A method, system and device for inferring a surrounding vehicle driving behavior intention

Abstract

The application discloses a kind of inference method, system and equipment of surrounding vehicle driving behavior intention, the inference method includes the following steps: establishing Frenet coordinate system, the driving path of target vehicle is modeled as cubic spline curve;Obtain the position (s t ,q t ) and initial heading angle θ t Of target vehicle in coordinate system, determine initial path formula (1);Obtain the calculation formula (3) of the terminal point s f ;Initial lane-changing time t prev Is preset, according to initial lane-changing time t prev And the calculation formula of terminal point s f , obtain the preset terminal point s f Of path;Obtain the calculation formula (4) of coefficient a, b, c, d, obtain current period k, according to formula (1), formula (3) and formula (4), estimate current period k adaptive lane-changing time t prev,k , according to formula (3) and t prev,k , obtain the actual terminal point s f Of path;According to the actual terminal point s f Of path, obtain three paths, calculate the probability of each path, according to the path of highest probability, infer the driving behavior intention of target vehicle.

Description

Technical Field

[0001] This invention belongs to the field of intelligent driving technology for vehicles, and specifically relates to a method, system, and device for inferring the driving intentions of surrounding vehicles. Background Technology

[0002] With rapid economic development, the number of vehicles on the road has increased dramatically, leading to frequent traffic accidents and making it difficult to effectively guarantee the safety of drivers and those of surrounding vehicles. Based on driver intent, driving behaviors can be categorized as left lane changes, going straight, and right lane changes. Lane changes not only alter the driver's own behavior but also influence the driving behavior of surrounding vehicles, resulting in numerous accidents caused by errors or oversights during lane changes. Therefore, accurately predicting the driving intentions of surrounding vehicles to improve the safety of lane changes and prevent traffic accidents has become an urgent problem to solve. Summary of the Invention

[0003] This invention presents a method for inferring the driving intentions of surrounding vehicles, comprising the following steps:

[0004] Step 1: Establish a Frenet coordinate system with the s-axis and q-axis as coordinate axes, and model the driving path of the target vehicle as a cubic spline curve;

[0005] Step 2: Obtain the position of the target vehicle in the coordinate system (s t ,q t ) and initial heading angle θ t Determine the initial path formula (1):

[0006] q(s)=a(ss t ) 3 +b(ss t ) 2 +c(ss t )+d,(s t ≤s≤s f (1)

[0007] The starting coordinates of the path are (s t q t The endpoint coordinates are (s) f q f ), where q f =q i i = 1, 2, 3, the q i For each lane centerline, the value of q is given, and the range of values ​​for s is [s...]. t ,s f ], where a, b, c, and d represent coefficients;

[0008] Step 3: Obtain the endpoint s fThe calculation formula (3) is as follows:

[0009] S f =v t t prev +s t (3)

[0010] Among them, v t t represents the longitudinal velocity of the target vehicle. prev Indicates lane change time, s f -s t This indicates the longitudinal distance traveled by the target vehicle from the start of the lane change to the completion of the lane change;

[0011] Preset initial lane change time t prev ,

[0012] Based on the initial lane change time t prev and the endpoint s f The calculation formula is used to obtain the preset endpoint s of the path. f ;

[0013] Step 4: Obtain the calculation formulas (4) for coefficients a, b, c, and d.

[0014]

[0015] The unique shape of the initial path is determined based on the calculation formulas for coefficients a, b, c, and d.

[0016] Step 5: Obtain the current cycle k, and use the linear recursive least squares estimation algorithm to estimate the adaptive lane-changing time t of the current cycle k according to formulas (1), (3) and (4). prev,k ,

[0017] According to formula (3) and t prev,k Obtain the actual endpoint s of the path. f ;

[0018] Step 6: Based on the actual destination s of the path f Given three paths, calculate the probability of each path.

