of the lidar vehicle;

[0092] S106, obtain the third distortion point in the point cloud, use the third distortion point to obtain the second distortion vector, use the distortion detection model to detect the distortion error of the second distortion vector and perform distortion compensation to obtain the first compensation vector;

[0093] S107, Set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed; otherwise, set the first compensation vector as the new second distortion vector and return to step S106.

[0094] For example, in step S101, the method for establishing the tunnel coordinate system is as follows: the initial position of the lidar vehicle is taken as the origin (0, 0, 0), the direction of the lidar vehicle's movement is the positive x-axis, the left side of the lidar vehicle is the positive y-axis, and the top of the lidar vehicle is the positive z-axis.

[0095] The first and second reference points refer to known points pre-marked on the tunnel wall, which are used as the basis for judging the motion distortion of the point cloud.

[0096] There are m first reference points and m second reference points. The first and second reference points are obtained by manually marking m first reference points and m second reference points in the tunnel with color. The m first reference points are numbered 1, 2, ..., m according to their position, where the first reference point i represents the i-th first reference point. The m second reference points are numbered 11, 22, ..., mm according to their position, where the second reference point ii represents the i-th second reference point.

[0097] The purpose of setting the first and second reference points is to randomly set m first reference points and m second reference points in the tunnel and mark them with colors. When the lidar scans the tunnel to obtain a point cloud, the first and second reference points will show a different color depth than the surrounding points in the point cloud, which makes it easier to distinguish them from other points in the point cloud.

[0098] Among them, the first reference point i and the second reference point ii are a pair of reference points;

[0099] By establishing a tunnel coordinate system and setting a first and second reference point, a beneficial effect was achieved: providing data for obtaining distortion points by subsequently scanning the tunnel with lidar.

[0100] In step S102, the process of the lidar vehicle moving and scanning to obtain point clouds is to scan according to the scanning cycle Z, and record a timestamp in each scanning cycle;

[0101] The first distortion point and the second distortion point refer to points in the point cloud obtained by lidar;

[0102] Since the lidar is carried and moved by the lidar vehicle to perform scanning, the point cloud acquired by the lidar vehicle will be distorted due to the movement of the lidar vehicle. The first distorted point and the second distorted point in the point cloud are the corresponding points of the first reference point and the second reference point on the point cloud with motion distortion, respectively.

[0103] There are m first distortion points and m second distortion points. The method for obtaining the first distortion points and the second distortion points is as follows: feature extraction is performed on the points in the point cloud to find the first distortion point corresponding to the first reference point. The m first reference points are numbered 1, 2, ..., m in positional order, where the first reference point i represents the i-th first reference point. The m second reference points are numbered 11, 22, ..., mm in positional order, where the second reference point ii represents the i-th second reference point.

[0104] The first distortion point i' and the second distortion point ii' are a pair of distortion points;

[0105] The timestamp refers to the time recorded when the lidar scans the point cloud, and each point in the point cloud has a corresponding timestamp.

[0106] The point cloud data includes the three-dimensional coordinates of the point cloud and the color of the point cloud;

[0107] By acquiring the first distortion point corresponding to the first reference point and the second distortion point corresponding to the second reference point in the point cloud, the acquired first distortion point and second distortion point are comparable with the corresponding first reference point and second reference point, thus achieving the beneficial effect of providing a data foundation for distortion analysis.

[0108] In step S103, the reference vector refers to the vector between the first reference point and the second reference point;

[0109] The distortion vector refers to the vector between the first distortion point and the second distortion point;

[0110] The distortion error refers to the error caused by motion distortion in the distortion vector between the first distortion point and the second distortion point compared to the reference vector between the first reference point and the second reference point. The distortion error D... i The error between the reference vector of the first reference point i and the second reference point ii and the first distortion vector of the first distortion point i′ and the second distortion point ii′ is represented. The error includes: rotation angle θ, scaling factor λ, and translation distance ξ.

