CN108445886A - A kind of automatic driving vehicle lane-change method and system for planning based on Gauss equation - Google Patents

Abstract

The present invention provides a kind of automatic driving vehicle lane-change method and system for planning based on Gauss equation, and wherein method includes the following steps：Step 1：The local waypoint format of definition；Step 2：Calculate barrier center, that is, longitudinal length；Step 3：It plans function design, and calculates the translational movement △ h of each waypoint；Step 4.The translational movement △ h are tagged to the abscissa variable quantity △ x of corresponding waypoint.The present invention is by calculating barrier center point coordinate and barrier longitudinal direction depth, Gauss origin is moved to barrier central point from car body local coordinate origin, while longitudinal depth of Use barriers object carries out Gaussian function segmentation and obtains the corresponding translation percentage of each waypoint；The cross directional variations amount of each waypoint in planning is marked with world coordinates, ensures the consistency in path when lane-change situation is constant.

Description

A method and system for lane-changing planning of autonomous vehicles based on Gaussian equations

technical field

The invention relates to the technical field of computer vision and image processing, in particular to a Gaussian equation-based lane-changing planning method and system for an automatic driving vehicle.

Background technique

Autonomous driving technology is becoming more and more mature, and path planning is the guarantee of the intelligence of unmanned vehicles. At present, the lane change method of most unmanned vehicles adopts the trajectory translation method. Although this method is simple and efficient, there is a problem of sudden changes in the trajectory. Such problems may easily lead to uneven lateral control of unmanned vehicles, resulting in a poor driving experience; At the same time, if the lane change situation remains unchanged during lane change planning, in order to ensure the consistency of the navigation path, unmanned vehicles should reduce planning as much as possible. In order to solve the above problems, the present invention proposes a lane change trajectory planning method combining trajectory translation and Gaussian low-pass filtering. By calculating the coordinates of the center point of the obstacle and the longitudinal depth of the obstacle, the Gaussian origin is moved from the origin of the local coordinates of the vehicle body to the obstacle At the same time, the longitudinal depth of the obstacle is used to segment the Gaussian function to obtain the translation percentage corresponding to each waypoint; the lateral change of each waypoint in the plan is marked with global coordinates to ensure the path when the lane change situation remains unchanged consistency. The present invention solves the problem of sudden change in trajectory at the source of lane changing, and at the same time not only can plan a lane changing path, a lane keeping path, but also can plan a path returning to the original trajectory. In addition, based on the characteristics of the Gaussian function, only the cut-off frequency of the Gaussian function needs to be adjusted for lane change path adjustment, which is simple and efficient.

The invention patent with the application number CN201710497273.5 discloses a lane change control method and device for an automatic driving vehicle. When changing lanes, the drivable area is first determined according to the distance between the obstacle and the vehicle and the current road conditions. When the last guidance trajectory is within the drivable area, it is preparing to change directions, avoiding the steering wheel shaking caused by repeated attempts to change lanes. This method requires path translation every cycle during the lane change process until the end of the lane change. This approach will cause the planned trajectory to be affected by vehicle body motion changes and sensor errors during the lane change process, resulting in a path that is not planned each time. Correlation is weak.

The invention patent with the application number CN1O5329238A discloses a monocular vision-based lane-changing control method for autonomous vehicles. This method installs a camera on the roof of the autonomous vehicle to collect images of lane lines; Line image processing and recognition to get the fitted lane line; the upper computer module calculates the steering wheel angle increment, and outputs the motor control signal to the execution unit. This method has relatively high requirements for the acquisition device, and can only use the camera to collect lane information. The lane changing method based on monocular vision image processing has insufficient adaptability.

