Method and system for simultaneous 2D positioning and 2D map creation using a 3D scanner

Abstract

A method for simultaneous 2D localization and 2D mapping using a 3D scanner (120) is described, wherein a surrounding environment (300) is scanned with the aid of the 3D scanner (120) in order to generate a three-dimensional representation of the surrounding environment (300) in the form of a 3D point cloud (320), the 3D point cloud consisting of a plurality of scan points (321). Subsequently, a two-dimensional representation of the surrounding environment (300) in the form of a 2D point cloud is generated from the 3D point cloud (320). Finally, the 2D point cloud is fed to a 2D-SLAM algorithm in order to generate a map of the surrounding environment (300) and simultaneously determine the current position (480) of the 3D scanner (120) within the map.

Description

Technical Field

[0001] The present invention relates to a method for simultaneous 2D positioning and 2D map creation using a 3D scanner and also to a corresponding system for carrying out the method. Background Art

[0002] Autonomous vehicles rely on continuous determination of their position or orientation. In addition to methods that rely on detailed maps of the surrounding environment for localization, methods are also known that can localize in unknown or constantly changing environments. Therefore, localization methods that also map the surrounding environment are used, particularly in mobile robots or autonomous vehicles (such as autonomous shuttles for transporting people or so-called AVGs (Automatic Guided Vehicles) for transporting containers in ports). In these so-called SLAM (Simultaneous Localization and Mapping) technologies, localization and mapping can interact continuously, enabling infrastructure-free localization. SLAM technology, particularly when using 2D lidar systems, is highly advanced. 2D lidar SLAM assumes a flat world in which structures are primarily two-dimensional. Therefore, these structures can be detected with sufficient accuracy by 2D laser scanners that perform level measurements at a defined height. While this method works well on flat surfaces (e.g., in enclosed spaces), it is generally not suitable for the 2D assumption of a flat world in outdoor areas. The ground plane or road is not necessarily flat, and there are also many structures that are not purely vertical (such as hills or bushes). Therefore, standard 2D lidar SLAM cannot provide the necessary reliability and accuracy in outdoor areas. An alternative is to use a 3D laser scanner and develop a 3D-SLAM system. However, for such a 3D-SLAM system, all six degrees of freedom (DOF) of motion (x, y, z, roll, pitch, yaw) must be determined. Therefore, compared with a 2D-SLAM system that only needs to determine three degrees of freedom (x, y, yaw), a significantly higher complexity is caused in the 3D-SLAM system. Due to the complex algorithm, the computational overhead and computing time will increase. Summary of the Invention

[0003] The object of the present invention is therefore to provide a method for reducing the computational overhead when using a 3D scanner. This object is achieved by a method for simultaneous 2D localization and 2D mapping when using a 3D scanner. Furthermore, this object is achieved by a system and a vehicle.

[0004] According to the present invention, a method for simultaneously performing 2D positioning and 2D mapping using a 3D scanner is provided, in which the surrounding environment is scanned with the aid of a 3D scanner in order to generate a three-dimensional representation of the surrounding environment in the form of a 3D point cloud. Next, a two-dimensional representation of the surrounding environment in the form of a 2D point cloud is generated from the 3D point cloud. Finally, the 2D point cloud is provided to a 2D-SLAM algorithm in order to generate a map of the surrounding environment and simultaneously determine the position (Lage) of the 3D scanner in the map. By converting the 3D point cloud into a 2D point cloud, the amount of data to be processed can be significantly reduced. This makes it possible to use a 2D-SLAM algorithm that allows positioning and simultaneous mapping of the environment with significantly less computing overhead. Therefore, the corresponding method can also be used for vehicles with relatively low computing power.

[0005] One embodiment provides for compensating for the movements of the 3D scanner during scanning. To do this, odometry information is first determined, representing the displacement of the scanner in six degrees of freedom. Subsequently, for each measurement result of the 3D point cloud, the position of the 3D scanner during the measurement of the corresponding point is determined based on the odometry information, and the measured point is then determined from this. These measures enable effective correction of 3D point clouds distorted by the movement of the 3D scanner during the scanning process.