[0111] The specific steps of step S103 include:

[0112] Let the three-dimensional coordinates of the first reference point i be represented as, (x i y i , z i The three-dimensional coordinates of the second reference point ii are represented as follows: Therefore, the formula for calculating the reference vector is obtained:

[0113] The three-dimensional coordinates of the first distortion point i′ are represented as (x′). i y′ i , z′ i The three-dimensional coordinates of the second distortion point ii′ are represented as follows: Therefore, the formula for calculating the first distortion vector is obtained:

[0114] The formula for calculating the rotation angle θ is as follows: The scaling factor λ is calculated using the following formula: The formula for calculating the translation distance ξ is as follows: The distortion error D is obtained. i =(θ i , λ i ξ i ), where θ i , λ i ξ i These represent the first distortion vectors respectively. The rotation angle, scaling factor, and translation distance.

[0115] By calculating the reference vector based on the first and second reference points, and calculating the distortion vector using the first and second distortion points, the distortion error of the reference vector and distortion vector is analyzed. The rotation angle, scaling factor, and translation distance are calculated respectively, which achieves the beneficial effect of providing distortion error data support for subsequent model training.

[0116] In step S104, the step of recording the motion state data of the LiDAR vehicle at each time corresponding to the timestamp while the LiDAR vehicle is scanning means that during the LiDAR scanning process, an inertial navigation system is installed on the LiDAR vehicle, and the motion state data of the LiDAR vehicle is collected by the inertial navigation system according to the time sequence of the timestamps.

[0117] The motion state data includes the displacement, velocity, acceleration, attitude information, and angular velocity of the lidar vehicle, and the attitude information includes the pitch angle, yaw angle, and roll angle of the lidar vehicle.

[0118] This step obtains the motion state data of the lidar vehicle at the corresponding time based on the tunnel coordinate system and timestamp. This motion state data serves as the training data for the subsequent training of the distortion detection model, providing a data foundation for the training of the subsequent model.

[0119] The specific steps of step S105 include:

[0120] The first reference point (x) i y i , z i The first reference point sequence {(x1, y1, z1), (x2, y2, z2), ..., (x...} is obtained by arranging the sequence according to the timestamps. m y m , z m )}, the second reference point The second reference point sequence is obtained by assembling a sequence according to the order of the timestamps. This leads to the corresponding reference vector sequence;

[0121] The first distortion point (x') i y' i , z' i The first distortion point sequence {(x'1, y'1, z'1), (x'2, y'2, z'2), ..., (x'1, y'1, z'1)} is obtained by arranging the sequences according to the timestamps. m y' m , z' m )}, the second distortion point The second distortion point sequence is obtained by assembling the sequence according to the timestamps. This leads to the corresponding first distortion vector sequence;

[0122] The motion state data of the lidar vehicle are arranged in the order of the timestamps to obtain the motion state sequence {V1, V2, ..., V...}. m};

[0123] The distortion error sequence {D1, D2, ..., D...} is calculated based on the reference vector sequence and the first distortion vector sequence. m};

[0124] Set the sliding window length, sliding step size, and prediction time step size, and obtain training samples from the motion state sequence and the distortion error sequence using the sliding window method. Use the training samples as training data for training the distortion detection model.

[0125] The first reference point, the second reference point, the first distortion point, the second distortion point, and the motion state data are used as input data for training the distortion detection model. The distortion error is used as output data for training the distortion detection model. The model aims to accurately determine the distortion error within a short prediction time as the prediction target, and to minimize the loss function between the actual distortion error and the predicted distortion error. As a training target, a final distortion detection model is trained to determine the distortion error of the lidar vehicle;

[0126] Where D i This represents the actual distortion error at the i-th timestamp. The loss function represents the distortion error predicted at the i-th timestamp. The calculation formula is:

[0127] The sliding window method is a commonly used technique in this field and will not be described in detail here;

[0128] Preferably, the distortion detection model is an LSTM model from deep learning models;

[0129] After the distortion detection model is trained, the distortion error is detected using the first distortion point, the second distortion point, and motion state data as input data to the model, and the distortion error of the distortion vector between the first distortion point and the second distortion point is output.