[0019] Based on the path with the highest probability, infer the driving intention of the target vehicle.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: After obtaining the three paths and their probabilities for the current period k, the three paths and their probabilities are updated once in each subsequent period, and the path with the highest probability is obtained to re-infer the driving intention of the target vehicle, which helps to improve the accuracy of the inference results. Through the inference method of this invention, the driving intentions of surrounding vehicles can be inferred, and their driving trajectories (left lane change, straight ahead, or right lane change) can be predicted in advance. This allows for the planning of the vehicle's own lane-changing trajectory, which helps to improve the safety of the vehicle when changing lanes, avoid traffic accidents, and ensure road safety.

[0021] Preferably, four boundary conditions of the initial path are obtained, and the calculation formulas for coefficients a, b, c, and d are obtained based on the boundary conditions;

[0022] The boundary conditions are as follows:

[0023]

[0024] Where q(s) t )=q t Indicates the starting boundary of the initial path; θ represents the slope of the initial path starting point. t q(s) represents the initial heading angle of the target vehicle; f )=q i q represents the endpoint boundary of the initial path. i This represents the distance from the center of each lane to the S-axis; Represents the slope of the initial path endpoint;

[0025] The formulas for calculating coefficients a, b, c, and d are obtained based on the four boundary conditions.

[0026] Preferably, based on formulas (1), (3), and (4), the modified formulas (5) for coefficient a and (6) for coefficient b are obtained:

[0027]

[0028]

[0029] Based on formulas (5) and (6), formula (1) is transformed to obtain formula (7a):

[0030]

[0031] Formula (7a) is expressed as the relationship between lane change time t prev Related paths, θ k It is the countdown to the lane change time. Among them, A k B kC k D k Corresponding to coefficients a, b, c, and d respectively, according to formulas (4), (5), and (6), we obtain A. k B k C k D k The calculation formula is as follows:

[0032]

[0033]

[0034]

[0035] D k =c(s) k -s t )+d (7e)

[0036] Among them, v t v is the longitudinal velocity value of the target vehicle at the starting point of the initial path. t For a constant value, (s) k ,q k ) represents the coordinates of the target vehicle at the current time k, and q represents the coordinates of the target vehicle. f Let s be the q value of the path endpoint. t The s value of the starting point of the path.

[0037] Preferably, the parameter θ is adaptively estimated using the recursive least squares method, and the recursive formula is as follows:

[0038]

[0039] in, In formula (7a), θ k The least squares estimate, The least squares estimate of θ for the previous period, φ(k) is the measurable signal for the current period k. T (k) is the transpose of φ(k), φ T (k) The predicted q value in the previous period, y(k) is the actual q value in the current period k, and L(k) is the recursion coefficient in the current period k.

[0040] The formula for calculating L(k) is:

[0041] L(k)=P(k)φ(k)=P(k-1)φ(k)(λ+φ T (k)P(k-1)φ(k)) -1 (9)

[0042] P(k)=(IL(k)φ T(k))λ -1 P(k-1) (10)

[0043] Where P(k) is the error covariance matrix, λ is the forgetting factor, and λ∈(0,1).

[0044] Preferably, formula (8) is nonlinearly transformed to obtain formula (11):

[0045] y(k)=f(k,θ)+v (11)

[0046] Where y(k) is the actual q value of the current period k, f(k,θ) is the formula for calculating the q value of the nonlinear system, and v is the noise during modeling;

[0047] Let v = 0, and approximate y(k) of formula (11) as follows:

[0048]

[0049] in, This indicates that the current period is k, and the q value calculated at time k-1 is obtained. F(k) represents the slope.

[0050]

[0051] Transform the above formula (12) into the form of formula (7a), and associate the calculation formula of q value with lane change time:

[0052]

[0053] in, The q-value is calculated using the prediction model;

[0054] Expanding formula (13) yields:

[0055]

[0056] in, Let F(k) be the slope.

[0057] Let the signal measured in the current period k be the slope of the path, i.e., φ. T (k)=F(k), transforming formula (10) into:

[0058]

[0059] Among them, P k Let P be the error covariance matrix for the current period k. k-1 Let λ be the error covariance matrix of the previous period, and let λ take values ​​in the range of (0,1].