Contents of the invention

In order to solve the above-mentioned technical problems, the present invention proposes a method and system for lane-changing planning of an automatic driving vehicle based on the Gaussian equation. By calculating the coordinates of the center point of the obstacle and the longitudinal depth of the obstacle, the Gaussian origin is moved from the origin of the local coordinates of the vehicle body to The center point of the obstacle, and at the same time use the longitudinal depth of the obstacle to perform Gaussian function segmentation to obtain the translation percentage corresponding to each waypoint; mark the lateral change amount of each waypoint in the plan with global coordinates to ensure that when the lane change situation remains unchanged Path consistency.

The first object of the present invention is to provide a method for lane-changing planning of automatic driving vehicles based on Gaussian equations, comprising the following steps:

Step 1: Define the local waypoint format;

Step 2: Calculate the obstacle center, that is, the longitudinal length;

Step 3: planning function design, and calculating the translation amount △h of each waypoint;

Step 4: Mark the translation amount Δh to the abscissa change amount Δx corresponding to the waypoint.

Preferably, each waypoint is composed of global coordinates (X, Y), local coordinates (x, y) and abscissa variation Δx.

In any of the solutions above, it is preferable that the global coordinates are a coordinate system that takes into account the positions of vehicles, obstacles, maps, etc., and the origin of the coordinates does not change.

In any of the above schemes, it is preferred that the local coordinates take the front of the vehicle as the origin, the direction of movement of the vehicle as the vertical axis, and the direction perpendicular to the direction of movement as the horizontal axis. The origin of the coordinates changes in real time with the vehicle body .

In any of the above solutions, it is preferred that the step 2 is to calculate the longitudinal depth s of the obstacle, that is, the local coordinates (j, i) of the center of the obstacle.

Preferably in any of the above schemes, the step 3 includes taking the longitudinal distance of the path as an independent variable, and taking the translation percentage H(y) corresponding to each distance obtained according to the Gaussian low-pass filter function, and the translation percentage H (y) is calculated as

Among them, D ₀ bit is about the measure of the expansion of the center, and y is the distance from the center of the frequency rectangle.

In any of the above schemes, preferably, the formula for calculating the translation amount Δh of each of the waypoints is:

△h=h×H(y)

Among them, h is the maximum translation distance value.

In any of the above schemes, it is preferred that the step 3 includes considering the longitudinal length of the obstacle to ensure the safety of lane keeping after changing lanes, and obtain the following translation distribution function

Among them, T is the value range of the longitudinal distance, which is symmetrical about the center point of the obstacle.

In any of the solutions above, preferably, the step 4 includes correcting the new local coordinates by Δx in the global.

Preferably in any of the above schemes, the Gaussian low-pass filter function is

Among them, μ is the probability density function value of normal distribution, and ^σ2 is the variance.

In any of the above schemes, it is preferred that a random variable X obeys the distribution of the function f(x, μ, σ), denoted as XN(μ, σ ² ).

Preferably in any of the above schemes, based on the characteristics of the Gaussian probability distribution, the Gaussian function can perform the evolution of the filter function in the frequency domain, and the following formula is obtained:

Among them, D(u, v) is the distance from the center of the frequency rectangle, and σ is a measure of the spread of the center.

In any of the above schemes, it is preferable to set σ=D ₀ to obtain a Gaussian low-pass filter

where _D0 is the cutoff frequency.

The second purpose of the invention is to provide a Gaussian equation-based automatic driving vehicle lane change planning system, including the following modules:

Waypoint definition module: used to define local waypoint format;

Calculation module: used to calculate the obstacle center, that is, the longitudinal length, plan function design, and calculate the translation amount Δh of each waypoint;

Marking module: used to mark the translation amount Δh to the abscissa change amount Δx of the corresponding waypoint. Preferably, each waypoint is composed of global coordinates (X, Y), local coordinates (x, y) and abscissa variation Δx.

In any of the solutions above, it is preferable that the global coordinates are a coordinate system that takes into account the positions of vehicles, obstacles, maps, etc., and the origin of the coordinates does not change.

