Laser SLAM map construction method and system based on stereo structure scene information

Abstract

The invention provides a laser SLAM map construction method and system based on three-dimensional structure scene information, and the method comprises the steps: calculating a current gravity direction through the measurement data of IMU equipment, and providing an absolute direction reference for the initialization of a map; according to the method, the feature lines and the feature surfaces are utilized to construct the residual equation, the information of the three-dimensional structures such as the ground and the wall surfaces in the initialized map is extracted, corresponding constraint conditions are applied according to the geometrical characteristics of ground flatness and wall body perpendicularity, and the residual equation of the plane and the vertical surface is constructed; the method comprises the following steps of: constructing a multi-dimensional residual equation, optimizing by using the constructed multi-dimensional residual equation, and further correcting data subjected to laser radar deviation correction according to an optimized pose to construct or adjust an initial map point, so that the initial map point is closer to a real environment in a geometric structure, error accumulation caused by sensor noise is reduced, and the accuracy of laser radar deviation correction is improved. And the precision and consistency of map construction are improved.

Description

Technical Field

[0001] The present invention relates to the field of real-time positioning and map construction, and in particular to the technical field of real-time positioning and map construction in an indoor three-dimensional environment, and discloses a laser SLAM map construction method and system based on three-dimensional structure scene information. Background Art

[0002] In many application scenarios, such as robot navigation, autonomous driving, augmented reality (AR), and indoor positioning systems, SLAM (Simultaneous Localization and Mapping) technology is widely used to achieve self-localization and map construction of devices in unknown environments. However, SLAM technology still faces many challenges in practical applications, especially in the absence of global navigation satellite system (GNSS) support.

[0003] Currently, most SLAM mapping solutions focus on the control of cumulative errors and the relative accuracy of data. These methods rely on the relative position information provided by devices such as inertial measurement units (IMUs) and visual sensors to build maps through continuous relative displacement and rotation information. However, this reliance leads to problems such as lack of absolute geometric constraints and insufficient utilization of environmental geometric information, resulting in low accuracy of the constructed maps. Summary of the invention

[0004] The present disclosure at least provides a laser SLAM map construction method and system based on stereoscopic structure scene information to improve the accuracy of the constructed three-dimensional map.

[0005] According to one aspect of the present disclosure, a laser SLAM map construction method based on stereoscopic structure scene information is provided, comprising:

[0006] S110, using a laser radar to scan data of a three-dimensional space to obtain laser data, and obtaining IMU data of an IMU device bound to the laser radar; performing spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data;

[0007] S120, extracting data of the IMU device at a stationary moment from the target IMU data, and determining the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data;

[0008] S130, for each frame of target laser data, using target IMU data with the same time as the current frame of target laser data, determining the attitude data of the IMU device; according to the preset installation position and installation attitude of the laser radar and the IMU device, converting the current frame of target laser data into the device coordinate system of the IMU device to obtain a frame of adjustment laser data; performing correction processing on the adjustment laser data according to the attitude data of the IMU device to obtain a frame of dedistorted laser data;

[0009] S140, constructing an initial map point cloud using the first frame of dedistorted laser data;

[0010] S150, for each frame of dedistorted laser data after the first frame of dedistorted laser data, extract the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame; determine the main direction of the characteristic line and the normal of the characteristic surface; construct the line characteristic residual equation using the characteristic line and the main direction, and construct the surface characteristic residual equation using the characteristic surface and the normal; optimize the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame to obtain the first optimized posture data of the dedistorted laser data of the current frame; register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and merge the registered laser data into the initial map point cloud; optimize the device parameters of the IMU device, and based on the optimized device parameters of the IMU device, cyclically execute steps S130 and S150 to correct and register the next frame of target laser data of the current frame and perform registration laser data merging operations on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions;

[0011] S160, extracting ground points and facade points from the initial map point cloud that meets preset conditions; fitting the plane equation of the ground using at least some of the ground points, and setting a first constraint condition of ground height consistency using the plane equation of the ground; clustering using at least some of the facade points, and fitting using each clustering result to obtain plane equations of multiple facades, and setting a second constraint condition that the facade is perpendicular to the ground using the plane equation of the facade; wherein the ground is determined according to the gravity direction;

[0012] S150, for each frame of dedistorted laser data after the first frame of dedistorted laser data, construct a ground height residual equation using the plane equation of the ground; and construct a vertical residual equation of the facade using the plane equation of the facade;

[0013] S160, determining second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data by using the surface feature residual equation, line feature residual equation, ground height residual equation, and vertical surface residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data;

[0014] S170, using the second optimized posture data of each frame of dedistorted laser data to respectively determine the mapping point cloud data of each frame of dedistorted laser data; and using the mapping point cloud data of each frame to construct a three-dimensional map.

[0015] In a possible implementation manner, constructing a ground height residual equation using a plane equation of the ground includes:

[0016] The ground height residual equation is constructed using the following formula:

[0017]

[0018] Among them, R i is the posture data of the dedistorted laser data of the i-th frame to be optimized, n is the number of frames before the current frame, P f z is the elevation of the dedistorted laser data point f in the i-th frame, q is the number of dedistorted laser data points in the i-th frame, and H is the fitted elevation of the ground; wherein H is determined based on the plane equation of the ground.