[0130] This step utilizes the first and second reference points, the first and second distortion points, motion state data, and distortion errors as training data to ensure the analyzability between the reference points and distortion points. This trains a distortion detection model to judge the distortion error of the lidar vehicle, providing a model foundation for subsequent detection of distortion errors of points in the point cloud.

[0131] In step S106, the third distortion point refers to a point in the point cloud that is not labeled with the first reference point and the second reference point;

[0132] The first compensation vector refers to the second distortion vector after distortion compensation, and the point on the first compensation vector is the first compensation point;

[0133] The specific steps of step S106 include:

[0134] Using any two points from the third distortion points as the starting and ending points of the second distortion vector, the second distortion vector (Δx') is obtained. i ,Δy' i , Δz' i );

[0135] Using the second distortion vector and the translation distance ξ i The formula for calculating translation compensation is as follows:

[0136]

[0137] The second distortion vector is rotated and compensated using the rotation angle θ, and the calculation formula is as follows:

[0138]

[0139] The second distortion vector is scaled and compensated using the scaling factor λ, and the calculation formula is as follows:

[0140]

[0141] Finally, the first compensation vector (L″) is obtained. x , L″ y , L″ z );

[0142] This step involves obtaining the third distortion point in the point cloud, using a distortion detection model to detect the distortion error of the third distortion point, and then translating, scaling, and rotating the second distortion vector of the third distortion point to compensate the distortion of the second distortion vector into the first compensation vector. This provides a data foundation for subsequently calculating the matching degree between the first compensation vector and the reference vector.

[0143] In step S107, the specific steps for calculating the matching degree E between the first compensation vector and the reference vector include:

[0144] Feature points are extracted from the first and second reference points on the known reference vector to obtain reference point features;

[0145] The first compensation point on the first compensation vector is matched with the known first reference point and second reference point to calculate the Euclidean distance between the first compensation point and the matched first reference point or second reference point. The Euclidean distance is the matching degree E between the first compensation point and the first reference point or second reference point.

[0146] The matching degree threshold Q is set manually. When the matching degree E is less than the matching degree threshold Q, it means that the first compensation vector obtained by the initial distortion compensation of the second distortion vector needs to be further distorted. The first compensation vector is used as the new second distortion vector, and the process returns to step S106 until the matching degree E between the first compensation vector after distortion compensation and the reference vector reaches the matching degree threshold Q, and the distortion compensation is completed.

[0147] This step obtains the first compensation vector through distortion compensation, sets a matching degree threshold, and calculates the matching degree to determine the matching degree between the second distortion vector and the first compensation vector after distortion compensation and the reference vector. If the matching degree does not meet the matching degree threshold, the matching degree of the first compensation vector after distortion compensation is further optimized through iteration until the matching degree reaches the matching degree threshold and the iteration ends, thus achieving the beneficial effect of gradually improving the distortion compensation effect.

[0148] In summary, the lidar-based tunnel detection method in the above embodiments of the present invention avoids acquiring a high-precision map of the entire tunnel by setting a first reference point and a second reference point, reducing the complexity of acquiring distortion reference objects. It uses a lidar vehicle to scan and obtain a point cloud, acquiring the first distortion point corresponding to the first reference point and the second distortion point corresponding to the second reference point, and recording timestamps. A reference vector is calculated using the first and second reference points, and a distortion vector is calculated using the first and second distortion points. The distortion error is calculated using the reference vector and the distortion vector. During the lidar scanning process, the motion state data of the lidar vehicle at the timestamps are recorded. The first and second reference points, the first and second distortion points, the motion state data, and the distortion error are used as training data to train a distortion detection method to judge the distortion error of the lidar vehicle. The model learns the distortion patterns of partial points in a point cloud, resulting in a distortion detection model used to analyze the overall distortion patterns of the point cloud. It identifies a third distortion point in the point cloud and uses this third distortion point to obtain a second distortion vector. The distortion detection model then detects the distortion error of the second distortion vector and performs distortion compensation to obtain a first compensation vector. The matching degree between the first compensation vector and the reference vector is calculated. If the matching degree reaches a threshold, distortion compensation is complete; otherwise, the first compensation vector is set as a new distortion vector for iterative optimization until the matching degree reaches the threshold. Ultimately, this achieves overall motion distortion compensation, improving the accuracy of the point cloud. By using a small number of reference points and distortion points to learn the motion distortion patterns of the point cloud, the complex steps of acquiring a high-precision tunnel map in the early stages are avoided, and the overall accuracy of the point cloud is improved through a small number of reference points.

