CN110705388A - A lane change recognition method for assisted driving based on predictive feedback - Google Patents

Abstract

The invention proposes a method for recognizing a lane change of a target vehicle for assisted driving based on prediction feedback. First, the motion state information and road structure information of the surrounding target vehicles are collected in real time to obtain the motion state of the target vehicle in the own vehicle coordinate system; then the ground coordinates are calculated. Based on the motion state of the target vehicle, extract the lane-changing intent feature of the target vehicle at the current moment; initially identify the lane-changing result of the target vehicle according to the above-mentioned lane-changing intent feature; use polynomial fitting and optimization method to predict the target vehicle on the preliminary identification result Motion trajectory; take the motion trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle, calculate the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory, and check the preliminary recognition result as the final target vehicle lane change recognition result . The invention improves the accuracy of identifying the lane-changing of the target vehicle under the observation noise, and enhances the robustness of the method for identifying the lane-changing of the target vehicle.

Description

A lane change recognition method for assisted driving based on predictive feedback

technical field

The invention relates to the technical field of environment perception of an advanced assisted driving system, in particular to a method for identifying a lane change of a target vehicle for assisted driving based on prediction feedback.

Background technique

The dynamic changes of other moving vehicles in the road range have a crucial impact on road traffic safety. In the advanced assisted driving system, the selection of the driving strategy of the intelligent networked vehicle needs to consider the motion behavior of other vehicles. Vehicle motion behaviors on structured roads include lane changing and lane keeping, among which vehicle lane changing behavior poses a greater threat to road traffic safety. According to the statistics of the European Union, the number of traffic accidents caused by lane changing accounts for 4%-10% of all traffic accidents, and the traffic delay time caused by lane changing traffic accidents accounts for 10% of the total delay time. Research by the U.S. Department of Transportation shows that it takes 1.5 seconds to recognize a vehicle's lane-changing behavior in order to have enough time to react. Therefore, identifying the lane-changing behavior of vehicles on structured roads in advance can effectively avoid lane-changing traffic accidents.

The identification of the lane-changing behavior of the target vehicle (that is, the vehicle to be identified) mainly includes two steps: first, extracting the target vehicle's lane-changing intent feature;

Existing research uses two types of lane-changing intention feature quantities: target vehicle motion state parameters and driving environment state parameters. The motion state parameters of the target vehicle include the lateral and longitudinal motion speeds of the target vehicle and the lateral distance between the vehicle and the lane line. The driving environment state parameters describe the relative motion relationship between the target vehicle and other vehicles on the road, such as the distance and relative speed between the target vehicle and the preceding vehicle in the same lane. The contribution of lane-changing intention feature in identifying lane-changing behavior is not equal. Schlechtriemen et al. compared the role and correlation of different feature quantities in identifying lane-changing behavior, and obtained the longitudinal relationship between the target vehicle and the preceding vehicle in the same lane. Speed, the lateral movement speed of the target vehicle and the lateral distance between the target vehicle and the lane line are the three most effective lane-changing intention feature quantities.

For lane change intention recognition algorithms, there are currently two categories: rule-based recognition methods and machine learning-based recognition methods. The rule-based recognition method usually proposes a series of conditions for the occurrence of lane-changing. When the feature quantity of the target vehicle's lane-changing intention meets the conditions for the occurrence of lane-changing, it is considered that the target vehicle is changing lanes. For example, Monot et al. obtained the probability value of the target vehicle changing lanes according to the lateral movement speed of the target vehicle and the lateral distance of the target vehicle from the lane line. When the probability value is greater than 50%, it is considered that the target vehicle will change lanes. The rule-based algorithm has high computational efficiency and is suitable for advanced assisted driving systems that require high real-time performance, but the recognition accuracy is not high. The methods based on machine learning include support vector machine SVM, hidden Markov model HMM, Bayesian decision-making and neural network. The lane-changing recognition method based on machine learning can handle high-dimensional feature input, and the recognition accuracy is high, but it requires high computing resources.

Existing target vehicle lane-changing recognition methods use multi-dimensional lane-changing intent feature quantities, and mostly use machine learning recognition algorithms. .htm) on the test results show that the lane change recognition accuracy is above 90%. However, the motion parameters and driving environment parameters of the target vehicle observed by the vehicle-mounted sensor system have observation noise, which has a great impact on the recognition accuracy of the lane-changing recognition algorithm. The vehicle lane change recognition accuracy rate is only 85%, while the recall rate is only 59%. In order to solve this problem, Woo et al. predicted the lane-changing trajectory of the target vehicle and the trajectory of other vehicles on the road on the basis of the recognition results based on the lane-changing feature quantity. If the future trajectory of the target vehicle intersects with the future trajectory of other vehicles, there is a collision risk. , indicating that the predicted trajectory of the target vehicle is unreasonable, and further indicates that the recognition result of the target vehicle's lane-changing state is unreasonable and should be corrected. Through this method of predictive trajectory feedback correction, Woo et al. improved the accuracy of lane change intention recognition. However, this feedback correction is suitable for high-traffic conditions, requiring other vehicles around the target vehicle to be used, and cannot be used in high-speed conditions with small traffic flow.