In any of the above schemes, it is preferred that the local coordinates take the front of the vehicle as the origin, the direction of movement of the vehicle as the vertical axis, and the direction perpendicular to the direction of movement as the horizontal axis. The origin of the coordinates changes in real time with the vehicle body .

In any of the above schemes, preferably, the calculation module is used to calculate the longitudinal depth s of the obstacle, that is, the local coordinates (j, i) of the center of the obstacle.

In any of the above schemes, preferably, the calculation module is also used to take the longitudinal distance of the path as an independent variable, and obtain the translation percentage H(y) corresponding to each distance obtained according to the Gaussian low-pass filter function, the translation The formula for calculating the percentage H(y) is

Among them, D ₀ bit is about the measure of the expansion of the center, and y is the distance from the center of the frequency rectangle.

In any of the above schemes, preferably, the formula for calculating the translation amount Δh of each of the waypoints is:

△h=h×H(y)

Among them, h is the maximum translation distance value.

In any of the above schemes, it is preferred that the step 3 includes considering the longitudinal length of the obstacle to ensure the safety of lane keeping after changing lanes, and obtain the following translation distribution function

Among them, T is the value range of the longitudinal distance, which is symmetrical about the center point of the obstacle.

In any of the solutions above, preferably, the marking module is used to correct the new local coordinates through Δx in the global.

Preferably in any of the above schemes, the Gaussian low-pass filter function is

Among them, μ is the probability density function value of normal distribution, and ^σ2 is the variance.

In any of the above schemes, it is preferred that a random variable X obeys the distribution of the function f(x, μ, σ), denoted as XN(μ, σ ² ).

Preferably in any of the above schemes, based on the characteristics of the Gaussian probability distribution, the Gaussian function can perform the evolution of the filter function in the frequency domain, and the following formula is obtained:

Among them, D(u, v) is the distance from the center of the frequency rectangle, and σ is a measure of the spread of the center.

In any of the above schemes, it is preferable to set σ=D ₀ to obtain a Gaussian low-pass filter

where _D0 is the cutoff frequency.

The present invention proposes a lane-changing planning method for autonomous driving vehicles based on Gaussian equations, which solves the problem of trajectory mutation at the source of lane-changing, and not only can plan lane-changing paths and lane keeping paths, but also can plan to return to the original trajectory Path, based on the characteristics of the Gaussian function, the adjustment of the lane change path only needs to adjust the cut-off frequency of the Gaussian function, the method is simple and efficient.

Description of drawings

FIG. 1 is a flow chart of a preferred embodiment of the Gaussian equation-based lane change planning method for an automatic driving vehicle according to the present invention.

FIG. 1A is a distribution diagram of Gaussian low-pass filter functions with different cutoff frequencies according to the Gaussian equation-based lane-changing planning method for an automatic driving vehicle of the present invention as shown in FIG. 1 .

FIG. 1B is an azimuth diagram of a single route under regular terrain according to the embodiment of the Gaussian equation-based lane change planning method for an automatic driving vehicle shown in FIG. 1 of the present invention.

FIG. 1C is a flowchart of a method for planning a waypoint index during a lane change in the embodiment shown in FIG. 1 of the Gaussian equation-based lane change planning method for an automatic driving vehicle according to the present invention.

FIG. 1D is an example diagram of trajectory planning of an unmanned vehicle according to the embodiment shown in FIG. 1 of the Gaussian equation-based lane-changing planning method for an automatic driving vehicle according to the present invention.

FIG. 2 is a block diagram of a preferred embodiment of the lane change planning system for automatic driving vehicles based on Gaussian equations according to the present invention.

Fig. 3 is a probability distribution diagram of a proposed planning algorithm according to another preferred embodiment of the Gaussian equation-based lane-changing planning method for an automatic driving vehicle of the present invention.

Detailed ways

The present invention will be further elaborated below in conjunction with the accompanying drawings and specific embodiments.

