WO2018066754A1

WO2018066754A1 - Method for estimating attitude of vehicle by using lidar sensor

Info

Publication number: WO2018066754A1
Application number: PCT/KR2016/013670
Authority: WO
Inventors: 박태형; 조준형
Original assignee: 충북대학교 산학협력단
Priority date: 2016-10-06
Filing date: 2016-11-25
Publication date: 2018-04-12
Also published as: KR101909953B1; KR20180038154A

Abstract

The present invention provides a method for estimating the attitude of a vehicle having a built-in inertial measurement unit and having a light detection and ranging (LiDAR) sensor mounted in the rear thereof, the method comprising the steps of: acquiring measurement point data corresponding to a distance measured using the LiDAR sensor; finding a road line by projecting the measurement point data on two-dimensional rectangular coordinates; calculating an attitude variation (α', β') of the vehicle from the road line; and calculating the attitude (α, β) of the vehicle by combining the attitude variation (α', β') with a measurement value measured by the inertial measurement unit. Using a LiDAR sensor which is conventionally mounted to a vehicle, the present invention can quickly estimate the attitude of a vehicle without adding any expensive inertial measurement unit or GPS device.

Description

Vehicle attitude estimation method using lidar sensor

The present invention relates to a method of estimating a current posture of a vehicle, and more particularly, to a method of estimating a current posture of a vehicle using a rearward downward-looking 2D LiDAR (LiDAR, Light Detection And Ranging) sensor.

In order to control the autonomous vehicle running at high speed, it is essential to estimate the attitude of the vehicle.

1 is a diagram for defining a posture of a vehicle.

Referring to FIG. 1, a three-dimensional coordinate system is fixed to a vehicle. When a rotation about each axis is referred to as α for x-axis rotation, β for y-axis rotation, and γ for z-axis rotation, α and β are poses. ), And γ is defined as the heading of the vehicle.

A device generally used for estimating a vehicle's attitude is an IMU (Inertial Measurement Unit), which measures three-axis angular velocities, that is, changes in α, β, and γ, and three-axis acceleration. At this time, the values of α, β, and γ are not measured immediately, but the amount of change is measured. However, since noise components accumulate together, an algorithm that mixes with other measurements is essential.

In the case of estimating the three-axis acceleration values of the inertial measurement unit without additional sensors, the accuracy is low and the delay speed due to the filter also becomes a problem. The use of high-priced products can be alleviated to some extent, but since the inertia measuring device becomes more expensive as the precision is higher, the vehicle attitude estimation system is generally configured by adding another sensor to the inertial measuring device.

In the case of the configuration by adding a GPS (Global Positioning System), a method of fusing the vehicle position information with the measurement value of the inertial measurement apparatus is used. Fusion achieves proper positional accuracy, but the vehicle's posture is not reliable because the operating cycle is very slow, such as 1 Hz, when using general GPS. When using an expensive receiver, the positional information is relatively accurate, but there is a problem that it is difficult to immediately estimate the attitude of the vehicle still driving at a high speed.

When the geomagnetic sensor is added, it is possible to directly measure the value of γ rather than the change in γ so that both the attitude and the direction of the vehicle can be estimated. However, there is a problem that the precision does not increase significantly and the delay speed due to the filter still does not decrease.

The present invention has been made to solve the above problems, and provides a method for estimating the attitude of the vehicle using a 2D LiDAR (Light Detection And Ranging) sensor mounted to the rear of the autonomous vehicle. There is a purpose.

The object of the present invention is not limited to the above-mentioned object, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, an inertial measurement unit (Inertial Measurement Unit) of the present invention is built in, and a lidar (LiDAR, Light Detection And Ranging) sensor is mounted on the rear of the vehicle attitude estimation method, the lidar sensor Acquiring measurement point data measuring a distance by using; projecting the measurement point data onto two-dimensional rectangular coordinates to find a road line; calculating a change in attitude of the vehicle (α ', β') from the road line And calculating the postures (α, β) of the vehicle by fusing the posture change amounts α 'and β' with the measured values of the inertial measurement apparatus.

The finding of the road line may include converting a point set P, which is a set of the measurement point data expressed in a polar coordinate system, into a rectangular coordinate system. Creating a transformation matrix by matching the point set P to the occupant grid map, continuously updating and optimizing the transformation matrix, and calculating the slope and intercept of the road line from the optimized transformation matrix. Deriving may be included.