SUMMARY OF THE INVENTION

In order to solve the problem of insufficient robustness of the target vehicle lane change identification method in the presence of observation noise in the vehicle-mounted sensor system, the present invention proposes a target vehicle lane change identification method based on trajectory prediction feedback.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for identifying a lane change of a target vehicle for assisted driving based on prediction feedback proposed by the present invention is characterized in that it includes the following steps:

S1 collects the motion state information and road structure information of the surrounding target vehicles in real time through the sensor system mounted on the ICV, and obtains the motion state of the target vehicle at the current moment in the self-vehicle coordinate system after processing by lane line recognition and target detection and tracking algorithms. Road lane line equation;

S2 calculates the motion state of the target vehicle and the rectangular envelope of the target vehicle at the current moment in the ground coordinate system according to the motion state of the target vehicle and the road lane line equation at the current moment in the self-vehicle coordinate system, and extracts the current moment of the target vehicle in the ground coordinate system. Lane-changing intention feature quantity; the lane-changing intention feature quantity includes the lateral velocity _vy of the target vehicle's center of mass along the direction perpendicular to the road lane line, the lateral distance _dy from the rectangular envelope frame of the target vehicle to the road lane line, and in the following scenario the longitudinal speed Δv _x and the longitudinal distance d _x of the target vehicle along the tangential direction of the road lane line relative to the moving vehicle in front of the same lane;

S3 defines the target vehicle's predicted lane-changing time and following risk coefficient as the lane-changing criterion according to the feature quantity of the target vehicle's lane-changing intention at the current moment in the extracted ground coordinate system; and obtains the predicted lane-changing time and following vehicle according to the statistics of the HighD data set The lane-changing threshold of the risk factor; if the predicted lane-changing time or the following risk factor is greater than the corresponding target vehicle's lane-changing threshold, it will preliminarily identify that the target vehicle is changing lanes or is about to change lanes, and determine whether to turn left or right according to the sign of the predicted lane-changing time. Change lanes to the right, otherwise initially identify the target vehicle and keep the current lane;

S4, according to the preliminary identification result obtained in step S3, use polynomial fitting and optimization method to predict the motion trajectory of the target vehicle: first determine the target vehicle motion trajectory constraint, use polynomial fitting to obtain a motion trajectory that conforms to the motion trajectory constraint, and then use the optimization method The most probable motion trajectory is obtained by screening as the predicted motion trajectory of the target vehicle;

S5 takes the motion trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle; calculates the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory of the target vehicle, and if the deviation exceeds the set trajectory deviation threshold, the target vehicle is determined The predicted motion trajectory does not conform to the motion trajectory constraint, that is, the preliminary identification result does not conform to the motion constraint. If the preliminary identification result is no lane change, it is corrected to change lane, and if the preliminary identification result is lane change, it is corrected to no lane change. After feedback correction The final target vehicle lane change recognition result is obtained; if the deviation value does not exceed the set trajectory deviation threshold, the preliminary recognition result remains unchanged as the final target vehicle lane change recognition result.

Features and beneficial effects of the present invention:

The present invention firstly uses the lane-changing intention feature quantity to identify whether the target vehicle changes lanes, and according to the lane-changing intention identification result, uses polynomial fitting to predict the target vehicle motion trajectory that conforms to the lane-changing intention, and at the same time uses the uniform acceleration motion model to generate the target conforming to the motion constraints The vehicle reference trajectory, if the predicted lane change trajectory is inconsistent with the reference trajectory in a short period of time, the lane change recognition result will be corrected, that is, the correction identified as a lane change is no lane change, and the correction identified as no lane change is a lane change. Due to the motion inertia of the vehicle, its motion is closer to the uniform acceleration motion model in a short period of time, so the motion trajectory generated by the uniform acceleration model is used as a reference trajectory in a short time to check whether the predicted target vehicle lane changing trajectory conforms to the motion inertia constraint. In addition, since the curvature of the motion trajectory generated by the polynomial fitting is continuous and conforms to the actual motion of the vehicle, the present invention adopts the polynomial fitting to predict the motion trajectory of the target vehicle.

The test results on the HighD data set (https://www.highd-dataset.com/) released by RWTH Aachen University in Germany show that by adding trajectory prediction feedback to the recognition results based on feature quantities, the present invention improves the performance under observation noise. The accuracy of target vehicle lane change recognition enhances the robustness of the target vehicle lane change recognition method.