Embodiment one

Autonomous driving technology is becoming more and more mature, and path planning is the guarantee of the intelligence of unmanned vehicles. At present, the lane change method of most unmanned vehicles adopts the trajectory translation method. Although this method is simple and efficient, there is a problem of sudden changes in the trajectory. Such problems may easily lead to uneven lateral control of unmanned vehicles, resulting in a poor driving experience; At the same time, if the lane change situation remains unchanged during lane change planning, in order to ensure the consistency of the navigation path, unmanned vehicles should reduce planning as much as possible. In order to solve the above problems, the present invention proposes a lane change trajectory planning method combining trajectory translation and Gaussian low-pass filtering. By calculating the coordinates of the center point of the obstacle and the longitudinal depth of the obstacle, the Gaussian origin is moved from the origin of the local coordinates of the vehicle body to the obstacle At the same time, the longitudinal depth of the obstacle is used to segment the Gaussian function to obtain the translation percentage corresponding to each waypoint; the lateral change of each waypoint in the plan is marked with global coordinates to ensure the path when the lane change situation remains unchanged consistency. The present invention solves the problem of sudden change in trajectory at the source of lane changing, and at the same time not only can plan a lane changing path, a lane keeping path, but also can plan a path returning to the original trajectory. In addition, based on the characteristics of the Gaussian function, only the cut-off frequency of the Gaussian function needs to be adjusted for lane change path adjustment, which is simple and efficient.

As shown in FIGS. 1 and 2 , step 100 is executed, and the waypoint definition module 200 defines a local waypoint format. The waypoint structure is shown in Table 1. Each waypoint is composed of global coordinates (X, Y), local coordinates (x, y) and abscissa variation Δx, and the local path is composed of several waypoints, as shown in Fig. 1A. The local coordinates are a Cartesian coordinate system with the front of the vehicle as the origin, the direction of movement of the vehicle as the vertical axis, and the direction perpendicular to the direction of movement as the horizontal axis. The origin of the coordinates changes with the vehicle body in real time, as shown in jMi in 1A; the global coordinates take into account For the coordinate system of the position of the car, obstacle, map, etc., the origin of the coordinates does not change, as shown in yOx in Figure 1A. The local coordinate dominant function guides the unmanned vehicle to perform real-time automatic driving, and the global coordinate dominant function plans the waypoint marking when the obstacle does not change. For the unmanned vehicle, this processing method only needs to be performed once for the same lane change planning, avoiding real-time secondary planning during lane changes.

global coordinates global coordinates local coordinates local coordinates Abscissa variation x Y x the y Δx

Table 1 Local waypoint structure

Step 110 is executed, and the calculation module 210 calculates the center of the obstacle, that is, the longitudinal length. As shown in Figure 1A, the longitudinal depth s of the obstacle and the local coordinates (j, i) of the center of the obstacle are calculated.

Step 120 is executed, the calculation module 210 plans the function design, and calculates the translation amount Δh of each waypoint. The function of this function is to take the longitudinal distance of the path as an independent variable, and take the translation percentage corresponding to each distance obtained according to the Gaussian low-pass filter function.

Multiply the horizontal translation percentage on the original path by the maximum translation distance value to obtain the translation amount of each waypoint, as shown in formula 2.

△h=h×H(y) (2)

The characteristic of Gaussian low-pass filtering is that the function probability distribution is centered on 0, and decreases smoothly and symmetrically to both sides. Based on this, in order to plan the lane-changing path and the lane-turning path, the Gaussian coordinate origin needs to be moved from the local coordinate origin of the vehicle body to the center point of the obstacle by using the characteristic that the Gaussian function decreases to both sides. At the same time, in order to ensure the safety of lane keeping after changing lanes, the longitudinal length of the obstacle needs to be considered, as shown in s in Figure 1A. Within the range covered by the length of the obstacle, the translation percentage should account for 100% of the translation amount, so we can get The following translational distribution function, shown in Equation 3.