It is possible to calculate the amount of change in the attitude of the vehicle α 'and β' from the transformation matrix.

The optimizing the transformation matrix may include calculating a linearly interpolated occupancy and occupancy slope value, updating a transformation matrix through the linearly interpolated occupancy and slope value, and a first predetermined error of the updated transformation matrix. If the threshold is greater than the threshold and the number of update repetitions is less than or equal to the second predetermined threshold, calculating the linearly interpolated occupancy rate and occupancy slope value again to update the transform matrix and the error of the updated transform matrix is less than the first threshold value, or If the number of update repetitions is greater than or equal to a second predetermined threshold, the optimization process may be terminated.

According to the present invention, the posture of the vehicle can be estimated quickly by using a lidar sensor attached to an existing vehicle without adding an expensive inertial measurement device or GPS equipment.

In addition, since the present invention uses a lidar sensor mounted on a vehicle, the vehicle attitude estimation function can be implemented at no additional cost.

1 is a diagram for defining a posture of a vehicle.

2 is a view showing an example of the measurement range and measured value of the 2D lidar sensor.

3 is an overall flowchart illustrating a method of estimating a vehicle attitude using a lidar sensor according to an exemplary embodiment of the present invention.

FIG. 4 is a diagram illustrating measured values of a lidar sensor viewed from a road in a Cartesian coordinate system.

5 is an exemplary diagram in which a transform is applied to an existing point set.

6 is a diagram illustrating a road reference line and a road line extracted while driving.

7 is a diagram illustrating an occupant lattice map used for optimization according to an embodiment of the present invention.

8 is a flowchart illustrating a transform matrix optimization method according to an embodiment of the present invention.

In the attitude estimation method of a vehicle in which an inertial measurement unit (Inertial Measurement Unit) of the present invention is built-in and a LiDAR (Light Detection And Ranging) sensor is mounted at the rear, a distance is measured using the lidar sensor. Acquiring measurement point data; projecting the measurement point data onto two-dimensional rectangular coordinates to find a road line; calculating a change in attitude of the vehicle from the road line; ', β') and the measured values of the inertial measurement device to calculate the attitude (α, β) of the vehicle.

As the present invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed to have meanings consistent with the meanings in the context of the related art, and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

In addition, in the description with reference to the accompanying drawings, the same components regardless of reference numerals will be given the same reference numerals and duplicate description thereof will be omitted. In the following description of the present invention, if it is determined that the detailed description of the related known technology may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

The present invention relates to a method of estimating a vehicle's attitude using a LiDAR (Light Detection And Ranging) sensor.

In the present invention, the subject performing the method of estimating the vehicle attitude using the lidar sensor may be an electronic system embedded in the vehicle. For example, a controller (processor) of an electronic control system module embedded in a vehicle, an electronic control system mounted on an autonomous vehicle, or the like may be a main agent.

The Lidar sensor is a sensor that directly measures the distance by shooting light. It is widely used in autonomous vehicles for location recognition, obstacle detection, and moving object tracking because the accurate distance can be obtained at high speed. The 2D lidar sensor can shoot light at different angles to obtain multiple distance data in one scan.

Referring to FIG. 2, the values reflected and measured by the vehicle lying within the measurement range indicated by the arc-shaped areas in the left figure are shown as points on two-dimensional coordinates in the right figure. As such, the 2D lidar sensor has a characteristic that distance and phase data called measurement points are displayed on one plane.

In general, autonomous vehicles may be equipped with a 2D lidar sensor at the rear for parking or rear obstacle detection, and thus, there is no additional cost when using them for estimating the attitude of the vehicle.

Referring to Figure 3, the vehicle attitude estimation method according to an embodiment of the present invention is as follows.

First, measurement point data obtained by measuring a distance from a lidar sensor is obtained (S310).

Next, the measurement point data is displayed on the two-dimensional rectangular coordinates, and the road line is found on the two-dimensional rectangular coordinates (S320).

The attitude change amounts α 'and β' of α and β are calculated from the road line (S330).

Then, the final posture values α and β are calculated by fusion with the measured values of the inertial measurement apparatus (S340).

In the present invention, every step of S310 to S340 is executed once every measurement cycle of the lidar sensor. Hereinafter, each step will be described in detail.

In step S310, a point set consisting of distance and phase, which is raw data measured by a 2D lidar sensor, is obtained. Each of these points is called a measuring point.