Description of drawings

FIG. 1 is a statistical probability density distribution diagram of a predicted lane change time feature in an embodiment of the present invention.

FIG. 2 is a statistical probability density distribution diagram of a vehicle-following risk coefficient feature quantity in an embodiment of the present invention.

FIG. 3 is a schematic diagram of generating a reference motion trajectory in an embodiment of the present invention.

FIG. 4 is a schematic diagram of a fitted motion trajectory in an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings.

A method for identifying a lane change of a target vehicle for assisted driving based on trajectory prediction feedback proposed by the present invention includes the following steps:

S1 collects the motion state information and road structure information of surrounding target vehicles in real time through the sensor system mounted on the intelligent networked vehicle, and obtains the motion of the target vehicle at the current moment in the self-vehicle coordinate system after processing by conventional lane line recognition and target detection and tracking algorithms. State and road lane line equations;

S2 calculates the motion state of the target vehicle and the rectangular envelope of the target vehicle at the current moment in the ground coordinate system according to the motion state of the target vehicle and the road lane line equation at the current moment in the own vehicle coordinate system obtained in step S1, and extracts the current moment in the ground coordinate system. The lane-changing intention feature quantity of the target vehicle. The lane-changing intention feature quantity includes the lateral velocity _vy of the target vehicle's center of mass along the direction perpendicular to the road lane line, the lateral distance _dy from the rectangular envelope frame of the target vehicle to the road lane line, and the following scenario. Information such as the longitudinal speed Δv _x and the longitudinal distance d _x of the target vehicle relative to the moving vehicle in front of the same lane along the tangential direction of the road lane line; the specific steps include:

S2.1 Convert the motion state of the target vehicle and the road lane line equation at the current time in the own vehicle coordinate system obtained in step S1 into the motion state of the target vehicle at the current time t ₀ in the ground coordinate system [x(t ₀ ), y(t ₀ ), v _x (t ₀ ), v _y (t ₀ ), a _x (t ₀ ), a _y (t ₀ )] and the target vehicle rectangular envelope, where [x(t ₀ ), y(t ₀ )] is the position coordinate of the center of mass of the target vehicle at the current moment, v _x (t ₀ ), v _y (t ₀ ) are the longitudinal and lateral velocity of the target vehicle at the current moment, respectively, a _x (t ₀ ), a _y (t ₀ ) are the longitudinal acceleration and lateral acceleration of the target vehicle at the current moment, respectively;

S2.2 According to the motion state of the target vehicle in the ground coordinate system and the lane line equation, calculate the lateral velocity _vy of the target vehicle's center of mass along the vertical lane line direction at the current moment and the vertical distance _dy between the rectangular envelope frame of the target vehicle and the lane line;

S2.3 judges whether there is a moving vehicle in front of the target vehicle in the same lane at the current moment, if so, calculate the longitudinal speed Δv _x and the longitudinal distance d _x of the target vehicle along the tangential direction of the road lane line relative to the moving vehicle in front of the same lane;

S3 defines the target vehicle's predicted lane-changing time TTLC and the following risk coefficient RP as the lane-changing criterion according to the extracted feature quantity of the target vehicle's lane-changing intention at the current moment in the ground coordinate system; and obtains the predicted lane-changing time according to the statistics of the HighD data set and the lane-changing threshold of the following risk factor; if the predicted lane-changing time or the following risk factor is greater than the corresponding target vehicle lane-changing threshold, the target vehicle is preliminarily identified as changing lanes or is about to be changed, and the direction of the vehicle is determined according to the sign of the predicted lane-changing time. Change lanes left or right, otherwise the target vehicle is initially identified to keep the current lane; it includes the following steps:

S3.1 According to the extracted feature quantity of the target vehicle's lane-changing intention at the current moment in the ground coordinate system, define the predicted lane-changing time TTLC and the following risk coefficient RP as the target vehicle's lane-changing criterion:

The predicted lane change time _{TTLC is defined as the ratio of the vertical distance dy between the rectangular envelope of the target vehicle and the road lane line at the current moment in the ground coordinate system and the lateral velocity vy of} the target vehicle's center of mass along the vertical lane line. The expression is as follows:

The following risk factor RP is defined as follows:

a) If there is no moving vehicle in front of the same lane of the target vehicle at the current moment, set the following risk coefficient RP to zero;

b) If there is a moving vehicle in front of the same lane of the target vehicle at the current moment, the following risk coefficient RP is defined as the weighted sum of the headway THW and the collision time TTC of the target vehicle, and the expression is as follows:

Wherein, the headway of the target vehicle THW=d _x /v _x , the collision time of the target vehicle TTC=d _x /Δv _x ; v _x is the longitudinal speed of the target vehicle along the tangential direction of the lane line, and d _x is the relative speed of the target vehicle The longitudinal distance of the moving vehicle in front of the lane along the tangential direction of the road lane line, Δv _x is the longitudinal speed of the target vehicle relative to the moving vehicle in front of the same lane along the tangential direction of the lane line, a, b are the headway and collision time of the target vehicle at the current moment, respectively Adjust the value of the weight coefficients a and b of the headway and the collision time, and test on the HighD data set, select the optimal weight coefficients a, b to make lane-changing vehicles and non-lane-changing vehicles follow. The difference in the risk factor RP value is the largest.

S3.2 Count the predicted lane-changing time TTLC and the following risk coefficient RP of lane-changing vehicles and non-lane-changing vehicles in the HighD data set, as shown in Figures 1 and 2, determine the predicted lane-changing time TTLC and the following risk coefficient RP The thresholds of α and β are used as the lane-changing thresholds α and β of the target vehicle, respectively, so that the accuracy of the test lane-changing recognition on the HighD data set is the highest.

S3.3 According to the lane-changing threshold of the target vehicle obtained in step S3.2, the lane-changing state of the target vehicle is divided into lane-changing to the left, lane-changing to the right, and no lane-changing; The judgment conditions are:

LCL=(0<TTLC<α)∪(RP>β) (3)

LCR=(-α＜TTLC＜0)∪(RP＜-β) (4)

In the formula, LCL indicates that the lane-changing status of the target vehicle is to change lanes to the left, LCR indicates that the lane-changing status of the target vehicle is to change lanes to the right; TTLC and RP are the predicted lane-changing time and Vehicle-following risk coefficient; α, β are respectively the predicted lane-changing time and the threshold of the vehicle-following risk coefficient of the target vehicle set in step S3.2; when neither the lane-changing conditions to the left nor the lane-changing conditions to the right are satisfied That is, do not change lanes.

S4 uses polynomial fitting and optimization methods to predict the trajectory of the target vehicle according to the preliminary identification results: first determine the trajectory constraints of the target vehicle, use polynomial fitting to obtain the trajectory that meets the trajectory constraints, and then filter through the optimization method to obtain the most likely trajectory The motion trajectory of the target vehicle is used as the predicted motion trajectory of the target vehicle; it specifically includes the following steps:

S4.1 uses polynomial fitting to fit the trajectory of the target vehicle that conforms to the lane-changing state

S4.11 Determine the starting point constraint of the motion trajectory of the target vehicle, and set the state of the starting point of the motion trajectory of the target vehicle as [x ₀ , y ₀ , v _x0 , v _y0 , a _x0 , a _y0 ], where [x ₀ , y ₀ ] is the starting point coordinate of the motion track, which is equal to the position coordinate of the target vehicle at the current moment, that is, x ₀ =x(t ₀ ), y ₀ =y(t ₀ ); [v _x0 , v _y0 ] is the starting point speed of the motion track, [ a _x0 , a _y0 ] is the acceleration of the starting point of the motion trajectory. Due to the noise in the state observation of the target vehicle, the velocity and acceleration of the starting point of the motion trajectory are obtained by means of mean filtering, that is, the speed and acceleration of the starting point of the motion trajectory are the speed and acceleration of the target vehicle in the past period of time Δt. The mean value of acceleration, the calculation formula is shown in (5)~(8):

v _x0 = mean(v _x (t)) (5)

v _y0 = mean(v _y (t)) (6)

a _x0 = mean(a _x (t)) (7)

a _y0 =mean(a _y (t)) (8)

Among them, t∈[t ₀ -Δt, t ₀ ], Δt is the time length of the historical trajectory; v _x (t), v _y (t) are the lateral velocity function perpendicular to the lane line on the historical trajectory of the target vehicle and the horizontal velocity function along the lane, respectively The longitudinal velocity function along the tangential direction of the line; a _x (t) and a _y (t) are the lateral acceleration function perpendicular to the lane line and the longitudinal acceleration function tangential along the lane line on the historical trajectory of the target vehicle, respectively.