Among them, T is the value range of the longitudinal distance, which is symmetrical about the center point of the obstacle. The probability distribution function of Equation 3 is shown in Figure 1B. In Fig. 1B, different cut-off frequency curves have different degrees of decrease in the vertical axis, which provides conditions for planning paths with different curvatures in different situations. At the same time, maintain a function segment with a vertical axis value of 1 on both sides of the origin, which is the position where the translation amount remains 100% during lane keeping. The unit of abscissa in Fig. 1B is dm.

Step 130 is executed, and the marking module 220 marks the translation amount Δh to the abscissa change amount Δx of the corresponding waypoint. In this way, the unmanned vehicle can find the global coordinates corresponding to each waypoint during the lane change process, and correct the new local coordinates through the global coordinates. Because the global coordinates of each waypoint do not change, this approach ensures that only one planning is performed when the lane change situation remains unchanged, ensuring the smoothness of the unmanned vehicle lane change. As shown in FIG. 1C , step 131 is executed. During the lane change process, when the current local waypoint is obtained, the one-to-one corresponding points between the global coordinates under the first plan and the current global coordinates are firstly indexed. Execute step 132, modify the abscissa translation amount of the current local coordinates according to the abscissa variation in the first planned global coordinates, and finally obtain the current navigation path. Step 133 is executed and the unmanned vehicle performs real-time automatic driving according to the navigation route.

Through the above four steps, the realization of the lane change planning of the present invention can be completed, and the realization result is shown in FIG. 1D . The unmanned vehicle has run into an obstacle on the driving route and has the condition to change lanes to the right. At this time, according to the lane change planning method proposed by the present invention, the unmanned vehicle can obtain 4 alternative paths and each path includes a road change section, The cut-off frequency of the Gaussian low-pass function of different paths is different for the lane keeping section and returning to the original trajectory section. Figure 1D effectively proves the validity and superiority of the present invention.

Embodiment two

Gaussian low-pass filter function

The probability density function of a normal distribution with mean μ and variance ^σ2 (or standard deviation) is an instance of a Gaussian function:

If a random variable X obeys this distribution, we write X:N(μ, ^σ2 ). Based on the characteristics of the Gaussian probability distribution, the Gaussian function can perform the evolution of the filter function in the frequency domain, such as:

Among them, D(u, v) is the distance from the center of the frequency rectangle, and σ is a measure of the spread of the center. A Gaussian low-pass filter can be obtained by setting σ=D ₀ :

Among them, D ₀ is the cut-off frequency, and the probability distribution diagram of H(u, v) with different cut-off frequencies is shown in Fig. 2 .

Embodiment Three

A common lane-changing method for autonomous vehicles is to directly shift the current local path by a fixed distance based on the position of the obstacle, and the new path obtained after the translation is the lane-changing path, as shown in the planned path 0 in Figure 2. The lane-changing path planned in this way does not take into account the future trajectory of unmanned vehicles, and lacks the planning to return to the original path after changing lanes. For unmanned vehicles, it is a step-type path change, so it will cause the lateral control of unmanned vehicles to vibrate easily and poor comfort during lane change.

The gist of the present invention is to obtain a smooth planning curve on the basis of shifting and changing lanes, that is, taking into account the future trajectory of the unmanned vehicle, and adding the planning of returning to the original trajectory after changing lanes.

Embodiment four

Compared with other similar prior art, the technical features of this application are as follows:

1. The technical circuit is different from the past. This application adds a Gaussian low-pass function on the basis of trajectory translation to plan lane-changing trajectories. The planned trajectory includes not only lane-changing trajectories, but also lane-keeping trajectories and lane-returning original path trajectories.

2. There are different ways to solve the problem of driving comfort when changing lanes. Directly use Gaussian function to smooth the trajectory, and solve the problem of poor driving comfort caused by sudden change of lane angle from the source.