[0019] In a possible implementation, constructing a vertical residual equation of a facade using a plane equation of the facade includes:

[0020] The vertical residual equation of the facade is constructed using the following formula:

[0021]

[0022] Among them, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, A j ,B j ,C j ,D j is the facade parameter of the jth facade of the i-th frame of dedistorted laser data, m is the number of facades of the i-th frame of dedistorted laser data, P i x,P i y,P i z is the point coordinate in the i-th frame of dedistorted laser data.

[0023] In a possible implementation, extracting ground points and elevation points from an initial map point cloud that meets preset conditions includes:

[0024] Constructing a KD tree of the initial map point cloud;

[0025] For each point in the initial map point cloud, use the KD tree to search for neighboring points of the current point, and use the neighboring points obtained by the search to determine the normal vector of the current point;

[0026] According to the normal vector of each point, ground points and facade points are extracted from the initial map point cloud.

[0027] In a possible implementation, constructing a line characteristic residual equation using the characteristic line and the main direction includes:

[0028] The line characteristic residual equation is constructed using the following formula:

[0029]

[0030] Among them, L k is the line parameter of the feature line k, the line parameter includes the main direction, h is the number of feature lines extracted from the dedistorted laser data of the i-th frame; R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P k is the line feature of the current feature line extracted.

[0031] In a possible implementation, the constructing a surface feature residual equation using the feature surface and the normal direction includes:

[0032] The surface feature residual equation is constructed using the following formula:

[0033]

[0034] Where d is the number of feature surfaces extracted from the dedistorted laser data of the i-th frame, N l is the plane parameter of the feature surface l, the surface parameter includes the normal, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P l is the surface feature of the current feature surface extracted.

[0035] In a possible implementation manner, determining the main direction of the characteristic line and the normal direction of the characteristic surface includes:

[0036] For the feature line, a neighborhood search is performed in the current initial map point cloud, and a principal component analysis is performed on the points on the current feature line using the first neighborhood point obtained by the search, and the direction corresponding to the maximum eigenvalue is taken as the main direction of the corresponding feature line;

[0037] For the feature surface, a neighborhood search is performed in the current initial map point cloud, and the second neighborhood points obtained by the search are used to perform principal component analysis on the points on the current feature surface, and the direction corresponding to the minimum eigenvalue is used as the normal of the corresponding feature surface.

[0038] In a possible implementation, determining the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data includes:

[0039] The Kalman filter is used to filter the data extracted when the IMU device is stationary, and the filtered data is used to determine the gravity direction in the device coordinate system corresponding to the IMU device.

[0040] According to another aspect of the present disclosure, a laser SLAM map construction system based on stereoscopic structure scene information is provided, comprising:

[0041] LiDAR is used to scan data in three-dimensional space and obtain laser data;

[0042] IMU device, bound to the lidar, is used to collect IMU data;

[0043] A data analysis and synchronization module is used to perform spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data;

[0044] A gravity direction determination module, used to extract data of the IMU device at a stationary moment from the target IMU data, and determine the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data;

[0045] The laser radar data correction module is used to determine the posture data of the IMU device for each frame of target laser data by using the target IMU data with the same time as the current frame of target laser data; according to the preset installation position and installation posture of the laser radar and the IMU device, the current frame of target laser data is converted into the device coordinate system of the IMU device to obtain a frame of adjustment laser data; according to the posture data of the IMU device, the adjustment laser data is corrected to obtain a frame of dedistorted laser data;

[0046] A map initialization module is used to construct an initial map point cloud using the first frame of dedistorted laser data;

[0047] A registration module is used to extract the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame for each frame of dedistorted laser data after the first frame of dedistorted laser data; determine the main direction of the characteristic line and the normal of the characteristic surface; construct a line characteristic residual equation using the characteristic line and the main direction, and construct a surface characteristic residual equation using the characteristic surface and the normal; optimize the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame to obtain the first optimized posture data of the dedistorted laser data of the current frame; register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and merge the registered laser data into the initial map point cloud; optimize the device parameters of the IMU device, and based on the optimized device parameters of the IMU device, use the laser radar data correction module and the registration module to cyclically perform correction and registration of the target laser data and registration laser data merging operations on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions;

[0048] A geometric feature extraction and constraint module, used to extract ground points and facade points from an initial map point cloud that meets preset conditions; use at least some of the ground points to fit the plane equation of the ground, and use the plane equation of the ground to set a first constraint condition for ground height consistency; use at least some of the facade points to cluster, and use each clustering result to fit to obtain plane equations of multiple facades, and use the plane equation of the facade to set a second constraint condition that the facade is perpendicular to the ground; wherein the ground is determined according to the gravity direction; and, for each frame of dedistorted laser data after the first frame of dedistorted laser data, use the plane equation of the ground to construct a ground height residual equation; use the plane equation of the facade to construct a facade vertical residual equation;

[0049] The optimization module is used to determine the second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data by using the surface feature residual equation, line feature residual equation, ground height residual equation, and vertical residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data; and to determine the mapping point cloud data of each frame of dedistorted laser data by using the second optimized posture data of each frame of dedistorted laser data; and to construct a three-dimensional map by using the mapping point cloud data of each frame.