[0006] In another embodiment, the distance information is generated by fusing the measurement data of the vehicle's odometry with the measurement data of the inertial measurement unit. This makes it possible to determine the displacement of the 3D scanner in all six degrees of freedom in a particularly simple manner.

[0007] In another embodiment, the 3D point cloud is transformed into a coordinate system whose z-axis is aligned parallel to the direction of gravity, so that the origin corresponds to the selected time. The 3D point cloud is transformed into a coordinate system whose z-axis is aligned parallel to the direction of gravity. These measures achieve an orientation of the 3D point cloud that is independent of any tilt of the 3D scanner that may have occurred at the time of measurement.

[0008] Another embodiment provides for extracting vertical structures from a 3D point cloud by estimating a normal for each point in the 3D point cloud and subsequently removing from the 3D point cloud points whose normal component in the vertical direction is greater than a predefined threshold. By removing points representing horizontal or inclined surfaces of the scanned object, the amount of measurement data to be processed can be significantly reduced. This significantly reduces the computational overhead required to execute the 2D-SLAM method.

[0009] In another embodiment, the 3D point cloud is projected into a two-dimensional coordinate system by removing the vertical component for each point of the 3D point cloud. By projecting the 3D point cloud onto a two-dimensional plane, the measurement data can be prepared for a 2D-SLAM algorithm.

[0010] Another embodiment provides for reducing the number of points in the 3D point cloud using voxel filters and / or regular downsampling. The resulting reduction in measurement data can significantly reduce the computational overhead required for executing 2D SLAM methods. The use of voxel filters allows, for example, the prioritization of specific spatial regions (e.g., frontal regions or regions at a specific height). Regular downsampling, in contrast, is a particularly simple data reduction method.

[0011] In another embodiment, a known starting position is first initialized for positioning. A SLAM graph is then established with the aid of an initial node and a corresponding unary factor, which encodes the initial position. Next, for each current scan, the current position is predicted with the aid of the currently acquired distance information. Scan matching is then performed as an initial estimate with the aid of the predicted current position. A new node is then generated in the SLAM graph, which has a binary edge to the previous node and has a scan match as a unary edge.

[0012] Furthermore, a system for simultaneous 2D localization and 2D mapping based on a 3D scanner is provided. The system is configured to generate a three-dimensional representation of the surroundings in the form of a 3D point cloud and includes a control device for simultaneous 2D localization and 2D mapping. The control device is configured to generate a two-dimensional representation of the surroundings in the form of a 2D point cloud from a 3D point cloud. Furthermore, the control device is configured to provide the 2D point cloud to a 2D-SLAM algorithm in order to generate a two-dimensional map of the surroundings and simultaneously determine the current position of the 3D scanner within the map. By converting the 3D point cloud into a 2D point cloud, the amount of data to be processed can be significantly reduced. This enables the use of a 2D-SLAM algorithm that can perform localization and simultaneously map the environment with significantly less computational effort. Consequently, the corresponding method can also be used for vehicles with relatively low computing power.

[0013] In one embodiment, the system further includes a distance measuring device for determining distance information representing the displacement of the 3D scanner in six degrees of freedom during the scanning process. The control unit is designed to compensate for the movement of the 3D scanner. The control unit is further designed to determine the current position of the 3D scanner for each scan point of the 3D point cloud based on the distance information and subsequently determine the measurement point based thereon. These measures enable effective correction of 3D point clouds distorted by the movement of the 3D scanner during the scanning process.

[0014] In another embodiment, the control device is also configured to transform the 3D point cloud into a coordinate system whose z-axis is oriented parallel to the direction of gravity so that the origin corresponds to a selected moment in time, and these measures enable an orientation of the 3D point cloud to be achieved that is independent of any tilt that may occur in the 3D scanner at the moment of measurement.

[0015] In another embodiment, the control device is further configured to extract vertical structures from the 3D point cloud by estimating a normal for each point in the 3D point cloud and subsequently removing those points in the 3D point cloud whose normal component in the vertical direction is greater than a predetermined threshold. By removing points representing horizontal or inclined surfaces of the scanned object, the amount of measurement data to be processed can be significantly reduced. This significantly reduces the computational overhead required to execute the 2D-SLAM method.