In FIG. 4, each point may be represented by the following equation.

Here, r represents a distance, θ represents an angle, and n measurement points can be obtained. The number n of measurements depends on the performance of the lidar. In the present invention, in order to reduce the complexity of the calculation and increase the accuracy, measurement points having too large or too small distance values are previously excluded from the point set.

Step S320 is a process of finding road lines by projecting measurement point data into a Cartesian coordinate system. Here, the road line is a straight line made up of measurement points indicating the distance measured by the light from the lidar sensor reflected on the road, that is, the distance and phase of the road.

In FIG. 4, measurement points obtained by a lidar sensor viewed from the road are illustrated. In Figure 4, the measuring points representing the road can be approximated by a straight line.

In order to find these road lines, a fast and accurate straight line extraction method is required. Although there are many fast and accurate methods among the existing straight line extraction methods, it has a problem that the robustness is poor when applied in the road environment that changes in real time. In order to solve this problem, the present invention proposes a method of finding a road line by matching a predefined road reference line for every measurement.

First, the conversion from the polar coordinate system to the Cartesian coordinate system is as follows.

In the present invention, since only the data measured by the reflection of the road is required, the range is arbitrarily reduced when the viewing angle of the lidar sensor is wide. The point set P is then multiplied by the two-dimensional transformation matrix.

5 is an exemplary diagram in which a transform is applied to an existing point set. Here, the transformation matrix is as follows.

Here, t _x represents x-axis movement, t _y represents y-axis movement, and phi means rotational movement angle. Finding this transformation matrix makes it easy to find the road line by simple calculation.

To find the transformation matrix, you need a road baseline to compare against. Road baseline refers to the portion of the lidar measurements taken on the flat road.

The dashed line in FIG. 6 is the road reference line, which depends on the mounting height and downward angle of the lidar sensor.

There are many ways to find the transformation matrix, but there are a limited number of methods that can be used in anomalous situations, such as driving on roads. This is because there is no guarantee that the number of point sets is the same each time, and the transformation matrix cannot be updated by simple matrix operation. Therefore, although a feature such as a straight line has been mainly selected or a supplementary point-to-point matching method has been used, the former has a high performance, but the robustness is very poor when the vehicle rotates, and the latter has a high accuracy but a real time. There is a problem that is difficult to operate. Therefore, in the present invention, a method of optimizing a function in the occupant grid map will be used.

Briefly, in order to find a transformation matrix, a method of finding a transformation matrix having the smallest error by matching the point set P with a predefined graticule map. Here, the occupant grid map is one of the methods of expressing a map, and has a share in each grid having a fixed size. In the present invention, an occupant grid map in which a road reference line is drawn in advance is used.

Referring to FIG. 7, the occupant lattice map is prepared in the form of a straight line along the Gaussian distribution, and is blanked line by line to prevent numerical divergence due to a zero y-axis gradient. Here, the distance between the road reference line and the origin of the map is calculated as follows through the height and downward angle of the lidar sensor.

Where ρ is the distance from the origin of the occupant grid map to the road reference line, h is the height of the lidar sensor, and ω is the downward angle of the sensor.

In the present invention, the size per cell of the graticule map, i.e. the resolution, depends on the performance of the LiDAR sensor. That is, as the number of points compared to the viewing angle increases, the resolution of the occupant grid map may be increased. In the present invention, the resolution of the occupant grid map must be high to increase the precision of the transformation matrix.

In the present invention, the road line can be found by deriving the slope and the intercept of the road line from the optimized transformation matrix.

Then, the vehicle posture change amounts α 'and β' can be calculated from the transformation matrix.

Now, the matching process in the present invention will be described in detail.

In the present invention, the Levenberg-Marquardt algorithm, which is one of nonlinear function optimization techniques, is applied to find the optimal transform T ^* . This algorithm has the advantage of high probability of finding global solution and fast convergence if the initial value is within a certain range.

Referring to FIG. 8, in the optimizing of the transformation matrix according to an embodiment of the present invention, first, a linear interpolated occupancy rate and occupancy slope value are calculated (S810).

The transformation matrix is updated through linearly interpolated occupancy and slope values (S820).

Here, if the error of the updated transformation matrix is greater than or equal to the first predetermined threshold (S830), and if the number of update iterations is less than or equal to the predetermined second threshold (S840), the linearly interpolated occupancy ratio and occupancy gradient value are calculated again. Update (S810, S820).