3. The planning times are different. Only one local path planning is carried out when the lane changing conditions are satisfied, and the lateral transformation of the planned path is marked in the global coordinates, so that the new periodic path only needs to use the global coordinates to index the change of the local coordinates during the lane changing process. The consistency of the lane change path is guaranteed.

4. The requirements for sensor configuration are inconsistent. There are no special requirements for sensors, whether it is radar, navigation, images, etc., as long as they can provide road guidance lines, the applicability is wider.

5. The nature of the technology is different. This application is based on meeting the lane changing conditions, that is to say, the focus is on lane changing path planning after meeting the lane changing conditions, and does not involve the selection of lane changing conditions;

6. The lane change control methods are different. The final result of this application is the lane change path, not the lane change control amount.

7. The scope of coverage is different. This application is based on the trajectory planned by the safe area, that is to say, the maximum value of the translation trajectory of this application is the width of the lane, and the lane keeping section is the longitudinal depth of the obstacle.

In order to better understand the present invention, the above has been described in detail in conjunction with specific embodiments of the present invention, but it is not intended to limit the present invention. Any simple modification made to the above embodiments according to the technical essence of the present invention still belongs to the scope of the technical solution of the present invention. What each embodiment in this specification focuses on is the difference from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. As for the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the related parts, please refer to the part of the description of the method embodiment.

Claims (10)

1. a kind of automatic driving vehicle lane-change planing method based on Gauss equation, includes the following steps：

Step 1：The local waypoint format of definition；

Step 2：Calculate barrier center, that is, longitudinal length；

Step 3：It plans function design, and calculates the translational movement △ h of each waypoint；

Step 4：The translational movement △ h are tagged to the abscissa variable quantity △ x of corresponding waypoint.

2. the automatic driving vehicle lane-change planing method based on Gauss equation as described in claim 1, it is characterised in that：It is described Each waypoint is by world coordinates (X, Y) and local coordinate (x, y) and abscissa variable quantity △ x compositions.

3. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 2, it is characterised in that：It is described World coordinates allows for the coordinate system of the positions such as vehicle, barrier, map, and coordinate origin does not change.

4. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 3, it is characterised in that：It is described Local coordinate is using headstock as origin, and the direction of motion of vehicle is the longitudinal axis, and the direction vertical with the direction of motion is that the right angle of horizontal axis is sat Mark system, coordinate origin change in real time with car body.

5. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 4, it is characterised in that：It is described Step 2 is to calculate barrier longitudinal direction depth s, that is, barrier center local coordinate (j, i).

6. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 5, it is characterised in that：It is described Step 3 includes taking each that foundation Gassian low-pass filter function obtains apart from right using the fore-and-aft distance in path as independent variable The translation percentage H (y) answered, the calculation formula for translating percentage H (y) are

Wherein, D₀Measurement of the position about the divergence at center, y is the distance according to frequency rectangular centre.

7. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 6, it is characterised in that：It is described The calculation formula of the translational movement △ h of each waypoint is

△ h=h × H (y)

Wherein, h is maximal translation distance value.

8. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 7, it is characterised in that：It is described Step 3 includes considering the longitudinal length of barrier to ensure track is kept after lane-change safety, obtains translating distribution letter below Number

Wherein, T is fore-and-aft distance value range, about barrier center point symmetry.

9. the automatic driving vehicle lane-change planing method based on Gauss equation as claimed in claim 8, it is characterised in that：It is described Step 4 includes correcting new local coordinate by the △ x in the overall situation.

10. a kind of automatic driving vehicle lane-change planning system based on Gauss equation, comprises the following modules：

Waypoint definition module：For defining local waypoint format；

Computing module：For calculating barrier center i.e. longitudinal length, planning function design, and calculate each waypoint Translational movement △ h；

Mark module：Abscissa variable quantity △ x for the translational movement △ h to be tagged to corresponding waypoint.