S4.12 Determine the end point constraint of the target vehicle's motion trajectory. If the lane changing state is to change lanes to the left, then make the end point of the trajectory on the center line of the left lane; On the center line of the lane, if the lane change status is not changing lanes, the end point of the trajectory is set on the center line of the current lane; the motion constraint of the end point of the target vehicle motion trajectory is:

y ₁ =y ₀ +Δy (9)

v _y1 = 0 (10)

a _y1 = 0 (11)

v _x1 = v _x0 +a _x0 t _pred (12)

a _x1 = a _x0 (13)

in,

Equation (9) is expressed as, the lateral coordinate y ₁ of the target vehicle's center of mass at the end point of the constraint trajectory perpendicular to the lane line is equal to the sum of the lateral coordinate y ₀ of the target vehicle's center of mass at the starting point of the trajectory and the lateral motion offset Δy of the target vehicle;

Formulas (10) and (11) are expressed as, respectively constraining the target vehicle lateral velocity v _y1 and the target vehicle lateral acceleration a _y1 perpendicular to the lane line at the end point of the trajectory to be zero;

Formula (12) is expressed as, the longitudinal motion of the target vehicle along the lane line is constrained to conform to uniform acceleration motion, v _x0 , v _x1 are the longitudinal velocity of the target vehicle along the tangential direction of the lane line at the starting point and the end point of the trajectory, respectively, a _x0 is the target vehicle Longitudinal acceleration, t _pred is the predicted movement duration from the track start point to the track end point, and the track end point time t ₁ =t ₀ +t _pred ;

Equation (13) is expressed as, the longitudinal acceleration a _x1 along the tangential direction of the lane line at the end point of the constraint trajectory is equal to the longitudinal acceleration a _x0 along the tangential direction of the lane line at the starting point of the trajectory.

S4.13 Fits the lateral movement of the target vehicle perpendicular to the lane line and the longitudinal movement along the tangential direction of the lane line with a 5th-order polynomial and a 4th-order polynomial respectively, and uses the least squares method to obtain the polynomial coefficients. Set the lateral motion trajectory equation of the target vehicle for:

y(t)=c _y5 t ⁵ +c _y4 t ⁴ +c _y3 t ³ +c _y2 t ² +c _y1 t+c _y0 (14)

Then the constraints of the starting point and the ending point of the lateral motion trajectory of the target vehicle are written as:

Similarly, the longitudinal motion trajectory equation of the target vehicle is set as:

x(t)=c _x4 t ₄ +c _x3 t ₃ +c _x2 t ₂ +c _x1 t+c _x0 (16)

Then the constraints of the starting point and end point of the longitudinal motion trajectory of the target vehicle are written as:

S4.2 Because the predicted movement time t _pred is different, the target vehicle lane change trajectory obtained by fitting is not unique. As shown in Figure 3, the optimization method is used to select the K candidate movement trajectories obtained in S4.13. The most likely motion trajectory, the optimized cost function includes motion time and lateral acceleration, and the optimized cost function used is:

C(T ⁱ )=t ⁱ +γmax(a _y (t)) (18)

Among them, t ⁱ is the predicted motion time corresponding to the ^ith candidate motion trajectory Ti of the target vehicle, i ₌ 1, 2, . The lateral acceleration function of the direction; γ is the coefficient, by selecting any vehicle motion trajectory in the HighD data set, adjusting the coefficient in the optimization cost function, so that the selected predicted trajectory and the real motion trajectory of the target vehicle are determined in a close manner;

The final selected trajectory minimizes the optimization cost function, which is used as the predicted motion trajectory T _opt of the target vehicle, and the expression is as follows:

T _opt =argmin(C(T ⁱ )) _i=1,2,...,K (19)

S5 takes the motion trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle; calculates the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory of the target vehicle, and if the deviation exceeds the set trajectory deviation threshold, the target vehicle is determined The predicted motion trajectory does not meet the motion trajectory constraint, that is, the initial lane change recognition result does not meet the motion constraint. If the initial lane change recognition result is no lane change, it will be corrected to lane change, and if the initial lane change recognition result is lane change, it will be corrected to no lane change. After feedback correction, the final target vehicle lane change recognition result is obtained; if the deviation value does not exceed the set trajectory deviation threshold, the preliminary recognition result remains unchanged as the final target vehicle lane change recognition result. Specifically include the following steps:

S5.1 takes the trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle

Due to the state inertia of vehicle motion, the vehicle motion is close to uniform acceleration in a short period of time, so the lane change trajectory that conforms to the uniform acceleration motion constraint is used as the reference motion trajectory of the target vehicle to check the predicted motion trajectory of the target vehicle. Let the motion state of the starting point of the reference motion trajectory of the target vehicle be [x ₀ , y ₀ , v _x0 , v _y0 , a _x0 , a _y0 ], and the target vehicle maintains uniform acceleration to obtain the reference trajectory within a period of time Δt _r :

where t ₂ =t ₀ +Δt _r . As shown in FIG. 4 , in a short time Δt _r (Δt _r is taken as 1 to 5 s, in this embodiment is 3 s), the uniform acceleration motion trajectory of the vehicle is relatively close to the real motion trajectory of the target vehicle. The value of Δt _r is chosen so that the uniform acceleration trajectory of any vehicle in the HighD dataset is close to the real trajectory within the time Δt _r . x _r (t) and y _r (t) are respectively the reference trajectory of the target vehicle’s lateral motion along the tangential direction of the lane line and the longitudinal motion reference trajectory of the vertical lane line within the prediction time range.