[0060] According to formulas (14), (15), and (16), the least squares estimate of parameter θ is obtained:

[0061]

[0062] Where, q k,m This represents the actual q value for the current period k. The q value is calculated for the current period k, and the prediction start time is set.

[0063] According to formula (17), the adaptive lane-changing time for the current cycle k is calculated as follows:

[0064]

[0065] Preferably, the preset lane-changing time threshold t th ,

[0066] Determine the adaptive lane-changing time t prev,k Is it less than t? th ,

[0067] If so, then according to formula (3) and t prev,k Obtain the actual endpoint s of the path. f According to the actual endpoint s of the path f Three paths are obtained, the probability of each path is calculated, and the driving intention of the target vehicle is inferred based on the path with the highest probability.

[0068] Otherwise, it is inferred that the target vehicle's driving intention is to maintain its current lane and continue straight.

[0069] Preferably, based on the actual endpoint s of the path f Given three paths, calculate the probability of each path, including the following steps:

[0070] Obtain the system modeling formula:

[0071]

[0072]

[0073] Where z(t) k ) is x(t k The measured value of M(t) k ) represents the path pattern of target vehicle k in the current period. w[t k ,M(t k )] represents the current period k-path pattern M(t) k The measurement noise of v[t) k-1 ,M(t k)] represents zero-mean Gaussian white noise, F[M(t) k ] represents the linearized equation x(t) for system modeling. k The slope of H[M(t) k )] is represented as z(t k The slope of )

[0074] Get t i To t k Measurement data at time,

[0075] Based on formulas (19) and (20), the path pattern M is obtained. j Posterior probability:

[0076]

[0077] in, Indicates t i To t k A sequence of measurements over a time period; z(t) k ) represents t k The measured value at time; j = 1, 2, 3; different values ​​of j represent different path patterns M. j ;

[0078] According to formula (21), the pattern probability a for each path is obtained. k [j]:

[0079]

[0080] Among them, a k [j] represents the pattern probability of the j-th path in the current period k, a k-1 [j] represents the pattern probability of the j-th path in the previous cycle;

[0081] Based on the path with the highest pattern probability, it can be inferred that the target vehicle's driving intention is to change lanes to the left, go straight, or change lanes to the right.

[0082] Preferably, obtain path mode M j In t k Likelihood function at time:

[0083]

[0084] Among them, v j Represented as in path mode M j A linear combination of measured values, P j For its covariance, For v j (t k The probability density function of v) j (tk Let P be a Gaussian random variable with a mean of 0. j (t k ) for v j (t k The covariance of )

[0085] Obtain the estimated error covariance P for each path pattern. k [j] and the mean r of the difference between the measured value and the actual value. k [j], according to P k [j] and r k [j], calculate the likelihood p(r) for each path. k [j]|M j ),

[0086] Set the likelihood p(r) k [j]|M j The distribution has a Gaussian distribution. According to formula (22), the likelihood p(r) is obtained. k [j]|M j The formula for calculating ) is:

[0087]

[0088] Where, r k [j] represents the mean of the differences between the measured value and the actual value under the current period k and the j-th path mode, P k [j] represents the estimation error covariance of the current period k and the j-th path pattern, where n is the norm of the measurement value sequence;

[0089] Substituting formula (23) into formula (21), we obtain the pattern probability a for each path. k [j].

[0090] The present invention also designs a system for inferring the driving intentions of surrounding vehicles, comprising:

[0091] Control module, configured for

[0092] A Frenet coordinate system is established with the s-axis and q-axis as coordinate axes, and the driving path of the target vehicle is modeled as a cubic spline curve.

[0093] Preset initial lane change time t prev ,

[0094] Information collection module, configured for

[0095] Obtain the position of the target vehicle in the coordinate system (s) t ,q t ) and initial heading angle θ t Determine the initial path formula (1).