[0149] Example 2

[0150] Please refer to Figure 2 The diagram shown is a schematic representation of steps S102 and S103 in the first embodiment of the present invention. Figure 2 The diagram shown is a planar schematic of setting the first and second reference points and obtaining the first and second distortion points by scanning with a lidar vehicle.

[0151] In this diagram, two vertical lines represent the two walls of the tunnel's overhead view. The black triangle on the left indicates the first reference point set on the tunnel, and the black circle in the middle indicates the second reference point set on the tunnel. After the lidar vehicle scans the tunnel, it obtains a point cloud with motion distortion errors. In the point cloud, the position of the first reference point shifts to the gray triangle in the diagram due to distortion, representing the first distortion point corresponding to the first reference point. The position of the second reference point shifts to the gray circle in the point cloud, representing the second distortion point corresponding to the second reference point. The arrow pointing from the first reference point to the second reference point represents the reference vector, and the arrow pointing from the first distortion point to the second distortion point represents the distortion vector.

[0152] Example 3

[0153] Please refer to Figure 3 The image shows a lidar-based tunnel detection device according to a third embodiment of the present invention, comprising:

[0154] Reference point setting module 11: used to establish a tunnel coordinate system and set the first and second reference points within the tunnel;

[0155] The reference point setting module 11 is specifically used to: set m reference points for both the first reference point and the second reference point, number the m first reference points in order of position as 1, 2, ..., m, where the first reference point i represents the i-th first reference point, and number the m second reference points in order of position as 11, 22, ..., mm, where the second reference point ii represents the i-th second reference point;

[0156] Distortion point acquisition module 12: used to obtain a point cloud by moving and scanning with a lidar vehicle, acquire a first distortion point corresponding to the first reference point and a second distortion point corresponding to the second reference point in the point cloud, and record a timestamp;

[0157] The distortion point acquisition module 12 is specifically used to: set m first distortion points and m second distortion points, number the m first distortion points as 1', 2', ..., m' according to the corresponding first reference points, wherein the first distortion point i' represents the i-th first distortion point, and number the m second distortion points as 11', 22', ..., mm' according to the corresponding second reference points, wherein the second distortion point ii' represents the i-th second distortion point;

[0158] Distortion error calculation module 13: used to calculate a reference vector using the first reference point and the second reference point, and to calculate a first distortion vector using the first distortion point and the second distortion point, and then to obtain the distortion error D using the reference vector and the first distortion vector. i ;

[0159] The distortion error calculation module 13 is specifically used to: represent the three-dimensional coordinates of the first reference point i as, (x i y i , z i The three-dimensional coordinates of the second reference point ii are represented as follows: Therefore, the formula for calculating the reference vector is obtained:

[0160] The three-dimensional coordinates of the first distortion point i' are represented as (x' i y'i , z' i The three-dimensional coordinates of the second distortion point ii′ are represented as follows: Therefore, the formula for calculating the first distortion vector is obtained:

[0161] The formula for calculating the rotation angle θ is as follows: The scaling factor λ is calculated using the following formula: The formula for calculating the translation distance ξ is as follows: The distortion error D is obtained. i =(θ i , λ i ξ i ), where θ i , λ i ξ i These represent the first distortion vectors respectively. The rotation angle, scaling factor, and translation distance;

[0162] Motion data acquisition module 14: used to record the motion state data of the lidar vehicle at each time corresponding to the timestamp while the lidar vehicle is scanning, based on the tunnel coordinate system and the timestamp;

[0163] Model training module 15: used to train a distortion detection model by using the first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data and the distortion error as training data, and the distortion detection model is used to determine the distortion error of the lidar vehicle;