[0016] Finally, a vehicle with a corresponding system is provided. This vehicle can create a map of its surroundings with relatively little computational effort and simultaneously determine its position within the map. This allows for particularly rapid vehicle orientation even in surroundings unknown to the vehicle, thereby increasing driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be described in more detail below with reference to the accompanying drawings.

[0018] Figure 1 schematically illustrates a 3D scanner when scanning its surroundings;

[0019] Figure 2 A schematic flow chart is shown for explaining the method according to the present invention;

[0020] Figure 3 A block diagram of a system according to the invention is shown, comprising a sensor device including a 3D scanner and a distance meter, and a control device;

[0021] Figure 4 Schematically showing the Figure 3 The system in the vehicle includes a 3D scanner and other sensors. DETAILED DESCRIPTION

[0022] The approach described herein enables the use of sensor data from a 3D scanner to create a map and simultaneously perform localization using a 2D-SLAM system. To this end, the 3D scanner's sensor data is adapted through a series of preprocessing steps and the use of additional distance-measuring sensors so that it can be used by the 2D-SLAM system. Each 3D scan is first motion-compensated and then placed in a standardized orientation. Vertical structures are then extracted from the 3D scans and used to generate a 2D point cloud. This 2D point cloud is then incorporated into a 2D LiDAR-SLAM system. For example, the entire system, consisting of sensors, data processing, and actuators, can be used to navigate an autonomous robot (e.g., an autonomous shuttle or autonomous vehicle). To this end, the invention described herein provides the vehicle's current position and orientation at all times, which are then used as the basis for path planning, execution, and ultimately, the generation of engine commands. However, in principle, applications are also conceivable in which localization is performed passively within another (e.g., manually) controlled system, while localization is used for other purposes (e.g., condition monitoring).

[0023] A 3D laser scanner is provided as the 3D scanner, which is in particular designed, for example, in the form of a rotating laser radar. Figure 1Such a rotating lidar system 120 is schematically shown in FIG. Lidar system 120 includes a housing in which a laser device 121 and a receiving device 122 are arranged. Laser device 121 includes at least one laser source for generating laser radiation in the form of a transmission laser beam 123. In contrast, receiving device 122 includes at least one detector for detecting laser radiation reflected from an object 310 in surroundings 300, which strikes receiving device 123 in the form of a reception laser beam 124. Lidar system 120 also typically includes a control unit 123 for controlling reading device 121 and receiving device 122, which is also arranged in the same housing in this example. In a typical rotating 3D laser scanner 120, laser device 121 simultaneously generates multiple laser beams 123, each of which scans different scanning points 321 of surroundings 300. Scanning points 321 detected simultaneously by 3D laser scanner 120 are typically arranged in a columnar pattern above one another. By means of a rotational motion 126 about a (typically vertical) rotation axis 127 (parallel to the Z axis), the lidar system 120 continuously scans a defined angular range of the surroundings 300 during each scanning cycle. The points 321 scanned during the entire scanning cycle form a 3D point cloud 320, which can be used as a three-dimensional representation of the surroundings 300 of the 3D scanner 120 for further processing.