[0096] Obtain the endpoint s f The calculation formula is (3);

[0097] Based on the initial lane change time t prev and the endpoint s f The calculation formula (3) is used to obtain the preset endpoint s of the path. f ,

[0098] Obtain the calculation formulas for coefficients a, b, c, and d (4);

[0099] The calculation module is configured for...

[0100] Using the linear recursive least squares estimation algorithm, the adaptive lane-changing time t of the current cycle k is estimated according to formulas (1), (3), and (4). prev,k ,

[0101] According to formula (3) and t prev,k Obtain the actual endpoint s of the path. f ,

[0102] Based on the actual endpoint s of the path f Given three paths, calculate the probability of each path.

[0103] Based on the path with the highest probability, infer the driving intention of the target vehicle.

[0104] The present invention also provides a computer device, including a memory and a processor;

[0105] The memory is used to store computer programs;

[0106] The processor is configured to implement the above-described inference method when executing the computer program. Attached Figure Description

[0107] Figure 1 This is a flowchart of the inference method according to an embodiment of the present invention.

[0108] Figure 2 This is a schematic diagram of the inference method according to an embodiment of the present invention.

[0109] Figure 3 This is a flowchart illustrating the inference of driving intentions of surrounding vehicles in an embodiment of the present invention.

[0110] Figure 4 This is a flowchart illustrating the update process of path pattern probability in an embodiment of the present invention. Detailed Implementation

[0111] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention.

[0112] This embodiment provides a method for inferring the driving intentions of surrounding vehicles, such as... Figure 1 As shown, it includes the following steps:

[0113] Step 1: Establish a Frenet coordinate system with the s-axis and q-axis as the coordinate axes, such as... Figure 2 As shown, the driving path (left lane change, straight ahead, and right lane change) that the target vehicle needs to follow is modeled as a cubic spline curve.

[0114] Step 2: Obtain the position of the target vehicle in the coordinate system (s t ,q t ) and initial heading angle θ t Determine the initial path formula (1):

[0115] q(s)=a(ss t ) 3 +b(ss t ) 2 +c(ss t )+d,(s t ≤s≤s f (1)

[0116] The starting coordinates of the path are (s t ,q t The endpoint coordinates are (s) f ,q f ), where q f =q i i = 1, 2, 3, the q i For each lane centerline, the value of q is given, and the range of values ​​for s is [s...]. t ,s f ], s is a changing value, each s corresponds to a unique coordinate on the path, and the set of these coordinates is the path.

[0117] Where a, b, c, and d are the coefficients of the cubic spline curve, the four boundary conditions of the initial path are obtained, the shape of the initial path is determined by the four boundary conditions, and the calculation formulas of coefficients a, b, c, and d are obtained according to the boundary conditions.

[0118] The boundary conditions are as follows:

[0119]

[0120] Where q(s) t )=q t Indicates the starting boundary of the initial path; θ represents the slope of the initial path starting point. tq(s) represents the initial heading angle of the target vehicle; f )=q i q represents the endpoint boundary of the initial path. i This represents the distance from the center of each lane to the S-axis; This represents the slope of the initial path endpoint.

[0121] Step 3: Obtain the calculation formulas (4) for coefficients a, b, c, and d.

[0122]

[0123] The unique shape of the initial path is determined based on the calculation formulas for coefficients a, b, c, and d.

[0124] Step 4: Obtain the endpoint s f The calculation formula (3) is as follows:

[0125] s f =v t t prev +s t (3)

[0126] Among them, v t t represents the longitudinal velocity of the target vehicle. prev Indicates lane change time, s f -s t This indicates the longitudinal distance traveled by the target vehicle from the start of the lane change to the completion of the lane change;

[0127] Preset initial lane change time t prev It lasts for 5 seconds.

[0128] Based on the initial lane change time t prev and the endpoint s f The calculation formula is used to obtain the preset endpoint s of the path. f .