[0164] The model training module 15 is specifically used for: training the first reference point (x) i y i , z i The first reference point sequence {(x1, y1, z1), (x2, y2, z2), ..., (x...} is obtained by arranging the sequence according to the timestamps. m y m , z m )}, the second reference point The second reference point sequence is obtained by assembling a sequence according to the order of the timestamps. This leads to the corresponding reference vector sequence;

[0165] The first distortion point (x') i y' i , z' i The first distortion point sequence {(x'1, y'1, z'1), (x'2, y'2, z'2), ..., (x'1, y'1, z'1)} is obtained by arranging the sequences according to the timestamps. m y'm , z' m )}, the second distortion point The second distortion point sequence is obtained by assembling the sequence according to the timestamps. This leads to the corresponding first distortion vector sequence;

[0166] The motion state data of the lidar vehicle are arranged in the order of the timestamps to obtain the motion state sequence {V1, V2, ..., V...}. m};

[0167] The distortion error sequence {D1, D2, ..., D...} is calculated based on the reference vector sequence and the first distortion vector sequence. m};

[0168] Set the sliding window length, sliding step size, and prediction time step size, and obtain training samples from the motion state sequence and the distortion error sequence using the sliding window method. Use the training samples as training data for training the distortion detection model.

[0169] The first reference point, the second reference point, the first distortion point, the second distortion point, and the motion state data are used as input data for training the distortion detection model. The distortion error is used as output data for training the distortion detection model. The model aims to accurately determine the distortion error within a short prediction time as the prediction target, and to minimize the loss function between the actual distortion error and the predicted distortion error. As a training target, a final distortion detection model is trained to determine the distortion error of the lidar vehicle;

[0170] Where D i This represents the actual distortion error at the i-th timestamp. The loss function represents the distortion error predicted at the i-th timestamp. The calculation formula is:

[0171] Distortion compensation module 16: used to obtain the third distortion point in the point cloud, obtain the second distortion vector using the third distortion point, detect the distortion error of the second distortion vector using the distortion detection model and perform distortion compensation to obtain the first compensation vector;

[0172] The distortion compensation module 16 is specifically used to: take any two points among the third distortion points as the starting and ending points of the second distortion vector to obtain the second distortion vector (Δx'). i ,Δy' i , Δz' i );

[0173] Using the second distortion vector and the translation distance ξ i The formula for calculating translation compensation is as follows:

[0174]

[0175] The second distortion vector is rotated and compensated using the rotation angle θ, and the calculation formula is as follows:

[0176]

[0177] The second distortion vector is scaled and compensated using the scaling factor λ, and the calculation formula is as follows:

[0178]

[0179] Finally, the first compensation vector (L″) is obtained. x , L″ y , L″ z );

[0180] Iterative optimization module 17: used to set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed. Otherwise, the first compensation vector is set as the new second distortion vector, and the process is returned to the distortion compensation module for iterative optimization.

[0181] The iterative optimization module 17 is specifically used to: extract feature points for the first and second reference points on the known reference vector to obtain reference point features;

[0182] The first compensation point on the first compensation vector is matched with the known first reference point and second reference point using feature points to calculate the Euclidean distance between the first compensation point and the matched first reference point or second reference point. The Euclidean distance is the matching degree E between the first compensation point and the first reference point or second reference point.

[0183] Example 4

[0184] The present invention also proposes a computer device, please refer to [link / reference]. Figure 4 The diagram shows a computer device according to a fourth embodiment of the present invention, including a memory 10, a processor 20, and a computer program 30 stored in the memory 10 and executable on the processor 20. When the processor 20 executes the computer program 30, it implements the above-described lidar-based tunnel detection method.

[0185] The memory 10 includes at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 10 can be an internal storage unit of a computer device, such as the hard disk of that computer device. In other embodiments, the memory 10 can be an external storage device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Furthermore, the memory 10 can include both internal and external storage units of the computer device. The memory 10 can be used not only to store application software and various types of data installed on the computer device, but also to temporarily store data that has been output or will be output.