[0024] In conventional approaches, the scanner data in the form of a 3D point cloud 320 are fed directly to a 3D-SLAM algorithm in order to create a three-dimensional map of the surroundings 300 and simultaneously determine its position and orientation within the three-dimensional map. In contrast, in the method according to the invention, the scanner data are first subjected to a special preprocessing, by which the 3D point cloud 320 is transformed into a 2D point cloud. This 2D point cloud is then fed to a 2D algorithm in order to create a two-dimensional map and simultaneously determine its position and orientation within the two-dimensional map. In order to transform the point cloud, the method provides a series of processing steps, which are explained in detail below. To this end, Figure 2 A simplified flow chart 400 is shown, which illustrates the flow of the method according to the present invention. In this case, in step 410, a 3D point cloud 120 is first generated by scanning the surroundings 300 using a 3D scanner 120. In parallel and essentially independently of this, in method step 420, odometry information is generated, which describes the change in the orientation of the vehicle 200 in six degrees of freedom (X, Y, Z, roll, pitch, yaw). This is done using an odometry measuring device 130, which includes a vehicle odometry device 132 based on wheel rotation (see Figure 4) and an inertial measurement unit (IMU), which detects the movement of vehicle 200 in space based on the accelerations acting on vehicle 200. The output of these steps is a 6-DOF prediction of the vehicle's trajectory. The term "6-DOF" here stands for six degrees of freedom (English: Degrees of Freedom), namely three spatial directions x, y and z and three angles, which describe the roll (English: roll), pitch (English: pitch) and yaw (English: yaw) of vehicle 200. To this end, the data of the vehicle odometry 132 connected to the wheels are fused with the data of the inertial measurement unit 131, which takes into account the wheel speed and, if necessary, the steering angle. Various methods are known for this purpose. In this case, the coordinate system of the fused odometry is oriented in a planar manner, that is, the z-axis of the coordinate system coincides with the establishment of the gravity axis. In the present exemplary embodiment, the data of the inertial measurement unit 131 are pre-fused and provide a complete orientation (roll, pitch, yaw). The traveled route is calculated from the odometry data of wheel-connected vehicle odometry 132 and mapped using the orientation determined using inertial measurement unit 131. In principle, corresponding odometry information (i.e., the distance traveled and the change in orientation of vehicle 200) can also be determined using other sensors of vehicle 200, for example, by evaluating sensor data detected using video camera 140 or radar sensor 160, or by evaluating position data determined using satellite navigation.

[0025] In method step 430, "scan preprocessing", i.e., preprocessing of the scan, is performed. The sensor data stream of the 3D scanner 120 generated continuously in method step 110 and the 6-DOF distance measurement information generated continuously in method step 420 are used as input variables. The term "scan" here refers to a data packet of the 3D scanner 120, which consists of the corresponding sensor data for a complete 360° rotation, in which a number n of points 321 are measured. Furthermore, the exact measurement time is known for each point 321. "Scan preprocessing" comprises a number of sub-steps: undistortion 431, leveling 432, extraction of vertical structures 433, and projection to 2D (434).

[0026] In the "Correction" method step 431, the movements (i.e., translations and rotations) of 3D scanner 120 during the scanning process are compensated. For each point 321 of 3D point cloud 320, the position of 3D scanner 120 is first determined using 6-DOF odometry 321 based on the temporal information of the corresponding point 321, and then the measured point 321 is determined from there. After this step, the movement of the 3D scanner in the coordinate system of the 6-DOF odometry is compensated.

[0027] In the "Orient" method step 432, the corrected 3D point cloud 120 is now transformed into an oriented coordinate system whose z-axis is oriented parallel to the direction of gravity. This transformation is performed so that the origin corresponds to the selected time (for example, the end of the currently processed scan).

[0028] In the "Extract Vertical Structures" method step 433, vertical structures (e.g., house walls) are extracted within 3D point cloud 320. To this end, a normal estimation is performed for each point 321 of the rectified and oriented 3D point cloud 320. The vertical and horizontal components of the surface normal for each point 321 are determined. Points 321 whose Z-weighted normal component is less than a predefined threshold are then extracted from 3D point cloud 320. Only these extracted points 321 are retained, while the remaining points 321 of 3D point cloud 320 are removed.

[0029] In the "Project to 2D" method step 434, the z components of points 321 remaining in 3D point cloud 320 are removed, as they are no longer needed. Optionally, the size of the point cloud is reduced for faster processing. This can be achieved, for example, using a voxel filter, in which only points 321 at a specific position (e.g., a specific height) are retained. Furthermore, the size of the point cloud can be reduced by regular downsampling, in which only specific points 321 are retained and all other points are removed.