[0129] Step 5: Obtain the current cycle k. In this embodiment, the cycle is set to 0.04 seconds. Using the linear recursive least squares estimation algorithm, the lane change time t of the current cycle k is calculated according to formulas (1), (3), and (4). prev,k Adaptive estimation is performed for lane-changing time t. prev For unknown variable parameters, which vary depending on the driver's style or changes in the surrounding environment, the specific estimation method is as follows:

[0130] Substituting formula (3) into formula (1), we can make the path and lane-changing time t equal. prev By applying the boundary conditions of the path endpoint, formula (4) is transformed to obtain the transformed formula (5) for coefficient a and the transformed formula (6) for coefficient b:

[0131]

[0132]

[0133] Formulas (5) and (6) are the coefficients of the path, and these coefficients are related to the lane change time t. prev Related, based on lane change time t prev The endpoint of the path, i.e. the boundary conditions of the endpoint, can be determined, thereby determining the shape of the path.

[0134] Substituting formulas (5) and (6) into formula (1), the shape of the path is related to the lane-changing time t. prev By relating them, formula (1) can be transformed to obtain formula (7a):

[0135]

[0136] Formula (7a) is expressed as the relationship between lane change time t prev Related paths, θ k It is the countdown to the lane change time. θ k It is an adaptive target parameter (the longer the lane change time, the longer the longitudinal distance the vehicle travels during the lane change process. Compared with a short lane change time, for a fixed longitudinal distance of the same length, the longer the lane change time, the shorter the corresponding lateral distance q).

[0137] Among them, A k B k C k D k Corresponding to coefficients a, b, c, and d respectively, according to formulas (4), (5), and (6), we obtain A. k B k C k D k The calculation formula is as follows:

[0138]

[0139]

[0140]

[0141] D k =c(s) k -s t )+d (7e)

[0142] Among them, v t v is the longitudinal velocity value of the target vehicle at the starting point of the initial path. t For a constant value, (s) k ,q k) represents the coordinates of the target vehicle at the current time k, and q represents the coordinates of the target vehicle. f Let s be the q value of the path endpoint. t The s value of the starting point of the path.

[0143] The parameter θ is adaptively estimated using the recursive least squares method, and the recursive formula is as follows:

[0144]

[0145] in, In formula (7a), θ k The least squares estimate, The least squares estimate of θ for the previous period, φ(k) is the measurable signal for the current period k. T (k) is the transpose of φ(k), φ T (k) The predicted q value in the previous period, y(k) is the actual q value in the current period k, and L(k) is the recursion coefficient in the current period k.

[0146] The formula for calculating L(k) is:

[0147] L(k)=P(k)φ(k)=P(k-1)φ(k)(λ+φ T (k)P(k-1)φ(k)) -1 (9)

[0148] P(k)=(IL(k)φ T (k))λ -1 P(k-1) (10)

[0149] Where P(k) is the error covariance matrix, λ is the forgetting factor, λ∈(0,1], which assigns higher weight to the most recently measured data.

[0150] By performing a nonlinear transformation on formula (8), we obtain formula (11):

[0151] y(k)=f(k,θ)+v (11)

[0152] Where y(k) is the actual q value of the current period k, f(k,θ) is the formula for calculating the q value of the nonlinear system, and v is the noise during modeling;

[0153] Assuming noise v = 0, the following approximation is made to y(k) in equation (11):

[0154]

[0155] in, This indicates that the current period is k, and the q value calculated at time k-1 is obtained. F(k) represents the slope.

[0156]

[0157] Transform the above formula (12) into the form of formula (7a), and associate the calculation formula of q value with lane change time:

[0158]

[0159] in, The q-value is calculated using the prediction model;

[0160] Expanding formula (13) yields:

[0161]

[0162] in, Let F(k) be the slope.

[0163] Let the signal measured in the current period k be the slope of the path, i.e., φ. T (k)=F(k), transforming formula (10) into:

[0164]

[0165] Among them, P k Let P be the error covariance matrix for the current period k. k-1 Let λ be the error covariance matrix of the previous period, and let λ take values ​​in the range of (0,1].