[0186] In some embodiments, the processor 20 may be an electronic control unit (ECU, also known as a vehicle computer), a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip, used to run program code stored in the memory 10 or process data, such as executing access restriction programs.

[0187] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the computer device. In other embodiments, the computer device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0188] This invention also proposes a readable storage medium storing a computer program that, when executed by a processor, implements the lidar-based tunnel detection method described above.

[0189] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0190] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for tunnel detection based on lidar, characterized in that, Includes the following steps: (1) Establish a tunnel coordinate system and set a first reference point and a second reference point inside the tunnel; (2) A point cloud is obtained by moving the lidar vehicle to scan, and a first distortion point corresponding to the first reference point and a second distortion point corresponding to the second reference point are obtained from the point cloud, and a timestamp is recorded. (3) Calculate the reference vector using the first reference point and the second reference point, and calculate the first distortion vector using the first distortion point and the second distortion point. Then, obtain the distortion error D using the reference vector and the first distortion vector. i ; (4) Based on the tunnel coordinate system and the timestamp, record the motion state data of the lidar vehicle at each time corresponding to the timestamp while the lidar vehicle is scanning; (5) The first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data and the distortion error are used as training data to train a distortion detection model, which is used to determine the distortion error of the lidar vehicle. (6) Obtain the third distortion point in the point cloud, use the third distortion point to obtain the second distortion vector, use the distortion detection model to detect the distortion error of the second distortion vector and perform distortion compensation to obtain the first compensation vector; (7) Set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed. Otherwise, set the first compensation vector as the new second distortion vector and return to step (6).

2. The lidar-based tunnel detection method according to claim 1, characterized in that, In step (1), there are m first reference points and m second reference points. The m first reference points are numbered 1, 2, ..., m according to their positional order, where the first reference point i represents the i-th first reference point. The m second reference points are numbered 11, 22, ..., mm according to their positional order, where the second reference point ii represents the i-th second reference point. In step (2), there are m first distortion points and m second distortion points. The m first distortion points are numbered 1′, 2′, ..., m′ according to the corresponding first reference points, where the first distortion point i′ represents the i-th first distortion point. The m second distortion points are numbered 11′, 22′, ..., mm′ according to the corresponding second reference points, where the second distortion point ii′ represents the i-th second distortion point. The distortion error D i The error between the reference vector of the first reference point i and the second reference point ii and the first distortion vector of the first distortion point i′ and the second distortion point ii′ is represented. The error includes: rotation angle θ, scaling factor λ, and translation distance ξ.

3. The lidar-based tunnel detection method according to claim 2, characterized in that, The specific steps of step (3) include: The three-dimensional coordinates of the first reference point i are represented as (x... i y i , z i The three-dimensional coordinates of the second reference point ii are represented as follows: Therefore, the formula for calculating the reference vector is obtained: The three-dimensional coordinates of the first distortion point i′ are represented as (x' i y' i , z' i The three-dimensional coordinates of the second distortion point ii′ are represented as follows: Therefore, the formula for calculating the first distortion vector is obtained: The rotation angle θ is calculated using the following formula: The scaling factor λ is calculated using the following formula: The formula for calculating the translation distance ξ is as follows: The distortion error D is obtained. i =(θ i , λ i ξ i ), where θ i , λ i ξ i These represent the first distortion vectors respectively. The rotation angle, scaling factor, and translation distance.

4. The lidar-based tunnel detection method according to claim 1, characterized in that, In step (4), the motion state data includes the displacement, velocity, acceleration, attitude information and angular velocity of the lidar vehicle, and the attitude information includes the pitch angle, yaw angle and roll angle of the lidar vehicle.