[0030] At the end of "scan preprocessing" 430 (i.e., preprocessing the scans), a 2D point cloud and the associated distance measurement information are present as a result. This information is then incorporated into a 2D-SLAM system having two modes: mapping mode 450 and positioning mode 470. In mapping mode, the 2D-SLAM system generates a map 460 of the surrounding environment 300. In the present embodiment, this map consists of the selected number of scans and their orientations. In positioning mode 470, a larger 2D point cloud is calculated from this. Map 460 is therefore a large point cloud, in which only the 2D components (x, y) are still used.

[0031] Localization provides a current position for each input scan, wherein the position describes the position (x, y) and the orientation (yaw angle or yaw) in 2D. At the beginning of the method, the positioning is initialized with a known starting position, wherein a SLAM graph is set up with the help of an initial node and a corresponding unary factor, which encodes the initial position. For each preprocessed scan, the position is predicted with the help of a provided range finder. Next, scan matching is performed on the predicted position as an initial estimate. This is achieved, for example, with the help of an ICP algorithm, in which correction calculations are iteratively performed. Then, in the SLAM graph, a new node is generated, which has a binary edge to the previous node and has the result of the scan matching step as a unary edge. This fusion scheme based on SLAM graph corresponds to the so-called "pose graph localization" method.

[0032] Figure 3 A block diagram of a system 100 according to the present invention is shown, which simultaneously performs 2D localization and 2D mapping using a 3D scanner 120. System 100 includes a sensor device 110 having a 3D scanner 120 for detecting a 3D point cloud 320 and an odometry measuring device 130 for detecting odometry information of a vehicle 200. Furthermore, system 100 includes a control unit 180 for executing a 2D-SLAM method using 3D point cloud 321. To this end, control unit 180 includes a first control unit 181 for preprocessing the sensor data provided by sensor device 110, transforming the 3D point cloud into a suitable 2D point cloud, and a second control unit 182 for executing the 2D-SLAM method based on the previously generated 2D point cloud. In this embodiment, the two control units 181 and 182 are configured as separate computing units. However, in principle, both control units can also be implemented as a common computing unit, on which all computing steps of the method according to the present invention are then performed.

[0033] exist Figure 4 The schematic diagram shows Figure 3Vehicle 200 of system 100 in FIG. System 100 includes an onboard sensor system 110, which, in addition to a corresponding 3D scanner 120 and an odometry device 130, typically includes at least one other environmental sensor (e.g., a video camera 140, a sensor 150 for satellite navigation, a radar sensor 160, or an ultrasonic sensor 170). A control unit 180 of vehicle 200 is configured to evaluate and further process the data acquired by the environmental sensors. In this example, odometry device 130, configured to determine the distance traveled, includes a wheel-integrated vehicle odometer 132, which determines the distance traveled based on the movement of wheels 210 relative to roadway 301. Odometry device 130 also includes an inertial measurement unit 131, which determines the rotation and acceleration of vehicle 200 during travel. The distance traveled is determined by integrating the measured acceleration over the travel time. The vehicle 200 is preferably an autonomous or automated vehicle (eg, an automated shuttle or AVG for transporting people). In addition, the vehicle 200 can also be configured in the form of a mobile robot.

[0034] In principle, in addition to the vehicle odometry 132 based on wheel speed measurement, other odometry methods for measuring the distance traveled can also be used for 6-DOF odometry—for example, a visual odometry based on the evaluation of video data, or a lidar odometry or radar odometry based on the evaluation of lidar or radar data.

[0035] Although the present invention has been described in detail through the preferred embodiments, the present invention is not limited to the disclosed examples. On the contrary, those skilled in the art can deduce other variations therefrom without departing from the scope of protection of the present invention.

Claims

1. A method for simultaneous 2D positioning and 2D map creation using a 3D scanner (120), in, Scanning a surrounding environment (300) with the aid of the 3D scanner (120) in order to generate a three-dimensional representation of the surrounding environment (300) in the form of a 3D point cloud (320), the 3D point cloud consisting of a plurality of scan points (321), wherein a two-dimensional representation of the surrounding environment (300) in the form of a 2D point cloud is generated from the 3D point cloud (320), wherein the 2D point cloud is provided to a 2D-SLAM algorithm in order to generate a map of the surrounding environment (300) and simultaneously determine the current position (480) of the 3D scanner (120) within the map, In order to perform positioning, it is first initialized with a known starting position. Therein, a SLAM graph is then set up with the aid of an initial node and a corresponding unary factor encoding the starting position (480), For each current scan, the current position (480) is predicted with the help of the currently obtained distance measurement information.