[0050] The disclosed laser SLAM map construction method and system based on three-dimensional structure scene information use the measurement data of the IMU device to calculate the current gravity direction, providing an absolute direction reference for map initialization. In addition to using feature lines and feature surfaces to construct residual equations, the disclosed method also extracts three-dimensional structure information such as the ground and walls in the initialization map, and imposes corresponding constraints based on the geometric characteristics of the flatness of the ground and the verticality of the walls to construct residual equations for planes and elevations, and uses the constructed residual equations in multiple dimensions for optimization. The laser radar corrected data is further corrected based on the optimized posture to construct or adjust the initial map points so that they are closer to the real environment in terms of geometry, thereby reducing the error accumulation caused by sensor noise and improving the accuracy and consistency of map construction.

[0051] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0053] Figure 1 is a flowchart of a laser SLAM map construction method based on three-dimensional structure scene information according to an embodiment of the present disclosure;

[0054] Figure 2 It is a structural schematic diagram of a laser SLAM map construction system based on three-dimensional structure scene information according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0055] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0056] At present, most SLAM mapping solutions have the following problems:

[0057] Cumulative error problem: During the initialization phase, the data from sensors such as IMU devices contain noise and bias, lacking the support of absolute reference information, which easily leads to cumulative errors in pose estimation. These errors gradually accumulate over time, affecting the overall accuracy and consistency of the map.

[0058] Lack of absolute geometric constraints: When initializing the map, traditional SLAM methods mainly focus on the relative consistency of the data, but ignore the geometric structure information in the environment. For example, the indoor ground in the real world should usually be flat, and the wall should be perpendicular to the ground, but these geometric constraints are not fully utilized in the relative coordinate system of traditional SLAM, resulting in the generated map may deviate from the real environment in terms of geometry.

[0059] Environmental geometry information is not fully utilized: Existing SLAM algorithms often ignore significant geometric features in the environment during the initialization phase, such as ground planes, vertical walls, etc. These geometric features are stable and predictable in space and can be used as important constraints to correct pose estimation, thereby improving the accuracy of map construction.

[0060] Although a lot of research has been done in the field of SLAM technology to optimize error accumulation and accurately calculate relative position information, there is relatively little research on how to organically integrate the geometric structure information of the environment into the SLAM initialization process. Existing methods are still insufficient in using geometric constraints to improve map construction accuracy, especially in the absence of absolute positioning information. How to effectively use the geometric characteristics of the environment to correct pose estimation is still a difficult problem that needs to be solved.

[0061] In response to the above problems, the present invention provides a laser SLAM map construction method and system based on three-dimensional structure scene information. The present invention uses the measurement data of the IMU device to calculate the current gravity direction to provide an absolute direction reference for map initialization. In addition to using feature lines and feature surfaces to construct residual equations, the present invention also extracts three-dimensional structure information such as the ground and walls in the initialization map, and imposes corresponding constraints based on the geometric characteristics of the flatness of the ground and the verticality of the walls to construct residual equations of the plane and the facade, and uses the constructed residual equations in multiple dimensions for optimization. The laser radar corrected data is further corrected according to the optimized posture to construct or adjust the initial map points so that they are closer to the real environment in terms of geometry, thereby reducing the error accumulation caused by sensor noise and improving the accuracy and consistency of map construction.

[0062] The technical solution of the present disclosure is described below through specific embodiments.

[0063] like Figure 1 As shown, it is a flowchart of the laser SLAM map construction method based on the three-dimensional structure scene information of this embodiment. The execution subject of this embodiment is a component, device or system with computing capability. Specifically, the laser SLAM map construction method based on the three-dimensional structure scene information of this embodiment includes the following steps:

[0064] S110, using a laser radar to scan data of a three-dimensional space to obtain laser data, using an IMU device to obtain environmental data of the three-dimensional space to obtain IMU data; performing spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data.

[0065] Here, a laser radar is used to collect three-dimensional space, that is, three-dimensional point cloud data in the environment, to obtain aurora data. At the same time, the acceleration and angular velocity information of the device are collected through the IMU device, that is, the above-mentioned IMU data.

[0066] After collecting laser data and IMU data, parse the laser data format, such as point cloud formats such as PCD, and parse the IMU data format (such as accelerometer and gyroscope data). After that, align the timestamps of the laser data and IMU data to ensure that the data at the same time is used for processing in subsequent steps. Use timestamp alignment algorithms, such as interpolation or interpolation methods, to handle the sampling frequency differences of different sensors, and perform posture calibration of the LiDAR and IMU devices in advance, that is, calibrate the relative relationship between the posture of the LiDAR and the posture of the IMU device.

[0067] S120, extracting data of the IMU device at a stationary moment from the target IMU data, and determining the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data.

[0068] When determining the direction of gravity, the data extracted when the IMU device is stationary is filtered using a Kalman filter, and the filtered data is used to determine the direction of gravity in the device coordinate system corresponding to the IMU device.