[0166] According to formulas (14), (15), and (16), the least squares estimate of parameter θ is obtained:

[0167]

[0168] Where, q k,m This represents the actual q value for the current period k. The q value is calculated for the current period k, and the prediction start time is set.

[0169] According to formula (17), the adaptive lane-changing time for the current cycle k is calculated as follows:

[0170]

[0171] Step 6: Based on the actual collected data samples, preset the lane-changing time threshold t. th For 10 seconds, such as Figure 3 As shown, the adaptive lane-changing time t is determined. prev,k Is it less than t? th ,

[0172] If so, then the adaptive lane-changing time t of the current cycle k will be... prev,k Substituting into formula (3), we obtain the actual endpoint s of the path. f ;

[0173] Otherwise, it indicates that the curvature of the path is small and the determined endpoint is far away. The path generated by using cubic spline curves is long. Therefore, regardless of the path pattern probability, it is determined that the target vehicle does not intend to change lanes, and it can be directly inferred that the target vehicle's driving behavior intention is to keep the current lane straight.

[0174] Step 7: Obtain the actual endpoint s of the path f Then, based on the actual endpoint s of the path f Given three paths, calculate the probability of each path using the following steps:

[0175] Obtain the system modeling formula:

[0176]

[0177]

[0178] Where z(t) k ) is x(t k The measured value of M(t) k ) represents the path pattern of target vehicle k in the current period. w[t k ,M(t k )] represents the current period k-path pattern M(t) k The measurement noise of v[t) k-1 ,M(t k )] represents zero-mean Gaussian white noise, F[M(t) k ] represents the linearized equation x(t) for system modeling. k The slope of H[M(t) k )] is represented as z(t k The slope of )

[0179] Get t i To t k Measurement data at time,

[0180] Based on formulas (19) and (20), the path pattern M is obtained. j Posterior probability:

[0181]

[0182] in, Indicates t i To t k A sequence of measurements over a time period; z(t) k) represents t k The measured value at time; j = 1, 2, 3; different values ​​of j represent different path patterns M. j ;

[0183] Get path pattern M j In t k Likelihood function at time:

[0184]

[0185] Among them, v j Represented as in path mode M j A linear combination of measured values, P j For its covariance, For v j (t k The probability density function of v) j (t k Let P be a Gaussian random variable with a mean of 0. j (t k ) for v j (t k The covariance of )

[0186] Obtain the estimated error covariance P for each path pattern. k [j] and the mean r of the difference between the measured value and the actual value. k [j], according to P k [j] and r k [j], calculate the likelihood p(r) for each path. k [j]|M j ),

[0187] Set the likelihood p(r) k [j]|M j The distribution has a Gaussian distribution. According to formula (22), the likelihood p(r) is obtained. k [j]|M j The formula for calculating ) is:

[0188]

[0189] Where, r k [j] represents the mean of the differences between the measured value and the actual value under the current period k and the j-th path mode, P k [j] represents the estimation error covariance of the current period k and the j-th path pattern, where n is the norm of the measurement value sequence;

[0190] Substituting formula (23) into formula (21), we obtain the pattern probability a for each path. k [j]:

[0191]

[0192] Among them, a k [j] represents the pattern probability of the j-th path in the current period k, a k-1 [j] represents the pattern probability of the j-th path in the previous cycle;

[0193] Based on the path with the highest pattern probability, it can be inferred that the target vehicle's driving intention is to change lanes to the left, go straight, or change lanes to the right.

[0194] The pattern probability of the j-th path in the previous period k-1 will affect the pattern probability of the current period k. Similarly, the pattern probability of the current period k will also affect the pattern probability of the j-th path in the next period k+1. Figure 4 As shown, the posterior value at the input of the previous period k-1 is... (i.e., the heading angle or slope of the target vehicle), estimation error covariance and Y k (The actual heading angle or slope of the target vehicle in the current period k), a k-1 [j] represents the path pattern probability calculated in the previous period k-1. The likelihood of each pattern is calculated by comparing the measured values ​​and the actual observed values, and the pattern probability is updated. The three path patterns and their probabilities in the current period are updated once in each period to perform real-time predictive analysis of the driving intentions of surrounding vehicles, thereby improving the accuracy of the inference results.