S5.2 Calculate the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory of the target vehicle

The deviation between the predicted motion trajectory of the target vehicle obtained in step S4 and the above-mentioned reference motion trajectory of the target vehicle is defined as the distance sum of discrete trajectory points, and the calculation formula is as follows:

Among them, N is the total number of discrete prediction durations, g is each discrete point within the prediction duration, and δt is the discrete time step of the prediction duration. x(t _g ), y(t _g ) are the horizontal and vertical predicted motion trajectories of the target vehicle discretized in the prediction time, respectively, x _r (t _g ), y _r (t _g ) are the discretized targets in the prediction time, respectively Vehicle lateral and vertical reference motion trajectories.

S5.3 If the trajectory deviation is greater than the set trajectory deviation threshold, the lane change recognition state does not conform to the motion constraints. If the initial lane change recognition is a lane change, the target vehicle state is corrected from lane change to no lane change, and vice versa. The lane change correction is lane change, and the final lane change recognition result is obtained after feedback correction.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, and substitutions can be made in these embodiments without departing from the principle and spirit of the invention and modifications, the scope of the present invention is defined by the appended claims and their equivalents.

Claims (5)

1. a target vehicle lane change identification method for assisted driving based on predictive feedback, is characterized in that, comprises the following steps: S1 collects the motion state information and road structure information of the surrounding target vehicles in real time through the sensor system mounted on the ICV, and obtains the motion state of the target vehicle at the current moment in the self-vehicle coordinate system after processing by lane line recognition and target detection and tracking algorithms. Road lane line equation; S2 calculates the motion state of the target vehicle and the rectangular envelope of the target vehicle at the current moment in the ground coordinate system according to the motion state of the target vehicle and the road lane line equation at the current moment in the self-vehicle coordinate system, and extracts the current moment of the target vehicle in the ground coordinate system. Lane-changing intention feature quantity; the lane-changing intention feature quantity includes the lateral velocity _vy of the target vehicle's center of mass along the direction perpendicular to the road lane line, the lateral distance _dy from the rectangular envelope frame of the target vehicle to the road lane line, and in the following scenario the longitudinal speed Δv _x and the longitudinal distance d _x of the target vehicle along the tangential direction of the road lane line relative to the moving vehicle in front of the same lane; S3 defines the target vehicle's predicted lane-changing time and following risk coefficient as the lane-changing criterion according to the feature quantity of the target vehicle's lane-changing intention at the current moment in the extracted ground coordinate system; and obtains the predicted lane-changing time and following vehicle according to the statistics of the HighD data set The lane-changing threshold of the risk factor; if the predicted lane-changing time or the following risk factor is greater than the corresponding target vehicle's lane-changing threshold, it will preliminarily identify that the target vehicle is changing lanes or is about to change lanes, and determine whether to turn left or right according to the sign of the predicted lane-changing time. Change lanes to the right, otherwise initially identify the target vehicle and keep the current lane; S4, according to the preliminary identification result obtained in step S3, use polynomial fitting and optimization method to predict the motion trajectory of the target vehicle: first determine the target vehicle motion trajectory constraint, use polynomial fitting to obtain a motion trajectory that conforms to the motion trajectory constraint, and then use the optimization method The most probable motion trajectory is obtained by screening as the predicted motion trajectory of the target vehicle; S5 takes the motion trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle; calculates the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory of the target vehicle, and if the deviation exceeds the set trajectory deviation threshold, the target vehicle is determined The predicted motion trajectory does not conform to the motion trajectory constraint, that is, the preliminary identification result does not conform to the motion constraint. If the preliminary identification result is no lane change, it is corrected to change lane, and if the preliminary identification result is lane change, it is corrected to no lane change. After feedback correction The final target vehicle lane change recognition result is obtained; if the deviation value does not exceed the set trajectory deviation threshold, the preliminary recognition result remains unchanged as the final target vehicle lane change recognition result. 2. The method for identifying a lane change of a target vehicle according to claim 1, wherein the step S3 specifically comprises the following steps: S3.1 According to the extracted feature quantity of the target vehicle's lane-changing intention at the current moment in the ground coordinate system, define the predicted lane-changing time TTLC and the following risk coefficient RP as the target vehicle's lane-changing criterion: The predicted lane change time _{TTLC is defined as the ratio of the vertical distance dy between the rectangular envelope frame of the target vehicle and the road lane line at the current moment in the ground coordinate system and the lateral speed vy of} the target vehicle's centroid along the direction of the vertical road lane line, the expression as follows:

The following risk factor RP is defined as follows: a) If there is no moving vehicle in front of the same lane of the target vehicle at the current moment, set the following risk coefficient RP to zero; b) If there is a moving vehicle in front of the same lane of the target vehicle at the current moment, the following risk coefficient RP is defined as the weighted sum of the headway THW and the collision time TTC of the target vehicle, and the expression is as follows:

in, The headway of the target vehicle THW=d _x /v _x , d _x is the longitudinal distance of the target vehicle along the tangential direction of the road lane line relative to the moving vehicle in front of the same lane at the current moment, v _x is the tangential direction of the target vehicle along the lane line at the current moment longitudinal speed; Collision time of the target vehicle TTC=d _x /Δv _x , Δv _x is the longitudinal speed of the target vehicle along the tangential direction of the road lane line relative to the moving vehicle in front of the same lane at the current moment; a, b are the weight coefficients of the headway and collision time of the target vehicle at the current moment, respectively. Adjust the values of the weight coefficients a and b of the headway and collision time, and test them on the HighD data set to select the optimal weight coefficient. a, b make the difference between the following hazard coefficients RP of lane-changing vehicles and non-lane-changing vehicles to be the largest; S3.2 Count the predicted lane-changing time TTLC and the following risk coefficient RP of lane-changing vehicles and non-lane-changing vehicles in the HighD data set, and determine the thresholds of the predicted lane-changing time TTLC and the following risk coefficient RP as the lane-changing of the target vehicle, respectively. Thresholds α, β make the test lane change recognition accuracy highest on the HighD data set; S3.3 According to the lane-changing thresholds α and β of the target vehicle, the lane-changing state of the target vehicle is divided into lane-changing to the left, lane-changing to the right and no lane-changing; the judgment conditions for lane-changing to the left and lane-right are respectively for: LCL=(0<TTLC<α)∪(RP>β) (3) LCR=(-α＜TTLC＜0)∪(RP＜-β) (4) In the formula, LCL indicates that the lane-changing status of the target vehicle is to change lanes to the left, LCR indicates that the lane-changing status of the target vehicle is to change lanes to the right, and if neither the left or right lane-changing conditions are satisfied, it is not lane-changing. . 3. The method for identifying a lane change of a target vehicle according to claim 1, wherein the step S4 specifically comprises the following steps: S4.1 uses polynomial fitting to fit the trajectory of the target vehicle that conforms to the lane-changing state S4.11 Determine the starting point constraint of the motion trajectory of the target vehicle, and set the state of the starting point of the motion trajectory of the target vehicle as [x ₀ , y ₀ , v _x0 , v _y0 , a _x0 , a _y0 ], where [x ₀ , y ₀ ] is the coordinate of the starting point of the motion trajectory, which is equal to the position coordinate of the center of mass of the target vehicle at the current time t ₀ in the ground coordinate system; [v _x0 , v _y0 ] is the starting point velocity of the motion trajectory, and [a _x0 , a _y0 ] is the starting point acceleration of the motion trajectory , due to the noise in the state observation of the target vehicle, the mean value filtering is used to obtain the velocity and acceleration of the starting point of the motion trajectory, that is, the velocity and acceleration of the starting point of the motion trajectory are the mean values of the speed and acceleration of the target vehicle in the past period of time Δt, and the calculation formula is as follows (5)～( 8) shown: v _x0 = mean(v _x (t)) (5) v _y0 = mean(v _y (t)) (6) a _x0 = mean(a _x (t)) (7) a _y0 = mean(a _y (t)) (8) Among them, t∈[t ₀ -Δt, t ₀ ], Δt is the time length of the historical trajectory; v _x (t), v _y (t) are the lateral velocity function perpendicular to the lane line on the historical trajectory of the target vehicle and the horizontal velocity function along the lane, respectively longitudinal velocity function along the tangential direction of the line; a _x (t), a _y (t) are the lateral acceleration function perpendicular to the lane line and the longitudinal acceleration function tangential along the lane line on the historical trajectory of the target vehicle, respectively; S4.12 Determine the end point constraint of the target vehicle's motion trajectory. If the lane changing state is to change lanes to the left, then make the end point of the trajectory on the center line of the left lane; On the center line of the lane, if the lane change status is not changing lanes, the end point of the trajectory is set on the center line of the current lane; the motion constraint of the end point of the target vehicle motion trajectory is: y ₁ =y ₀ +Δy (9) v _y1 = 0 (10) a _y1 = 0 (11) v _x1 = v _x0 +a _x0 t _pred (12) a _x1 = a _x0 (13) in, Equation (9) is expressed as, the lateral coordinate y ₁ of the target vehicle's center of mass at the end point of the constraint trajectory perpendicular to the lane line is equal to the sum of the lateral coordinate y ₀ of the target vehicle's center of mass at the starting point of the trajectory and the lateral motion offset Δy of the target vehicle; Formulas (10) and (11) are expressed as, respectively constraining the target vehicle lateral velocity v _y1 and the target vehicle lateral acceleration a _y1 perpendicular to the lane line at the end point of the trajectory to be zero; Formula (12) is expressed as, the longitudinal motion of the target vehicle along the lane line is constrained to conform to uniform acceleration motion, v _x0 , v _x1 are the longitudinal velocity of the target vehicle along the tangential direction of the lane line at the starting point and the end point of the trajectory, respectively, a _x0 is the target vehicle Longitudinal acceleration, t _pred is the predicted movement duration from the track start point to the track end point, and the track end point time t ₁ =t ₀ +t _pred ; Formula (13) is expressed as, the longitudinal acceleration a _x1 along the tangential direction of the lane line at the end point of the constraint trajectory is equal to the longitudinal acceleration a _x0 along the tangential direction of the lane line at the starting point of the trajectory; S4.13 Fits the lateral motion perpendicular to the lane line and the longitudinal motion along the tangential direction of the lane line with a 5th-order polynomial and a 4th-order polynomial, respectively, and uses the least squares method to obtain the polynomial coefficients. Set the lateral motion trajectory equation of the target vehicle for: y(t)=c _y5 t ⁵ +c _y4 t ⁴ +cy ₃ t ³ +cy ₂ t ² +c _y1 t+c _y0 (14) Then the constraints of the starting point and the ending point of the lateral motion trajectory of the target vehicle are written as: The longitudinal motion trajectory equation of the target vehicle is set as: x(t)=c _x4 t ⁴ +c _x3 t ³ +c _x2 t ² +c _x1 t+c _x0 (16) The constraints on the start and end points of the longitudinal motion trajectory of the target vehicle are written as:

S4.2 selects the most probable motion trajectory from the K candidate motion trajectories obtained in step S4.13 by an optimization method. The optimized cost function includes the movement time and lateral acceleration, and the optimized cost function used is: C(T ⁱ )=t ⁱ +γmax(a _y (t)) (18) Among them, t ⁱ is the predicted motion time corresponding to the ^ith candidate motion trajectory Ti of the target vehicle, i=1, 2, ..., K, a _y (t) is the target vehicle from the predicted motion trajectory starting point to the predicted motion trajectory end point The lateral acceleration function in the direction of the vertical lane line during the process; γ is the coefficient, by selecting any vehicle motion trajectory in the HighD data set, adjusting the coefficient in the optimization cost function, so that the selected predicted trajectory and the real motion trajectory of the target vehicle are used as a way of approaching Sure; The final selected trajectory minimizes the optimization cost function, which is used as the predicted motion trajectory T _opt of the target vehicle, and the expression is as follows: T _opt =argmin(C(T ⁱ )) _{i=1, 2, . . . , K} (19). 4. The method for identifying a lane change of a target vehicle according to claim 3, wherein the step S5 specifically comprises the following steps: S5.1 takes the trajectory that conforms to the uniform acceleration motion constraint as the reference motion trajectory of the target vehicle Let the motion state of the starting point of the reference motion trajectory of the target vehicle be [x ₀ , y ₀ , v _x0 , v _y0 , a _x0 , a _y0 ], and the target vehicle maintains a uniform acceleration motion to obtain the reference motion trajectory within a period of time Δt _r as follows:

Among them, t ₂ =t ₀ +Δt _r ; the value of Δt _r is selected so that the uniform acceleration trajectory of any vehicle in the HighD data set within the time Δt _r is closest to its real trajectory; S5.2 Calculate the cumulative distance deviation between the predicted motion trajectory of the target vehicle and the reference motion trajectory of the target vehicle The deviation between the predicted motion trajectory of the target vehicle obtained in step S4 and the above-mentioned reference motion trajectory of the target vehicle is defined as the distance sum of discrete trajectory points, and the calculation formula is as follows:

Among them, N is the discrete total number of prediction duration, g is each discrete point within the prediction duration, and δt is the discrete time step of the prediction duration; S5.3 If the trajectory deviation is greater than the set trajectory deviation threshold, the lane change recognition state does not conform to the motion constraints. If the initial lane change recognition is a lane change, the target vehicle state is corrected from lane change to no lane change, and vice versa. The lane change correction is lane change, and the final lane change recognition result is obtained after feedback correction. 5 . The method for recognizing a lane change of a target vehicle according to claim 4 , wherein Δt _r is set to be 1 to 5 s. 6 .

Images