[0069] Specifically:

[0070] Identify the moment when the IMU device is stationary and extract the IMU data at that moment (accelerometer data mainly reflects the direction of gravity, and gyroscope data reflects the angular velocity of the device). Use the Kalman filter to process the IMU data at the stationary moment to estimate the current direction of gravity. The Kalman filter can fuse the IMU data, filter out noise, and provide a stable estimate of the direction of gravity. Based on the output of the Kalman filter, determine the direction of gravity in the device coordinate system to provide a reference for subsequent ground horizontal and wall vertical constraints.

[0071] S130. For each frame of target laser data, the target IMU data with the same time as the current frame of target laser data is used to determine the posture data of the IMU device; according to the preset installation position and installation posture of the laser radar and the IMU device, the current frame of target laser data is converted into the device coordinate system to obtain a frame of adjustment laser data; according to the posture data of the IMU device, the adjustment laser data is corrected to obtain a frame of dedistorted laser data.

[0072] Specifically, the attitude information provided by the IMU device (estimated by the Kalman filter) and the device parameters of the IMU device itself, such as the IMU device acceleration I a and angular velocity noise I g , calculate the posture of the IMU device in space (including roll angle, pitch angle and yaw angle).

[0073] The point cloud data collected by the laser radar, that is, the aligned target laser data, is converted from the radar coordinate system to the device coordinate system according to the conversion relationship between the coordinate systems of the laser radar and the IMU device. After that, the attitude data of the IMU device is estimated, and the rotation and translation transformation is performed to correct the laser attitude motion distortion and obtain the dedistorted laser data.

[0074] S140, constructing an initial map point cloud using the first frame of dedistorted laser data.

[0075] S150, registering the dedistorted laser data and performing a registration laser data merging operation on the initial map point cloud: for each frame of dedistorted laser data after the first frame of dedistorted laser data, extracting the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame; determining the main direction of the characteristic lines and the normal of the characteristic surfaces; constructing a line characteristic residual equation using the characteristic lines and the main direction, and constructing a surface characteristic residual equation using the characteristic surfaces and the normal; optimizing the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame; to obtain first optimized posture data of the dedistorted laser data of the current frame; to register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and to merge the registered laser data into the initial map point cloud; to optimize the device parameters of the IMU device, and based on the device parameters of the optimized IMU device, to loop through steps S130 and S150 to perform deviation correction and registration on the next frame of target laser data of the current frame and to perform registration laser data merging operations on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions.

[0076] In some embodiments, the above-mentioned determination of the main direction of the characteristic line and the normal direction of the characteristic surface can be implemented by the following steps:

[0077] For the feature line, a neighborhood search is performed in the current initial map point cloud, and a plurality of first neighborhood points obtained by the search are used to perform principal component analysis on the points on the current feature line, and the direction corresponding to the maximum eigenvalue is used as the main direction of the corresponding feature line; for the feature surface, a neighborhood search is performed in the current initial map point cloud, and a plurality of second neighborhood points obtained by the search are used to perform principal component analysis on the points on the current feature surface, and the direction corresponding to the minimum eigenvalue is used as the normal of the corresponding feature surface.

[0078] The line characteristic residual equation is constructed using the following formula:

[0079]

[0080] Among them, L k is the line parameter of the feature line k, the line parameter includes the main direction, h is the number of feature lines extracted from the dedistorted laser data of the i-th frame; R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P k is the line feature of the current feature line extracted.

[0081] The surface feature residual equation is constructed using the following formula:

[0082]

[0083] Where d is the number of feature surfaces extracted from the dedistorted laser data of the i-th frame, N l is the plane parameter of the feature surface l, the surface parameter includes the normal, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P l is the surface feature of the current feature surface extracted.

[0084] The registered laser data are merged into the initial map point cloud to gradually expand the coverage of the map.

[0085] During the map update process, an optimization algorithm (such as nonlinear least squares optimization) is used to adjust the device parameters of the IMU device, such as the acceleration I of the IMU device. a and angular velocity noise I g , correct sensor bias and improve the accuracy of attitude estimation.

[0086] Apply the optimized device parameters and repeat the above correction, registration and laser data merging operations on the initial map point cloud. Specifically, in the continuously released data stream, for each frame of target laser data, the acceleration I of the optimized IMU device is a and angular velocity noise I g Applied to the correction of the next frame of target laser data, the IMU is used to calculate the posture of the IMU device in space (including roll angle, pitch angle and yaw angle) based on the latest device parameters and the acceleration and angular velocity parameters of the IMU at the same time. The optimized posture of the radar at the same time is calculated based on the calibration parameters, and the map is continuously aligned and optimized to gradually build a high-precision initialization map point cloud.

[0087] The initial map point cloud meets the preset conditions, that is, as the initial map point cloud gradually accumulates, the map can reflect the key structural features of the current environment, such as the ground plane and the building facades. In specific practice, the number of frames of target laser data or dedistorted laser data processed when the preset conditions are met can be determined based on experience, and when the number of data frames merged by the initial map point cloud exceeds the number of frames, it is determined that the preset conditions are met.

