阅读说明：本技术 用于对点云的测量点分类的方法、控制设备、计算机程序和存储介质 (Method, control device, computer program and storage medium for classifying measurement points of a point cloud ) 是由 傅承煊 于 2021-05-31 设计创作，主要内容包括：公开一种用于通过控制设备对通过至少一个传感器所求取的点云(尤其是由激光雷达传感器、雷达传感器和/或摄像机传感器所求取的点云)的测量点进行分类的方法,其中,针对所述点云的任意测量点求取到相邻测量点的局部表面向量,针对任意局部表面向量分别计算所述局部表面向量相对于重力向量之间的角度,针对所述点云的任意测量点,基于所计算的角度求取具有相对于所述重力向量的最大角度的最大表面向量和归一化表面向量,将所述点云的具有如下归一化表面向量和/或最大表面向量的任意测量点分类为非地面点：所述归一化表面向量和/或所述最大表面向量相对于所述重力向量的角度大于界限值。还公开一种控制设备、一种计算机程序以及一种机器可读的存储介质。(Disclosed is a method for classifying measurement points of a point cloud determined by at least one sensor (in particular a point cloud determined by a lidar sensor, a radar sensor and/or a camera sensor) by means of a control device, wherein for any measurement point of the point cloud a local surface vector to an adjacent measurement point is determined, the angle between the local surface vector with respect to a gravity vector is calculated for any local surface vector in each case, for any measurement point of the point cloud a maximum surface vector and a normalized surface vector with a maximum angle with respect to the gravity vector are determined on the basis of the calculated angles, and any measurement point of the point cloud with the following normalized surface vector and/or maximum surface vector is classified as a non-ground point: the angle of the normalized surface vector and/or the maximum surface vector relative to the gravity vector is greater than a threshold value. A control device, a computer program and a machine-readable storage medium are also disclosed.)

1. Method for classifying measurement points (4) of a point cloud (P) determined by at least one sensor (2), in particular by a lidar sensor, a radar sensor and/or a camera sensor, by means of a control device (6), wherein,

for any measuring point (4) of the point cloud (P), local surface vectors (10) of adjacent measuring points (12) are determined,

calculating an angle (WD) between the local surface vector (10) with respect to a gravity vector (g) separately for any local surface vector (10),

for any measurement point (4) of the point cloud (P), a maximum surface vector (20) and a normalized surface vector (16, 18) with a maximum angle (WD) relative to the gravity vector (g) are determined on the basis of the calculated angle (WD),

classifying any measurement point of the point cloud having a normalized surface vector (16, 18) and/or a maximum surface vector (20) as a non-ground point (16): the angle (WD) of the normalized surface vector and/or the maximum surface vector with respect to the gravity vector (g) is larger than a threshold value.

2. The method according to claim 1, wherein unclassified neighboring measurement points (12) of measurement points classified as non-ground points are found, wherein the following unclassified neighboring measurement points (12) of measurement points classified as non-ground points are classified as non-ground points: the unclassified adjacent measurement points have the same azimuth angle (WA) and higher or the same elevation angle (WE).

3. Method according to claim 1 or 2, wherein for any unclassified measurement point (12) of the point cloud (P) a height value (z) is compared with the height of the sensor (2) above the ground, wherein an unclassified measurement point (12) of the point cloud (P) is classified as a ground point if its height value (z) substantially coincides with the height of the sensor (2) above the ground.

4. Method according to one of claims 1 to 3, wherein measurement points (4) of the point cloud (P) classified as ground points with at least one unclassified neighboring measurement point (12) are determined and a region growing method (24) is applied.

5. Method according to any one of claims 1 to 4, wherein the measurement points (4) of the point cloud (P) are at least temporarily stored in a storage unit (8) in a structured form having a plurality of rows and columns.

6. A control device (6), wherein the control device (6) is provided for carrying out the method according to any one of claims 1 to 5.