If the error of the updated transformation matrix is less than the first threshold (S830) or if the number of update iterations is greater than or equal to a second predetermined threshold (S840), the optimization process is terminated.

This optimization process is described in more detail as follows.

First, we update the transformation matrix T ^* until it converges sufficiently into the following equation.

Here, p _{k +} ₁ is the variables t _x , t _y , φ of the transformation matrix T ^* to be optimized, and p _k is the previous transformation matrix T. R (p _k ) is the error when the current transformation matrix is applied, J _r is the first derivative of the error function, and H _r is the second derivative of the error function.

H _r can be obtained as Jr as in the following equation.

Next, the previous transformation matrix T is substituted to obtain a linearly interpolated occupancy rate and occupancy slope value as in the following equation.

Where v is occupancy,

Is the slope in the x direction,

Denotes the slope of the y direction.

The Jr matrix is calculated using the occupancy ratio and the first derivative of the point, and the transformation matrix is updated. J _r is calculated as follows.

Here, x and y are the coordinates of each point in the point set, and φ is the rotation angle of the road line.

Then, H _r is obtained from the calculated J _r , and the updated conversion matrix T ^* is calculated.

In the present invention, if the error is reduced, the buffer coefficient μ _k is reduced, otherwise the μ _k is increased to be used in the next iteration. When the error is less than the threshold or reaches the set number of repetitions, the optimization process is terminated and the updated transform matrix T ^* is stored.

In operation S330, an amount of change in α and β of the vehicle is calculated from the transformation matrix.

Referring to FIG. 6, when a three-dimensional x- and y-axis rotation transformation is applied to a straight line obtained from a transformation matrix indicated by a solid blue line, a straight line in a flat plane represented by a green dotted line may be obtained. Then, α and β of the vehicle are calculated by the following equation.

Where p is a point on the road reference line considering the attachment position and the downward angle of the lidar sensor when p is flat, p 'is a point on the calculated straight line, R _x is the x-axis 4x4 rotation matrix, and R _y is y Axis 4x4 rotation matrix. This calculation is easily solved by the quadratic system of equations for α and β, and compared with the road baseline on the plain, so it can be said that the change is actually obtained.

The final step S340 is a step of compensating by mixing the amount of change in α and β obtained by performing step S330 with the amount of α and β estimated by the inertial measurement sensor inside the vehicle. Here, the vehicle attitude is estimated using the vehicle speed, the measurement cycle of the lidar sensor, and the posture change amount acquired by the lidar sensor. The Kalman filter is then used to fuse the position estimated by the inertial measurement sensor.

While the invention has been described using some preferred embodiments, these embodiments are illustrative and not restrictive. Those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the invention and the scope of the rights set forth in the appended claims.

Claims

In a vehicle attitude estimation method in which an inertial measurement unit is embedded and a LiDAR (LiDAR, Light Detection And Ranging) sensor is mounted at the rear,

Acquiring measurement point data of a distance using the lidar sensor;

Finding the road line by projecting the measurement point data onto two-dimensional rectangular coordinates;

Calculating an amount of change in attitude of the vehicle from the road line (α ', β'); And

And calculating the postures (α, β) of the vehicle by fusing the posture change amounts (α ', β') with the measured values of the inertial measurement apparatus.
The method according to claim 1,

Finding the road line,

Converting a point set P, which is a set of measurement point data expressed in a polar coordinate system, into a rectangular coordinate system;

Creating an occupant grid map, which is a map having a share for each fixed grid, in the form of a straight line along the Gaussian distribution;

Obtaining a transformation matrix by matching the point set P to the occupant grid map;

Continuously updating and optimizing the transformation matrix; And

And deriving a slope and an intercept of the road line from the optimized transformation matrix.
The method according to claim 2,

And calculating a vehicle posture change amount (α ', β') from the transformation matrix.
The method according to claim 2,

Optimizing the transformation matrix,

Calculating linearly interpolated occupancy and occupancy slope values;

Updating a transformation matrix with the linearly interpolated occupancy and slope values;

If the error of the updated transformation matrix is greater than or equal to the first predetermined threshold and the number of update iterations is less than or equal to the second predetermined threshold, calculating the linearly interpolated occupancy ratio and occupancy gradient value again to update the transformation matrix; And

And if the error of the updated transformation matrix is less than the first threshold, or if the number of update iterations is greater than or equal to a second predetermined threshold, terminating the optimization process.