5. The lidar-based tunnel detection method according to claim 3, characterized in that, The specific steps of step (5) include: The first reference point (x) i y i , z i The first reference point sequence {(x1, y1, z1), (x2, y2, z2), ..., (x...} is obtained by arranging the sequence according to the timestamps. m y m , z m )}, the second reference point The second reference point sequence is obtained by assembling a sequence according to the order of the timestamps. This leads to the corresponding reference vector sequence; The first distortion point (x'1, y'1, z') i The first distortion point sequence {(x'1, y'1, z'1), (x'2, y'2, z'2), ..., (x'1, y'1, z'1)} is obtained by arranging the sequences according to the timestamps. m y' m , z' m )}, the second distortion point The second distortion point sequence is obtained by assembling the sequence according to the timestamps. This leads to the corresponding first distortion vector sequence; The motion state data of the lidar vehicle are arranged in the order of the timestamps to obtain the motion state sequence {V1, V2, ..., V...}. m }; The distortion error sequence {D1, D2, ..., D...} is calculated based on the reference vector sequence and the first distortion vector sequence. m }; Set the sliding window length, sliding step size, and prediction time step size, and obtain training samples from the motion state sequence and the distortion error sequence using the sliding window method. Use the training samples as training data for training the distortion detection model. The first reference point, the second reference point, the first distortion point, the second distortion point, and the motion state data are used as input data for training the distortion detection model. The distortion error is used as output data for training the distortion detection model. The model aims to accurately determine the distortion error within a short prediction time as the prediction target, and to minimize the loss function between the actual distortion error and the predicted distortion error. As a training target, a final distortion detection model is trained to determine the distortion error of the lidar vehicle; Where D i This represents the actual distortion error at the i-th timestamp. The loss function represents the distortion error predicted at the i-th timestamp. The calculation formula is:

6. The lidar-based tunnel detection method according to claim 5, characterized in that, The specific steps of step (6) include: Using any two points from the third distortion points as the starting and ending points of the second distortion vector, the second distortion vector (Δx') is obtained. i ,Δy' i , Δz' i ); Using the second distortion vector and the translation distance ξ i The formula for calculating translation compensation is as follows: The second distortion vector is rotated and compensated using the rotation angle θ, and the calculation formula is as follows: The second distortion vector is scaled and compensated using the scaling factor λ, and the calculation formula is as follows: Finally, the first compensation vector (L″) is obtained. x , L″ y , L″ z ).

7. The lidar-based tunnel detection method according to claim 1, characterized in that, In step (7), the specific steps for calculating the matching degree E between the first compensation vector and the reference vector include: Feature points are extracted from the first and second reference points on the known reference vector to obtain reference point features; The first compensation point on the first compensation vector is matched with the known first reference point and second reference point using feature points to calculate the Euclidean distance between the first compensation point and the matched first reference point or second reference point. The Euclidean distance is the matching degree E between the first compensation point and the first reference point or second reference point.

8. A tunnel detection device based on lidar, characterized in that, include: Reference point setting module: used to establish the tunnel coordinate system and set the first and second reference points within the tunnel; Distortion point acquisition module: used to obtain point cloud by moving and scanning with lidar vehicle, acquire the first distortion point corresponding to the first reference point and the second distortion point corresponding to the second reference point in the point cloud, and record the timestamp; Distortion error calculation module: used to calculate a reference vector using the first reference point and the second reference point, and to calculate a first distortion vector using the first distortion point and the second distortion point, and then to obtain the distortion error D using the reference vector and the first distortion vector. i ; Motion data acquisition module: used to record the motion state data of the lidar vehicle at each time corresponding to the timestamp, based on the tunnel coordinate system and the timestamp; Model training module: used to train a distortion detection model by using the first reference point, the second reference point, the first distortion point, the second distortion point, the motion state data and the distortion error as training data. The distortion detection model is used to determine the distortion error of the lidar vehicle. Distortion compensation module: used to obtain the third distortion point in the point cloud, obtain the second distortion vector using the third distortion point, detect the distortion error of the second distortion vector using the distortion detection model and perform distortion compensation to obtain the first compensation vector; Iterative optimization module: used to set the matching degree threshold Q, calculate the matching degree E between the first compensation vector and the reference vector. If the matching degree E reaches the matching degree threshold Q, the distortion compensation is completed; otherwise, the first compensation vector is set as the new second distortion vector, and the process is returned to the distortion compensation module for iterative optimization.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the lidar-based tunnel detection method as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the lidar-based tunneling detection method as described in any one of claims 1-7.

Images