7. A computer program comprising instructions which, when the computer program is implemented by a computer or control device (6), cause the computer or control device to carry out the method according to any one of claims 1 to 5.

8. A machine-readable storage medium (8) on which a computer program according to claim 7 is stored.

Technical Field

The invention relates to a method for classifying measurement points of a point cloud determined by at least one sensor, in particular a point cloud determined by a lidar sensor, a radar sensor and/or a camera sensor. The invention also relates to a control device, a computer program and a machine-readable storage medium.

Background

In the fields of automated driving assistance functions and automated driving, a laser radar sensor, a radar sensor, or a camera sensor is generally used as an environment sensor to perform environment sensing. The environment can be scanned by means of an environment sensor in order to determine a plurality of measurement points in the form of a three-dimensional point cloud having distance information to the object in the scanning region. In this case, for example, a travel Time measurement or a so-called Time-of-Flight (Time-of-Flight) measurement is carried out, and the distance covered by the emitted beam is calculated from the measured travel Time.

In order to detect objects from the measurement points of the point cloud, it is generally necessary to classify the measurement points into ground points and non-ground points, which are associated with the ground. Subsequent object identification is performed based on the measurement points classified as non-ground points. Methods for classifying measurement points of a point cloud are known. However, the known methods are complex and therefore require high computational power. Due to the complexity of these methods, the real-time processing of the measurement data can only be carried out with high technical expenditure. Furthermore, the performance of the known method in terms of false alarm rate (false-postiv-Raten) and false alarm rate (false-postiv-Raten) is insufficient.

Disclosure of Invention

The task on which the invention is based can be seen as proposing a method for classifying measurement data which has reduced computational power requirements and is real-time.

The object is achieved by a method for classifying measurement points of a point cloud determined by at least one sensor, in particular a point cloud determined by a lidar sensor, a radar sensor and/or a camera sensor, a control device, a computer program and a machine-readable storage medium. The following also gives advantageous configurations of the invention.

According to one aspect of the invention, a method for classifying measurement points of a point cloud determined by at least one sensor is provided. The at least one sensor can determine measurement data in the form of measurement points and can be configured, for example, as a lidar sensor, a radar sensor and/or a camera sensor.

The method may be implemented by a control device. The control device can be configured as a Field Programmable Gate Array (FPGA), an Application Specific Integrated Circuit (ASIC), a microprocessor, a computer or as a hardware accelerator.

The algorithm may also be implemented in an FPGA, ASIC, or other type of hardware accelerator to reduce CPU load.

In one step of the method, for any measurement point of the point cloud, a local surface vector of a neighboring measurement point is determined. The local surface vector can be determined as a normal vector and oriented towards the adjacent measuring point. In this case, for any measurement point, the surface vector to its neighboring point or neighboring measurement points is first calculated. Alternatively, the respective unknown surface vector can also be calculated by a cross product (Kreuzprodukt) of two known surface vectors.

The angle between the corresponding local surface vector with respect to the gravity vector is calculated for each local surface vector. Subsequently, for any measurement point of the point cloud, a maximum surface vector and a normalized surface vector having a maximum angle with respect to the gravity vector are found based on the calculated angle.

The gravity vector is oriented perpendicular to the ground (unorgrun or Boden) corresponding to the gravitational force.

The angle between the local surface vector relative to the gravity vector may preferably lie in the range between 0 ° and 90 °, including 0 ° and 90 °. In the case of an angle greater than 90 °, this angle can be found as a subtraction from 180 °.

The maximum surface vector can be determined for any measurement point. At least one surface vector can be associated with any measurement point. In the case where a plurality of surface vectors pointing from one measurement point to an adjacent measurement point are assigned to one measurement point, a surface vector having the largest angle with respect to the gravity vector may be defined as the largest surface vector. Thus, the maximum surface vector may be oriented substantially parallel with respect to the X-Y plane, thereby maximizing the angle with respect to the gravity vector.