[0088] S160. Extract ground points and facade points from an initial map point cloud that meets preset conditions; use at least some of the ground points to fit a plane equation of the ground, and use the plane equation of the ground to set a first constraint condition for ground height consistency; use at least some of the facade points to cluster, and use each clustering result to fit to obtain plane equations of multiple facades, and use the plane equation of the facade to set a second constraint condition that the facade is perpendicular to the ground; wherein the ground is determined based on the gravity direction.

[0089] In some embodiments, the ground points and facade points extracted from the initial map point cloud that meet the preset conditions can be determined by the following steps:

[0090] Construct a KD tree of the initial map point cloud; for each point in the initial map point cloud, use the KD tree to search for neighboring points of the current point, and use the searched neighboring points to determine the normal vector of the current point; according to the normal vectors of each point, extract ground points and facade points from the initial map point cloud.

[0091] The above steps construct a KD tree index for the initial map point cloud data, which can speed up the neighbor search process. Then, for each map point, the normal vector is calculated using the neighboring points. The normal vector is mainly determined by the PCA (Principal Component Analysis) method. Then the normal vector classification processing is performed. According to the direction of the normal vector, the ground points and facade points are extracted from the map points: the ground point is a spatial point with a normal pointing upward (parallel to the direction of gravity) or close to the horizontal, and the facade point is a spatial point with a normal perpendicular to the direction of gravity, reflecting the vertical characteristics of the wall.

[0092] Fitting the plane equation of the ground: Filter out the point set that meets the plane characteristics from the classified ground points, perform plane fitting on the ground point set, and use the RANSAC algorithm or the least squares method to determine the plane equation of the ground. The ground equation can be obtained by P z =H to express, P z Indicates the height of the current ground point, and H is the height of the fitted plane.

[0093] Add ground constraint settings for ground points: According to the plane equation of the fitted ground, set the constraint condition for ground height consistency (the first constraint condition), that is, the height of all ground points should be consistent within a certain error range, reflecting the flatness of the ground.

[0094] Fitting the plane equation of the facade: Filter out the point set that meets the vertical characteristics from the classified facade points, then cluster them through clustering analysis (such as DBSCAN algorithm), and then perform plane fitting on the facade point set of the clustering results, and use the RANSAC algorithm or least squares method to determine the plane equation of the facade.

[0095] Facade clustering segmentation: Through clustering analysis (such as DBSCAN algorithm), the facade points obtained by clustering are further segmented into different facades, reflecting different walls or building structures. The plane equation of a single facade can be expressed as A*P x +B*P y +C*P z =D. A, B, C, D are the parameters of the plane equation, P x , P y , P z are the coordinates of the corresponding elevation point.

[0096] Add facade constraint settings to facade points: According to the plane equation of the fitted facade, set the constraint condition that the facade is perpendicular to the ground (the second constraint condition), that is, the normal of all facade points should be perpendicular to the direction of gravity to ensure the vertical characteristics of the wall.

[0097] S170, for each frame of dedistorted laser data after the first frame of dedistorted laser data, construct a ground height residual equation using the plane equation of the ground; and construct a facade vertical residual equation using the plane equation of the facade.

[0098] The ground height residual equation is constructed using the following formula:

[0099]

[0100] Among them, R i is the posture data of the dedistorted laser data of the i-th frame to be optimized, n is the number of frames before the current frame, P f z is the elevation of the dedistorted laser data point f in the i-th frame, q is the number of dedistorted laser data points in the i-th frame, and H is the fitted elevation of the ground; wherein H is determined based on the plane equation of the ground.

[0101] The vertical residual equation of the facade is constructed using the following formula:

[0102]

[0103] Among them, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, A j ,B j ,C j ,D j is the facade parameter of the jth facade of the i-th frame of dedistorted laser data, m is the number of facades of the i-th frame of dedistorted laser data, P i x,P i y,P i z is the point coordinate in the i-th frame of dedistorted laser data.

[0104] S180, using the surface feature residual equation, line feature residual equation, ground height residual equation, and facade vertical residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data, determine the second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data.

[0105] In the specific implementation, the following constraint equations are constructed using the surface feature residual equation, line feature residual equation, ground height residual equation, and facade vertical residual equation:

[0106]

[0107] where φ all is the preset paradigm. By solving the function, the residuals of the feature lines, feature surfaces, ground constraints, and elevation constraints are optimized to obtain the final pose R that needs to be optimized. i , a nonlinear least squares optimization algorithm can be used for multiple iterations to obtain the optimal posture data of each frame of dedistorted laser data.

[0108] S190, using the second optimized posture data of each frame of dedistorted laser data to respectively determine the mapping point cloud data of each frame of dedistorted laser data; and using the mapping point cloud data of each frame to construct a three-dimensional map.

[0109] Using the second optimized posture data and the dedistorted laser data, the dedistorted laser data is reprojected and transformed, and the formula is as follows:

[0110] X w =R i *X i

[0111] Among them, X w is the w-th frame of the point cloud data, R i is the second optimized posture data of the dedistorted laser data of the i-th frame, X i is the dedistorted laser data of the i-th frame.

[0112] Taking the ground and facade as constraints, the final re-optimized mapping point cloud data can significantly improve the accuracy of the initialized map point cloud. Therefore, the re-converted final mapping point cloud data can be used instead of the initialized map point cloud for subsequent optimization, so that its geometric structure is more consistent with the real environment characteristics, and the final optimized point cloud map has higher accuracy.