[0195] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for inferring the driving intentions of surrounding vehicles, characterized in that, Includes the following steps: Step 1: Establish a Frenet coordinate system with the s-axis and q-axis as coordinate axes, and model the driving path of the target vehicle as a cubic spline curve; Step 2: Obtain the position of the target vehicle in the coordinate system and initial heading angle Determine the initial path formula (1): The starting coordinates of the path are The coordinates of the endpoint are ,in, = i=1,2,3, the The center line corresponding to each lane q The value of s is [ ]. The a , b , c , d Indicates coefficient; Step 3: Obtain the coordinates of the endpoint. The calculation formula (3): in, This represents the longitudinal velocity of the target vehicle. Indicates lane change time. - This indicates the longitudinal distance traveled by the target vehicle from the start of the lane change to the completion of the lane change; Preset initial lane change time , Based on the initial lane change time and the coordinates of the endpoint The calculation formula is used to obtain the preset endpoint coordinates of the path. ; Step 4: Obtain the coefficients a , b , c , d The calculation formula is (4). According to the coefficient a , b , c , d The calculation formula determines the unique shape of the initial path; Step 5: Obtain the current cycle k, and use the linear recursive least squares estimation algorithm to estimate the adaptive lane-changing time of the current cycle k according to formulas (1), (3) and (4). , According to formula (3) and To obtain the actual endpoint coordinates of the path ; Step 6: Based on the actual endpoint coordinates of the path Given three paths, calculate the probability of each path. Based on the path with the highest probability, infer the driving intention of the target vehicle.

2. The method for inferring the driving intentions of surrounding vehicles according to claim 1, characterized in that, Obtain the four boundary conditions for the initial path, and obtain the coefficients based on the boundary conditions. a , b , c , d The calculation formula; The boundary conditions are as follows: in, Indicates the starting boundary of the initial path; ( )=tan( The slope of the initial path starting point is represented by ). Indicates the initial heading angle of the target vehicle; Indicates the endpoint boundary of the initial path. This represents the distance from the center of each lane to the S-axis; ( )=0 indicates the slope of the initial path endpoint; The coefficients are obtained based on the four boundary conditions. a , b , c , d The calculation formula.

3. The method for inferring the driving intentions of surrounding vehicles according to claim 1, characterized in that, Based on formulas (1), (3), and (4), obtain the coefficients. a The transformed formula (5) and b The transformed formula (6): Based on formulas (5) and (6), formula (1) is transformed to obtain formula (7). a ): The formula (7) a () indicates the time to change lanes. Related paths, It is the countdown to the lane change time. ,in, , , , Corresponding coefficients a , b , c , d According to formulas (4), (5) and (6), we get , , , The calculation formula is as follows: in, The longitudinal velocity value of the target vehicle at the starting point of the initial path, It is a constant value. , () represents the coordinates of the target vehicle at the current period k. The q value represents the endpoint of the path. The s value of the starting point of the path.

4. The method for inferring the driving intentions of surrounding vehicles according to claim 3, characterized in that, Using recursive least squares method for parameters For adaptive estimation, the recursive formula is: in, For formula (7) a )middle The least squares estimate, For the previous cycle The least squares estimate, For the signal that can be measured in the current period k, for transpose, The q value predicted in the previous period, This represents the actual q value for the current period k. The recursive coefficient at the current period k; The The calculation formula is: in, Let be the error covariance matrix. Forgetting factor, (0,1).