The normalized surface vector may be calculated by constructing an average from all the surface vectors of the measurement points. Here, the angle between the normalized surface vector and the gravity vector may also be found.

In another step, any measurement point of the point cloud having a normalized surface vector and/or a maximum surface vector as follows is classified as a non-ground point: the angle of the normalized surface vector and/or the maximum surface vector with respect to the gravity vector is greater than a threshold value. The limit value can in particular lie in an angle range of 45 ° to 90 °. Thereby, measurement points with normalized surface vectors and/or maximum surface vectors as follows are classified as non-ground points: the normalized surface vector and/or the maximum surface vector has an excessive angle with respect to the gravity vector. This step is important to avoid false positive results in further processing, since measurement points classified as non-ground points are not excluded in the algorithm of the alternative region growing method, thereby enabling the region growing method to be accelerated.

In the case of false positive results, the measurement points of the object may be classified as ground points. In the case of a false negative result, the measurement points that depict the ground may be classified as non-ground points.

According to a further aspect of the invention, a control device is provided, wherein the control device is provided for carrying out the method. The control device may be, for example, a vehicle-side control device, a control device outside the vehicle, or an external server unit (e.g., a cloud system).

Furthermore, according to an aspect of the invention, a computer program is provided, which comprises instructions which, when the computer program is implemented by a computer or a control device, cause the computer or the control device to carry out the method according to the invention. According to a further aspect of the invention, a machine-readable storage medium is provided, on which a computer program according to the invention is stored.

The control device can be used, for example, in vehicles which can be operated in an assisted, partially automated, highly automated and/or fully automated or driverless manner according to the BASt standard. Such vehicles may be, for example, passenger cars, trucks, robotic taxis, and the like. The vehicle is not limited to operation on roads. Rather, the vehicle may also be configured as a watercraft, an aircraft (e.g., a transport drone), or the like.

According to one embodiment, unclassified neighboring measurement points of measurement points classified as non-ground points are evaluated, wherein the following unclassified neighboring measurement points of measurement points classified as non-ground points are classified as non-ground points: these unclassified neighboring measurement points have the same azimuth and higher or the same elevation. By this measure, all unclassified neighboring measurement points are compared in terms of their height deviation with measurement points that have been classified as non-ground points. Measurement points classified as non-ground points may be equally classified as non-ground points if the height or z-value of adjacent unclassified measurement points is greater than the height or z-value of measurement points classified as non-ground points. The method may be repeated for each row and column of measurement points of the point cloud. Thereby the probability of false positive results can be further reduced.

According to a further embodiment, for any unclassified measurement point of the point cloud, the height value or z value is compared with the height of the sensor above the ground, wherein an unclassified measurement point of the point cloud is classified as a ground point if its height value substantially coincides with the height of the sensor above the ground. The remaining measurement points are thus compared for the z-value without classification. A measurement point is similarly classified as a ground point if its z-value substantially coincides with the gravity vector and the direction of the surface vector substantially coincides with the direction of the gravity vector.

According to a further embodiment, measurement points of the point cloud classified as ground points are determined with at least one unclassified neighboring measurement point, and a region growing method is applied. Here, the remaining neighboring measurement points are classified as neither non-ground points nor ground points. By this measure, other ground points are identified using key features, wherein there is similarity of surface vectors of neighboring ground points. Additionally, other criteria (e.g., differences in z-values) may be considered by the region growing method. By this step, it is possible in particular to avoid false-positive results which may occur, for example, in hills.

According to a further embodiment, the measurement points of the point cloud are at least temporarily stored in a storage unit in a structured form having a plurality of rows and columns. The memory unit may be integrated in the control device or may be an external memory unit. By providing the measurement points structurally, any measurement point can be accessed based on the row number and the column number. In particular, adjacent measurement points can thereby be identified for any measurement point.

The measurement points of the same column preferably have the same elevation angle and the measurement points of the same row (Reihe or Zeile) have the same azimuth angle.