[0113] In some embodiments, the above step S150 realizes the registration of subsequent frames after the first frame of dedistorted laser data with the initial map point cloud: for each frame of dedistorted laser data after correction, the residual equation is constructed using the feature lines and feature surfaces in the point cloud, so that the distance from the surface feature point to the fitting plane is minimized, and the distance from the line feature point to the fitting space line is minimized, and the dedistorted laser data after correction is optimized using the least squares method, and it is accurately registered with the initial map point cloud. The specific operations may include:

[0114] 1) Using edge detection algorithm and plane fitting algorithm, feature line points and feature surface points are extracted from the dedistorted laser data after correction as the basic features for subsequent matching and optimization.

[0115] 2) Use the KDTREE search algorithm to perform neighborhood search on the points on the extracted feature lines and feature surfaces in the initial map, use PCA to perform principal component analysis based on the searched neighborhood points, and select the maximum eigenvalue as the main direction of the feature line and the minimum eigenvalue as the normal direction of the feature surface;

[0116] 3) According to the extracted line features and the main direction calculated by the neighborhood points, the residual equation based on the line features can be constructed:

[0117]

[0118] Among them, L k is the line parameter of the feature line k, the line parameter includes the main direction, h is the number of feature lines extracted from the dedistorted laser data of the i-th frame; R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P k is the line feature of the current feature line extracted.

[0119] According to the extracted surface features and the normals calculated from the neighborhood points, the residual equation based on the surface features can be constructed:

[0120]

[0121] Where d is the number of feature surfaces extracted from the dedistorted laser data of the i-th frame, N l is the plane parameter of the feature surface l, the surface parameter includes the normal, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P l is the surface feature of the current feature surface extracted.

[0122] According to the set of line feature residual equations and surface feature residual equations, the nonlinear least squares method is used for iteration, and the residual set can be expressed as:

[0123] The smaller the better, the optimized pose R is finally obtained. i .

[0124] The disclosed solution introduces structured geometric information such as the ground and plane elevation in the environment during the SLAM initialization phase, and optimizes the pose estimation and map construction process through specific geometric constraints. The following are the performance improvements of the disclosed solution:

[0125] Error control: The gravity direction and geometric constraints provided by the IMU device effectively reduce the cumulative error in pose estimation and improve the overall accuracy of the map. Geometric consistency: The ground height consistency and facade vertical constraints ensure that the generated map is highly consistent with the real environment in terms of geometric structure, enhancing the realism and practicality of the map. Optimize stability: The posture parameters are continuously adjusted by the optimization algorithm to improve the stability and reliability of the SLAM solution in complex environments. Enhanced adaptability: The scheme disclosed in this disclosure is applicable to a variety of complex environments, such as multi-story buildings and industrial warehouses, and improves the adaptability of the SLAM solution in different application scenarios. Reduce cumulative errors: In indoor and outdoor areas with rich buildings, the scheme disclosed in this disclosure can reduce the overall cumulative error, for example, in indoor environments, the ground height of the collected data is consistent, and in outdoor areas with rich buildings, the facade is ensured to be perpendicular to the ground, avoiding large tilts in the map data, which is inconsistent with the actual situation. Enhance the fit between the map and the real scene: The map is closer to the real environment in terms of geometric structure, and will not cause large tilts in the data or structural characteristics that are inconsistent with the actual situation, thereby improving the realism and practicality of the map.

[0126] In summary, the solution disclosed in the present invention not only effectively reduces the cumulative error and improves the geometric consistency of the map, but also enhances the adaptability and stability of the solution in complex environments, significantly improves the fit between the map and the real scene, and has significant technical advantages and broad application prospects.

[0127] Based on the same inventive concept, the present disclosure provides a laser SLAM map construction device based on stereoscopic structure scene information, and the steps performed by the components of the device are the same or similar to the above method, so similar parts are not repeated. Figure 2 As shown, the laser SLAM map construction device based on three-dimensional structure scene information of this embodiment includes:

[0128] The laser radar 210 is used to scan data in a three-dimensional space to obtain laser data.

[0129] The IMU device 220 is bound to the laser radar and is used to collect IMU data.

[0130] The data analysis and synchronization module 230 is used to perform spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data.

[0131] The gravity direction determination module 240 is used to extract the data of the IMU device at a stationary moment from the target IMU data, and determine the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data.

[0132] The laser radar data correction module 250 is used to determine the posture data of the IMU device for each frame of target laser data using the target IMU data with the same time as the current frame of target laser data; convert the current frame of target laser data into the device coordinate system according to the preset installation position and installation posture of the laser radar and IMU device to obtain a frame of adjustment laser data; and perform correction processing on the adjustment laser data according to the posture data of the IMU device to obtain a frame of dedistorted laser data.

[0133] The map initialization module 260 is used to construct an initial map point cloud using the first frame of dedistorted laser data.