5. The method for inferring the driving intentions of surrounding vehicles according to claim 4, characterized in that, By performing a nonlinear transformation on formula (8), we obtain formula (11): in, This represents the actual q value for the current period k. Here is the formula for calculating the q-value of a nonlinear system. This refers to noise during modeling. set up For formula (11) Make the following approximation: in, This indicates that the current period is k, and the value of q calculated at time k-1 is obtained. Expressed as slope, Transform the above formula (12) into the form of formula (7a), and associate the calculation formula of q value with lane change time: in, The q-value is calculated using the prediction model; Expanding formula (13) yields: in, slope , Let the signal measured in the current period k be the slope of the path, i.e. = Equation (10) can be transformed into: in, Let k be the error covariance matrix for the current period k. The error covariance matrix of the previous period. The value range is (0,1]; According to formulas (14), (15), and (16), the parameters are obtained. Least squares estimate: in, This represents the actual q value for the current period k. The q value is calculated for the current period k, and the prediction start time is set. = ; According to formula (17), the adaptive lane-changing time for the current cycle k is calculated as follows: 。 6. The method for inferring the driving intentions of surrounding vehicles according to claim 1, characterized in that, Preset lane change time threshold , Determine the adaptive lane changing time Is it less than , If so, then according to formula (3) and To obtain the actual endpoint coordinates of the path Based on the actual endpoint coordinates of the path Three paths are obtained, the probability of each path is calculated, and the driving intention of the target vehicle is inferred based on the path with the highest probability. Otherwise, it is inferred that the target vehicle's driving intention is to maintain its current lane and continue straight.

7. The method for inferring the driving intentions of surrounding vehicles according to claim 1, characterized in that, Based on the actual endpoint coordinates of the path Given three paths, calculate the probability of each path, including the following steps: Obtain the system modeling formula: in, for The measured value, This represents the path pattern of target vehicle k in the current period. ; Represented as the current period k-path pattern Measurement noise, Represented as zero-mean Gaussian white noise, Represented as linearized equations for system modeling The slope, Represented as The slope; Get to Measurement data at time, Based on formulas (19) and (20), the path pattern is obtained. Posterior probability: in, express to A sequence of measurements over a period of time; express The measured value at time; j=1,2,3; different values ​​of j represent different path patterns. ; According to formula (21), the pattern probability of each path is obtained. : in, Let be the pattern probability of the j-th path in the current period k. This represents the pattern probability of the j-th path in the previous period; Based on the path with the highest pattern probability, it can be inferred that the target vehicle's driving intention is to change lanes to the left, go straight, or change lanes to the right.

8. The method for inferring the driving intentions of surrounding vehicles according to claim 7, characterized in that, Get path mode exist Likelihood function at time: in, Indicated in path mode A linear combination of the measured values, For its covariance, for The probability density function, Let be a Gaussian random variable with a mean of 0. for covariance; Obtain the estimated error covariance for each path pattern and the mean of the differences between the measured value and the actual value ,according to and The likelihood of each path is calculated. , Setting likelihood Having a Gaussian distribution, the likelihood is obtained according to formula (22). The calculation formula is as follows: in, This represents the mean of the differences between the measured value and the actual value under the current period k and the j-th path mode. Let be the estimated error covariance of the current period k and the j-th path pattern, and n be the norm of the measurement value sequence; Substituting formula (23) into formula (21), we obtain the pattern probability for each path. .

9. A system for inferring the driving intentions of surrounding vehicles, characterized in that, include: Control module, configured for A Frenet coordinate system is established with the s-axis and q-axis as coordinate axes, and the driving path of the target vehicle is modeled as a cubic spline curve. Preset initial lane change time , Information collection module, configured for Obtain the position of the target vehicle in the coordinate system and initial heading angle Determine the initial path formula Get the coordinates of the destination Calculation formula Based on the initial lane change time and the coordinates of the endpoint Calculation formula Get the preset endpoint coordinates of the path , Obtaining coefficients a , b , c , d Calculation formula The calculation module is configured for... Using the linear recursive least squares estimation algorithm, according to the formula ,formula and formula Estimate the adaptive lane-changing time for the current period k. , According to the formula and To obtain the actual endpoint coordinates of the path , Based on the actual endpoint coordinates of the path Given three paths, calculate the probability of each path. Based on the path with the highest probability, infer the driving intention of the target vehicle.

10. A computer device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the inference method as described in any one of claims 1-8 when executing the computer program.

Images