Drawings

The preferred embodiments of the invention are further explained below on the basis of an extremely simplified schematic diagram. Shown here are:

fig. 1 shows a schematic diagram with an arrangement of an exemplary point cloud, to illustrate a method according to an embodiment,

figure 2 shows a schematic comparison of the normalized surface vector of the measurement points with respect to the gravity vector,

figure 3 shows a schematic comparison of the maximum surface vector of the measurement points with respect to the gravity vector,

fig. 4 shows a schematic comparison between a measurement point classified as a non-ground point and an unclassified measurement point with a larger z-value.

Detailed Description

Fig. 1 to 4 show schematic diagrams to illustrate a method according to an embodiment. The method is used for classifying measurement points 4 determined by at least one sensor 2.

A schematic view of an arrangement 1 with an exemplary point cloud P is shown in fig. 1. The arrangement 1 has a sensor 2, for example in the form of a lidar sensor. Alternatively or additionally, the sensor 2 may have a radar sensor and/or a camera sensor.

The sensor 2 can scan the scanning area a and collect measurement data in the form of measurement points 4. The measuring points 4 are present in a grid or table and can be assigned to rows and columns of the table. Here, the rows correspond to the azimuth angle WA and the columns correspond to the elevation angle WE.

The sensor 2 is connected to the control device 6 in a data-conducting manner. The control device 6 can receive the measuring points 4 of the sensor 2 and store them at least temporarily in the memory unit 8.

The memory unit 8 can be configured as a machine-readable storage medium on which a computer program is stored, which computer program comprises instructions that, when the computer program is executed by the control device 6, cause the control device to execute the method.

In fig. 1, a measuring point 4 with four surface vectors 10 to adjacent measuring points 12 is shown by way of example. For any measurement point 4 of the point cloud P, a local surface vector 10 to a neighboring measurement point 12 is determined.

Fig. 2 shows a schematic comparison of the normalized surface vectors 16, 18 of the measurement points 4 with respect to the gravity vector g. For clarity, arrangement 1 is not shown. Here, normalized surface vectors 16 of the measured data 4 classified as non-ground points and normalized surface vectors 18 of the measured data 4 classified as ground points are shown.

In the step shown in fig. 2, the angle WD between these local surface vectors 10 with respect to the gravity vector g is calculated separately for any local surface vector 10. Here, the following relationship of ground points may be used: these ground points extend more parallel to the gravity vector g than non-ground points.

Fig. 3 shows a schematic comparison of the maximum surface vector 20 of the measuring point 4 with respect to the gravity vector g. In particular, the largest surface vector 20 on an exemplary building wall is shown oriented in an almost perpendicular manner relative to the normalized surface vector 18. Therefore, both the maximum surface vector 20 and the normalized surface vectors 16, 18 must be considered for classification in order to robustly implement the method.

For any measurement point 4 of the point cloud P, a maximum surface vector 20 and normalized surface vectors 16, 18 are found based on the calculated angle WD. The maximum surface vector 20 corresponds to one of the surface vectors 10 of the measuring point 4, which has the maximum angle WD with respect to the center of gravity vector g. Thus, the maximum surface vector 20 is substantially parallel to the x-y plane.

In a further step, any measurement point 4 of the point cloud P having the following normalized surface vector 16 and/or maximum surface vector 20 is classified as a non-ground point: the angle WD of the normalized surface vector and/or the maximum surface vector with respect to the gravity vector g is greater than a threshold value.

A schematic comparison between a measurement point 4 classified as a ground point and an unclassified measurement point 12 with a larger z-value is shown in fig. 4. The measuring points 4 of the point cloud P classified as ground points are determined with at least one unclassified neighboring measuring point 12, and a region growing method is applied. Here, the remaining neighboring measurement points 12 are classified as neither non-ground points nor ground points. By this measure, the similarity of the surface vectors of adjacent ground points can be used. For example, a slightly raised surface 22 can thus be registered and classified as belonging to the ground. Arrow 24 shows an exemplary direction of the region growing method.