[0134] The registration module 270 is used to extract the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame for each frame of dedistorted laser data after the first frame of dedistorted laser data; determine the main direction of the characteristic line and the normal of the characteristic surface; construct the line characteristic residual equation using the characteristic line and the main direction, and construct the surface characteristic residual equation using the characteristic surface and the normal; optimize the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame to obtain the first optimized posture data of the dedistorted laser data of the current frame; register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and merge the registered laser data into the initial map point cloud; optimize the device parameters of the IMU device, and based on the optimized device parameters of the IMU device, use the laser radar data correction module and the registration module to cyclically perform the correction and registration of the target laser data and the registration laser data merging operation on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions.

[0135] The geometric feature extraction and constraint module 280 is used to extract ground points and facade points from the initial map point cloud that meets preset conditions; use at least some of the ground points to fit the plane equation of the ground, and use the plane equation of the ground to set a first constraint condition for ground height consistency; use at least some of the facade points to cluster, and use each clustering result to fit to obtain plane equations of multiple facades, and use the plane equation of the facade to set a second constraint condition that the facade is perpendicular to the ground; wherein the ground is determined according to the gravity direction; and, for each frame of dedistorted laser data after the first frame of dedistorted laser data, use the plane equation of the ground to construct a ground height residual equation; use the plane equation of the facade to construct a facade vertical residual equation.

[0136] The optimization module 290 is used to determine the second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data by using the surface feature residual equation, line feature residual equation, ground height residual equation, and facade vertical residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data; and to determine the mapping point cloud data of each frame of dedistorted laser data by using the second optimized posture data of each frame of dedistorted laser data; and to construct a three-dimensional map by using the mapping point cloud data of each frame.

[0137] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0138] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0139] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0140] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0141] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0142] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0143] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0144] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A laser SLAM map construction method based on three-dimensional structure scene information, characterized in that: include: S110, using a laser radar to scan data of a three-dimensional space to obtain laser data, and obtaining IMU data of an IMU device bound to the laser radar; performing spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data; S120, extracting data of the IMU device at a stationary moment from the target IMU data, and determining the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data; S130, for each frame of target laser data, using target IMU data with the same time as the current frame of target laser data, determining the attitude data of the IMU device; according to the preset installation position and installation attitude of the laser radar and the IMU device, converting the current frame of target laser data into the device coordinate system of the IMU device to obtain a frame of adjustment laser data; performing correction processing on the adjustment laser data according to the attitude data of the IMU device to obtain a frame of dedistorted laser data; S140, constructing an initial map point cloud using the first frame of dedistorted laser data; S150, for each frame of dedistorted laser data after the first frame of dedistorted laser data, extract the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame; determine the main direction of the characteristic line and the normal of the characteristic surface; construct the line characteristic residual equation using the characteristic line and the main direction, and construct the surface characteristic residual equation using the characteristic surface and the normal; optimize the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame to obtain the first optimized posture data of the dedistorted laser data of the current frame; register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and merge the registered laser data into the initial map point cloud; optimize the device parameters of the IMU device, and based on the optimized device parameters of the IMU device, cyclically execute steps S130 and S150 to correct and register the next frame of target laser data of the current frame and perform registration laser data merging operations on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions; S160, extracting ground points and elevation points from the initial map point cloud that meets preset conditions; fitting a plane equation of the ground using at least some of the ground points, and setting a first constraint condition of ground height consistency using the plane equation of the ground; Clustering is performed using at least part of the facade points, and fitting is performed using each clustering result to obtain plane equations of a plurality of facades, and the plane equations of the facades are used to set a second constraint condition that the facade is perpendicular to the ground; wherein the ground is determined according to the gravity direction; S170, constructing a ground height residual equation using a plane equation of the ground for each frame of dedistorted laser data after the first frame of dedistorted laser data; Use the plane equation of the facade to construct the vertical residual equation of the facade; S180, determining second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data by using the surface feature residual equation, line feature residual equation, ground height residual equation, and vertical surface residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data; S190, using the second optimized posture data of each frame of dedistorted laser data to respectively determine the mapping point cloud data of each frame of dedistorted laser data; and using the mapping point cloud data of each frame to construct a three-dimensional map.

2. The method according to claim 1, characterized in that The method of constructing a ground height residual equation using a plane equation of the ground includes: The ground height residual equation is constructed using the following formula: Among them, R i is the posture data of the dedistorted laser data of the i-th frame to be optimized, n is the number of frames before the current frame, P f z is the elevation of the dedistorted laser data point f in the i-th frame, q is the number of dedistorted laser data points in the i-th frame, and H is the fitted elevation of the ground; wherein H is determined based on the plane equation of the ground.

3. The method according to claim 1, characterized in that: The method of constructing a vertical residual equation of a facade using the plane equation of the facade includes: The vertical residual equation of the facade is constructed using the following formula: Among them, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, A j ,B j ,C j ,D j is the facade parameter of the jth facade of the i-th frame of dedistorted laser data, m is the number of facades of the i-th frame of dedistorted laser data, P i x,P i y,P i z is the point coordinate in the i-th frame of dedistorted laser data.

4. The method according to claim 1, characterized in that The extracting of ground points and elevation points from the initial map point cloud that meets the preset conditions includes: Constructing a KD tree of the initial map point cloud; For each point in the initial map point cloud, use the KD tree to search for neighboring points of the current point, and use the neighboring points obtained by the search to determine the normal vector of the current point; According to the normal vector of each point, ground points and facade points are extracted from the initial map point cloud.