2. The method according to claim 1, in, compensating for movements performed by the 3D scanner (120) during the scanning process, First, distance measurement information is obtained, wherein the distance measurement information represents the displacement of the 3D scanner (120) in six degrees of freedom. Therein, for each measured scan point (321) of the 3D point cloud (320), the current position (480) of the 3D scanner (120) during the measurement of the corresponding point (321) is then determined based on the distance measurement information, and the measured point (321) is then determined from there.

3. The method according to claim 2, in, The distance measurement information is generated by fusing measurement data of a vehicle distance meter (132) with measurement data of an inertial measurement unit (131).

4. The method according to any one of claims 1 to 3, in, The 3D point cloud (320) is transformed into a coordinate system in which the z-axis is oriented parallel to the direction of gravity.

5. The method according to any one of claims 1 to 3, in, Vertical structures are extracted from the 3D point cloud (320) by performing a normal estimation for each point (321) of the 3D point cloud (320) and subsequently removing the following points (321) of the 3D point cloud (320): the normal component of the point in the vertical direction (z) is greater than a predetermined threshold value.

6. The method according to any one of claims 1 to 3, in, The 3D point cloud (320) is projected into a two-dimensional coordinate system by removing a vertical component for each of the scan points (321) of the 3D point cloud (320).

7. The method according to any one of claims 1 to 3, in, The number of points (321) of the 3D point cloud (320) is reduced by means of a voxel filter and / or by regular downsampling.

8. The method according to any one of claims 1 to 3, in, performing scan matching with the predicted current position (480) as an initial estimate, In this case, a new node is subsequently generated in the SLAM graph, which has a binary edge to the previous node and has the scan match as a unary edge.

9. A system (100) for performing the method according to any one of claims 1 to 8, the system comprising: a 3D scanner (120) configured to generate a three-dimensional representation of the surrounding environment (300) in the form of a 3D point cloud (320), a control device (180) for simultaneously performing 2D positioning and 2D map creation, wherein the control device (180) is designed to generate a two-dimensional representation of the surroundings (300) in the form of a 2D point cloud from the 3D point cloud (320), The control device (180) is further configured to provide the 2D point cloud to a 2D-SLAM algorithm in order to generate a two-dimensional map of the surroundings (300) and simultaneously determine the current position (480) of the 3D scanner (120) within the map (460). In order to perform positioning, it is first initialized with a known starting position. Therein, a SLAM graph is then set up with the aid of an initial node and a corresponding unary factor encoding the starting position (480), For each current scan, the current position (480) is predicted with the help of the currently obtained distance measurement information.

10. The system (100) according to claim 9, comprising a distance measuring device (130) for determining distance information, the distance information representing the displacement of the 3D scanner (120) in six degrees of freedom during a scanning process, in, The control device (180) is designed to compensate for the movement of the 3D scanner (120), The control device (180) is further configured to determine, for each scanning point (321) of the 3D point cloud (320), the current position (480) of the 3D scanner (120) during the measurement of the corresponding point (321) based on the distance measurement information, and subsequently determine the measured point (321) from there.

11. The system (100) according to claim 9 or 10, in, The control device (180) is further designed to transform the 3D point cloud (320) into a coordinate system whose z-axis is oriented parallel to the direction of gravity.

12. The system (100) according to claim 9 or 10, in, The control device (180) is further configured to extract vertical structures based on the 3D point cloud (320) by performing a normal estimation for each point (321) of the 3D point cloud (320) and subsequently removing the following points (321) of the 3D point cloud (320): the normal component of the point in the vertical direction (z) is greater than a predetermined threshold value.

13. A vehicle (200) having a system (100) according to any one of claims 9 to 12.

Images