5. The method according to claim 1, characterized in that The constructing of a line characteristic residual equation using the characteristic line and the main direction includes: The line characteristic residual equation is constructed using the following formula: Among them, L k is the line parameter of the feature line k, the line parameter includes the main direction, h is the number of feature lines extracted from the dedistorted laser data of the i-th frame; R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P k is the line feature of the current feature line extracted.

6. The method according to claim 1, characterized in that The method of constructing a surface characteristic residual equation using the characteristic surface and the normal direction includes: The surface feature residual equation is constructed using the following formula: Where d is the number of feature surfaces extracted from the dedistorted laser data of the i-th frame, N l is the plane parameter of the feature surface l, the surface parameter includes the normal, R i is the posture data of the i-th frame of dedistorted laser data to be optimized, n is the number of frames of the current frame and the frames before the current frame; I i is the installation position and installation posture of the IMU device and the laser radar at the time of collecting the i-th frame of dedistorted laser data, P l is the surface feature of the current feature surface extracted.

7. The method according to claim 1, characterized in that The determining the main direction of the characteristic line and the normal direction of the characteristic surface comprises: For the feature line, a neighborhood search is performed in the current initial map point cloud, and a principal component analysis is performed on the points on the current feature line using the first neighborhood point obtained by the search, and the direction corresponding to the maximum eigenvalue is taken as the main direction of the corresponding feature line; For the feature surface, a neighborhood search is performed in the current initial map point cloud, and the second neighborhood points obtained by the search are used to perform principal component analysis on the points on the current feature surface, and the direction corresponding to the minimum eigenvalue is used as the normal of the corresponding feature surface.

8. The method according to claim 1, characterized in that The determining the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data includes: The Kalman filter is used to filter the data extracted when the IMU device is stationary, and the filtered data is used to determine the gravity direction in the device coordinate system corresponding to the IMU device.

9. A laser SLAM map construction system based on three-dimensional structure scene information, characterized in that: include: LiDAR is used to scan data in three-dimensional space and obtain laser data; IMU device, bound to the lidar, is used to collect IMU data; A data analysis and synchronization module is used to perform spatiotemporal synchronization processing on the laser data and the IMU data to obtain target laser data and target IMU data; A gravity direction determination module, used to extract data of the IMU device at a stationary moment from the target IMU data, and determine the gravity direction in the device coordinate system corresponding to the IMU device based on the extracted data; The laser radar data correction module is used to determine the posture data of the IMU device for each frame of target laser data by using the target IMU data with the same time as the current frame of target laser data; according to the preset installation position and installation posture of the laser radar and the IMU device, the current frame of target laser data is converted into the device coordinate system of the IMU device to obtain a frame of adjustment laser data; according to the posture data of the IMU device, the adjustment laser data is corrected to obtain a frame of dedistorted laser data; A map initialization module is used to construct an initial map point cloud using the first frame of dedistorted laser data; A registration module is used to extract the characteristic lines and characteristic surfaces of the dedistorted laser data of the current frame for each frame of dedistorted laser data after the first frame of dedistorted laser data; determine the main direction of the characteristic line and the normal of the characteristic surface; construct a line characteristic residual equation using the characteristic line and the main direction, and construct a surface characteristic residual equation using the characteristic surface and the normal; optimize the line characteristic residual equation and the surface characteristic residual equation corresponding to the dedistorted laser data of the current frame and the frames before the current frame to obtain the first optimized posture data of the dedistorted laser data of the current frame; register the dedistorted laser data of the current frame using the first optimized posture data to obtain a frame of registered laser data, and merge the registered laser data into the initial map point cloud; optimize the device parameters of the IMU device, and based on the optimized device parameters of the IMU device, use the laser radar data correction module and the registration module to cyclically perform correction and registration of the target laser data and registration laser data merging operations on the initial map point cloud until the initial map point cloud obtained by the merging operation meets the preset conditions; A geometric feature extraction and constraint module is used to extract ground points and elevation points from an initial map point cloud that meets preset conditions; use at least some of the ground points to fit a plane equation of the ground, and use the plane equation of the ground to set a first constraint condition for ground height consistency; Clustering is performed using at least part of the facade points, and fitting is performed using each clustering result to obtain plane equations of multiple facades, and the plane equations of the facades are used to set a second constraint condition that the facade is perpendicular to the ground; wherein the ground is determined according to the gravity direction; and for each frame of dedistorted laser data after the first frame of dedistorted laser data, the plane equation of the ground is used to construct a residual equation for ground height; and the plane equation of the facade is used to construct a residual equation for verticality of the facade; The optimization module is used to determine the second optimized posture data of each frame of dedistorted laser data after the first frame of dedistorted laser data by using the surface feature residual equation, line feature residual equation, ground height residual equation, and vertical residual equation of each frame of dedistorted laser data after the first frame of dedistorted laser data; and to determine the mapping point cloud data of each frame of dedistorted laser data by using the second optimized posture data of each frame of dedistorted laser data; and to construct a three-dimensional map by using the mapping point cloud data of each frame.

Images